Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * predicate.c
4 : : * POSTGRES predicate locking
5 : : * to support full serializable transaction isolation
6 : : *
7 : : *
8 : : * The approach taken is to implement Serializable Snapshot Isolation (SSI)
9 : : * as initially described in this paper:
10 : : *
11 : : * Michael J. Cahill, Uwe Röhm, and Alan D. Fekete. 2008.
12 : : * Serializable isolation for snapshot databases.
13 : : * In SIGMOD '08: Proceedings of the 2008 ACM SIGMOD
14 : : * international conference on Management of data,
15 : : * pages 729-738, New York, NY, USA. ACM.
16 : : * http://doi.acm.org/10.1145/1376616.1376690
17 : : *
18 : : * and further elaborated in Cahill's doctoral thesis:
19 : : *
20 : : * Michael James Cahill. 2009.
21 : : * Serializable Isolation for Snapshot Databases.
22 : : * Sydney Digital Theses.
23 : : * University of Sydney, School of Information Technologies.
24 : : * http://hdl.handle.net/2123/5353
25 : : *
26 : : *
27 : : * Predicate locks for Serializable Snapshot Isolation (SSI) are SIREAD
28 : : * locks, which are so different from normal locks that a distinct set of
29 : : * structures is required to handle them. They are needed to detect
30 : : * rw-conflicts when the read happens before the write. (When the write
31 : : * occurs first, the reading transaction can check for a conflict by
32 : : * examining the MVCC data.)
33 : : *
34 : : * (1) Besides tuples actually read, they must cover ranges of tuples
35 : : * which would have been read based on the predicate. This will
36 : : * require modelling the predicates through locks against database
37 : : * objects such as pages, index ranges, or entire tables.
38 : : *
39 : : * (2) They must be kept in RAM for quick access. Because of this, it
40 : : * isn't possible to always maintain tuple-level granularity -- when
41 : : * the space allocated to store these approaches exhaustion, a
42 : : * request for a lock may need to scan for situations where a single
43 : : * transaction holds many fine-grained locks which can be coalesced
44 : : * into a single coarser-grained lock.
45 : : *
46 : : * (3) They never block anything; they are more like flags than locks
47 : : * in that regard; although they refer to database objects and are
48 : : * used to identify rw-conflicts with normal write locks.
49 : : *
50 : : * (4) While they are associated with a transaction, they must survive
51 : : * a successful COMMIT of that transaction, and remain until all
52 : : * overlapping transactions complete. This even means that they
53 : : * must survive termination of the transaction's process. If a
54 : : * top level transaction is rolled back, however, it is immediately
55 : : * flagged so that it can be ignored, and its SIREAD locks can be
56 : : * released any time after that.
57 : : *
58 : : * (5) The only transactions which create SIREAD locks or check for
59 : : * conflicts with them are serializable transactions.
60 : : *
61 : : * (6) When a write lock for a top level transaction is found to cover
62 : : * an existing SIREAD lock for the same transaction, the SIREAD lock
63 : : * can be deleted.
64 : : *
65 : : * (7) A write from a serializable transaction must ensure that an xact
66 : : * record exists for the transaction, with the same lifespan (until
67 : : * all concurrent transaction complete or the transaction is rolled
68 : : * back) so that rw-dependencies to that transaction can be
69 : : * detected.
70 : : *
71 : : * We use an optimization for read-only transactions. Under certain
72 : : * circumstances, a read-only transaction's snapshot can be shown to
73 : : * never have conflicts with other transactions. This is referred to
74 : : * as a "safe" snapshot (and one known not to be is "unsafe").
75 : : * However, it can't be determined whether a snapshot is safe until
76 : : * all concurrent read/write transactions complete.
77 : : *
78 : : * Once a read-only transaction is known to have a safe snapshot, it
79 : : * can release its predicate locks and exempt itself from further
80 : : * predicate lock tracking. READ ONLY DEFERRABLE transactions run only
81 : : * on safe snapshots, waiting as necessary for one to be available.
82 : : *
83 : : *
84 : : * Lightweight locks to manage access to the predicate locking shared
85 : : * memory objects must be taken in this order, and should be released in
86 : : * reverse order:
87 : : *
88 : : * SerializableFinishedListLock
89 : : * - Protects the list of transactions which have completed but which
90 : : * may yet matter because they overlap still-active transactions.
91 : : *
92 : : * SerializablePredicateListLock
93 : : * - Protects the linked list of locks held by a transaction. Note
94 : : * that the locks themselves are also covered by the partition
95 : : * locks of their respective lock targets; this lock only affects
96 : : * the linked list connecting the locks related to a transaction.
97 : : * - All transactions share this single lock (with no partitioning).
98 : : * - There is never a need for a process other than the one running
99 : : * an active transaction to walk the list of locks held by that
100 : : * transaction, except parallel query workers sharing the leader's
101 : : * transaction. In the parallel case, an extra per-sxact lock is
102 : : * taken; see below.
103 : : * - It is relatively infrequent that another process needs to
104 : : * modify the list for a transaction, but it does happen for such
105 : : * things as index page splits for pages with predicate locks and
106 : : * freeing of predicate locked pages by a vacuum process. When
107 : : * removing a lock in such cases, the lock itself contains the
108 : : * pointers needed to remove it from the list. When adding a
109 : : * lock in such cases, the lock can be added using the anchor in
110 : : * the transaction structure. Neither requires walking the list.
111 : : * - Cleaning up the list for a terminated transaction is sometimes
112 : : * not done on a retail basis, in which case no lock is required.
113 : : * - Due to the above, a process accessing its active transaction's
114 : : * list always uses a shared lock, regardless of whether it is
115 : : * walking or maintaining the list. This improves concurrency
116 : : * for the common access patterns.
117 : : * - A process which needs to alter the list of a transaction other
118 : : * than its own active transaction must acquire an exclusive
119 : : * lock.
120 : : *
121 : : * SERIALIZABLEXACT's member 'perXactPredicateListLock'
122 : : * - Protects the linked list of predicate locks held by a transaction.
123 : : * Only needed for parallel mode, where multiple backends share the
124 : : * same SERIALIZABLEXACT object. Not needed if
125 : : * SerializablePredicateListLock is held exclusively.
126 : : *
127 : : * PredicateLockHashPartitionLock(hashcode)
128 : : * - The same lock protects a target, all locks on that target, and
129 : : * the linked list of locks on the target.
130 : : * - When more than one is needed, acquire in ascending address order.
131 : : * - When all are needed (rare), acquire in ascending index order with
132 : : * PredicateLockHashPartitionLockByIndex(index).
133 : : *
134 : : * SerializableXactHashLock
135 : : * - Protects both PredXact and SerializableXidHash.
136 : : *
137 : : * SerialControlLock
138 : : * - Protects SerialControlData members
139 : : *
140 : : * SLRU per-bank locks
141 : : * - Protects SerialSlruCtl
142 : : *
143 : : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
144 : : * Portions Copyright (c) 1994, Regents of the University of California
145 : : *
146 : : *
147 : : * IDENTIFICATION
148 : : * src/backend/storage/lmgr/predicate.c
149 : : *
150 : : *-------------------------------------------------------------------------
151 : : */
152 : : /*
153 : : * INTERFACE ROUTINES
154 : : *
155 : : * housekeeping for setting up shared memory predicate lock structures
156 : : * InitPredicateLocks(void)
157 : : * PredicateLockShmemSize(void)
158 : : *
159 : : * predicate lock reporting
160 : : * GetPredicateLockStatusData(void)
161 : : * PageIsPredicateLocked(Relation relation, BlockNumber blkno)
162 : : *
163 : : * predicate lock maintenance
164 : : * GetSerializableTransactionSnapshot(Snapshot snapshot)
165 : : * SetSerializableTransactionSnapshot(Snapshot snapshot,
166 : : * VirtualTransactionId *sourcevxid)
167 : : * RegisterPredicateLockingXid(void)
168 : : * PredicateLockRelation(Relation relation, Snapshot snapshot)
169 : : * PredicateLockPage(Relation relation, BlockNumber blkno,
170 : : * Snapshot snapshot)
171 : : * PredicateLockTID(Relation relation, ItemPointer tid, Snapshot snapshot,
172 : : * TransactionId tuple_xid)
173 : : * PredicateLockPageSplit(Relation relation, BlockNumber oldblkno,
174 : : * BlockNumber newblkno)
175 : : * PredicateLockPageCombine(Relation relation, BlockNumber oldblkno,
176 : : * BlockNumber newblkno)
177 : : * TransferPredicateLocksToHeapRelation(Relation relation)
178 : : * ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
179 : : *
180 : : * conflict detection (may also trigger rollback)
181 : : * CheckForSerializableConflictOut(Relation relation, TransactionId xid,
182 : : * Snapshot snapshot)
183 : : * CheckForSerializableConflictIn(Relation relation, ItemPointer tid,
184 : : * BlockNumber blkno)
185 : : * CheckTableForSerializableConflictIn(Relation relation)
186 : : *
187 : : * final rollback checking
188 : : * PreCommit_CheckForSerializationFailure(void)
189 : : *
190 : : * two-phase commit support
191 : : * AtPrepare_PredicateLocks(void);
192 : : * PostPrepare_PredicateLocks(TransactionId xid);
193 : : * PredicateLockTwoPhaseFinish(TransactionId xid, bool isCommit);
194 : : * predicatelock_twophase_recover(TransactionId xid, uint16 info,
195 : : * void *recdata, uint32 len);
196 : : */
197 : :
198 : : #include "postgres.h"
199 : :
200 : : #include "access/parallel.h"
201 : : #include "access/slru.h"
202 : : #include "access/transam.h"
203 : : #include "access/twophase.h"
204 : : #include "access/twophase_rmgr.h"
205 : : #include "access/xact.h"
206 : : #include "access/xlog.h"
207 : : #include "miscadmin.h"
208 : : #include "pgstat.h"
209 : : #include "port/pg_lfind.h"
210 : : #include "storage/predicate.h"
211 : : #include "storage/predicate_internals.h"
212 : : #include "storage/proc.h"
213 : : #include "storage/procarray.h"
214 : : #include "utils/guc_hooks.h"
215 : : #include "utils/rel.h"
216 : : #include "utils/snapmgr.h"
217 : :
218 : : /* Uncomment the next line to test the graceful degradation code. */
219 : : /* #define TEST_SUMMARIZE_SERIAL */
220 : :
221 : : /*
222 : : * Test the most selective fields first, for performance.
223 : : *
224 : : * a is covered by b if all of the following hold:
225 : : * 1) a.database = b.database
226 : : * 2) a.relation = b.relation
227 : : * 3) b.offset is invalid (b is page-granularity or higher)
228 : : * 4) either of the following:
229 : : * 4a) a.offset is valid (a is tuple-granularity) and a.page = b.page
230 : : * or 4b) a.offset is invalid and b.page is invalid (a is
231 : : * page-granularity and b is relation-granularity
232 : : */
233 : : #define TargetTagIsCoveredBy(covered_target, covering_target) \
234 : : ((GET_PREDICATELOCKTARGETTAG_RELATION(covered_target) == /* (2) */ \
235 : : GET_PREDICATELOCKTARGETTAG_RELATION(covering_target)) \
236 : : && (GET_PREDICATELOCKTARGETTAG_OFFSET(covering_target) == \
237 : : InvalidOffsetNumber) /* (3) */ \
238 : : && (((GET_PREDICATELOCKTARGETTAG_OFFSET(covered_target) != \
239 : : InvalidOffsetNumber) /* (4a) */ \
240 : : && (GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \
241 : : GET_PREDICATELOCKTARGETTAG_PAGE(covered_target))) \
242 : : || ((GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \
243 : : InvalidBlockNumber) /* (4b) */ \
244 : : && (GET_PREDICATELOCKTARGETTAG_PAGE(covered_target) \
245 : : != InvalidBlockNumber))) \
246 : : && (GET_PREDICATELOCKTARGETTAG_DB(covered_target) == /* (1) */ \
247 : : GET_PREDICATELOCKTARGETTAG_DB(covering_target)))
248 : :
249 : : /*
250 : : * The predicate locking target and lock shared hash tables are partitioned to
251 : : * reduce contention. To determine which partition a given target belongs to,
252 : : * compute the tag's hash code with PredicateLockTargetTagHashCode(), then
253 : : * apply one of these macros.
254 : : * NB: NUM_PREDICATELOCK_PARTITIONS must be a power of 2!
255 : : */
256 : : #define PredicateLockHashPartition(hashcode) \
257 : : ((hashcode) % NUM_PREDICATELOCK_PARTITIONS)
258 : : #define PredicateLockHashPartitionLock(hashcode) \
259 : : (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + \
260 : : PredicateLockHashPartition(hashcode)].lock)
261 : : #define PredicateLockHashPartitionLockByIndex(i) \
262 : : (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + (i)].lock)
263 : :
264 : : #define NPREDICATELOCKTARGETENTS() \
265 : : mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
266 : :
267 : : #define SxactIsOnFinishedList(sxact) (!dlist_node_is_detached(&(sxact)->finishedLink))
268 : :
269 : : /*
270 : : * Note that a sxact is marked "prepared" once it has passed
271 : : * PreCommit_CheckForSerializationFailure, even if it isn't using
272 : : * 2PC. This is the point at which it can no longer be aborted.
273 : : *
274 : : * The PREPARED flag remains set after commit, so SxactIsCommitted
275 : : * implies SxactIsPrepared.
276 : : */
277 : : #define SxactIsCommitted(sxact) (((sxact)->flags & SXACT_FLAG_COMMITTED) != 0)
278 : : #define SxactIsPrepared(sxact) (((sxact)->flags & SXACT_FLAG_PREPARED) != 0)
279 : : #define SxactIsRolledBack(sxact) (((sxact)->flags & SXACT_FLAG_ROLLED_BACK) != 0)
280 : : #define SxactIsDoomed(sxact) (((sxact)->flags & SXACT_FLAG_DOOMED) != 0)
281 : : #define SxactIsReadOnly(sxact) (((sxact)->flags & SXACT_FLAG_READ_ONLY) != 0)
282 : : #define SxactHasSummaryConflictIn(sxact) (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_IN) != 0)
283 : : #define SxactHasSummaryConflictOut(sxact) (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_OUT) != 0)
284 : : /*
285 : : * The following macro actually means that the specified transaction has a
286 : : * conflict out *to a transaction which committed ahead of it*. It's hard
287 : : * to get that into a name of a reasonable length.
288 : : */
289 : : #define SxactHasConflictOut(sxact) (((sxact)->flags & SXACT_FLAG_CONFLICT_OUT) != 0)
290 : : #define SxactIsDeferrableWaiting(sxact) (((sxact)->flags & SXACT_FLAG_DEFERRABLE_WAITING) != 0)
291 : : #define SxactIsROSafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_SAFE) != 0)
292 : : #define SxactIsROUnsafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_UNSAFE) != 0)
293 : : #define SxactIsPartiallyReleased(sxact) (((sxact)->flags & SXACT_FLAG_PARTIALLY_RELEASED) != 0)
294 : :
295 : : /*
296 : : * Compute the hash code associated with a PREDICATELOCKTARGETTAG.
297 : : *
298 : : * To avoid unnecessary recomputations of the hash code, we try to do this
299 : : * just once per function, and then pass it around as needed. Aside from
300 : : * passing the hashcode to hash_search_with_hash_value(), we can extract
301 : : * the lock partition number from the hashcode.
302 : : */
303 : : #define PredicateLockTargetTagHashCode(predicatelocktargettag) \
304 : : get_hash_value(PredicateLockTargetHash, predicatelocktargettag)
305 : :
306 : : /*
307 : : * Given a predicate lock tag, and the hash for its target,
308 : : * compute the lock hash.
309 : : *
310 : : * To make the hash code also depend on the transaction, we xor the sxid
311 : : * struct's address into the hash code, left-shifted so that the
312 : : * partition-number bits don't change. Since this is only a hash, we
313 : : * don't care if we lose high-order bits of the address; use an
314 : : * intermediate variable to suppress cast-pointer-to-int warnings.
315 : : */
316 : : #define PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash) \
317 : : ((targethash) ^ ((uint32) PointerGetDatum((predicatelocktag)->myXact)) \
318 : : << LOG2_NUM_PREDICATELOCK_PARTITIONS)
319 : :
320 : :
321 : : /*
322 : : * The SLRU buffer area through which we access the old xids.
323 : : */
324 : : static SlruCtlData SerialSlruCtlData;
325 : :
326 : : #define SerialSlruCtl (&SerialSlruCtlData)
327 : :
328 : : #define SERIAL_PAGESIZE BLCKSZ
329 : : #define SERIAL_ENTRYSIZE sizeof(SerCommitSeqNo)
330 : : #define SERIAL_ENTRIESPERPAGE (SERIAL_PAGESIZE / SERIAL_ENTRYSIZE)
331 : :
332 : : /*
333 : : * Set maximum pages based on the number needed to track all transactions.
334 : : */
335 : : #define SERIAL_MAX_PAGE (MaxTransactionId / SERIAL_ENTRIESPERPAGE)
336 : :
337 : : #define SerialNextPage(page) (((page) >= SERIAL_MAX_PAGE) ? 0 : (page) + 1)
338 : :
339 : : #define SerialValue(slotno, xid) (*((SerCommitSeqNo *) \
340 : : (SerialSlruCtl->shared->page_buffer[slotno] + \
341 : : ((((uint32) (xid)) % SERIAL_ENTRIESPERPAGE) * SERIAL_ENTRYSIZE))))
342 : :
343 : : #define SerialPage(xid) (((uint32) (xid)) / SERIAL_ENTRIESPERPAGE)
344 : :
345 : : typedef struct SerialControlData
346 : : {
347 : : int headPage; /* newest initialized page */
348 : : TransactionId headXid; /* newest valid Xid in the SLRU */
349 : : TransactionId tailXid; /* oldest xmin we might be interested in */
350 : : } SerialControlData;
351 : :
352 : : typedef struct SerialControlData *SerialControl;
353 : :
354 : : static SerialControl serialControl;
355 : :
356 : : /*
357 : : * When the oldest committed transaction on the "finished" list is moved to
358 : : * SLRU, its predicate locks will be moved to this "dummy" transaction,
359 : : * collapsing duplicate targets. When a duplicate is found, the later
360 : : * commitSeqNo is used.
361 : : */
362 : : static SERIALIZABLEXACT *OldCommittedSxact;
363 : :
364 : :
365 : : /*
366 : : * These configuration variables are used to set the predicate lock table size
367 : : * and to control promotion of predicate locks to coarser granularity in an
368 : : * attempt to degrade performance (mostly as false positive serialization
369 : : * failure) gracefully in the face of memory pressure.
370 : : */
371 : : int max_predicate_locks_per_xact; /* in guc_tables.c */
372 : : int max_predicate_locks_per_relation; /* in guc_tables.c */
373 : : int max_predicate_locks_per_page; /* in guc_tables.c */
374 : :
375 : : /*
376 : : * This provides a list of objects in order to track transactions
377 : : * participating in predicate locking. Entries in the list are fixed size,
378 : : * and reside in shared memory. The memory address of an entry must remain
379 : : * fixed during its lifetime. The list will be protected from concurrent
380 : : * update externally; no provision is made in this code to manage that. The
381 : : * number of entries in the list, and the size allowed for each entry is
382 : : * fixed upon creation.
383 : : */
384 : : static PredXactList PredXact;
385 : :
386 : : /*
387 : : * This provides a pool of RWConflict data elements to use in conflict lists
388 : : * between transactions.
389 : : */
390 : : static RWConflictPoolHeader RWConflictPool;
391 : :
392 : : /*
393 : : * The predicate locking hash tables are in shared memory.
394 : : * Each backend keeps pointers to them.
395 : : */
396 : : static HTAB *SerializableXidHash;
397 : : static HTAB *PredicateLockTargetHash;
398 : : static HTAB *PredicateLockHash;
399 : : static dlist_head *FinishedSerializableTransactions;
400 : :
401 : : /*
402 : : * Tag for a dummy entry in PredicateLockTargetHash. By temporarily removing
403 : : * this entry, you can ensure that there's enough scratch space available for
404 : : * inserting one entry in the hash table. This is an otherwise-invalid tag.
405 : : */
406 : : static const PREDICATELOCKTARGETTAG ScratchTargetTag = {0, 0, 0, 0};
407 : : static uint32 ScratchTargetTagHash;
408 : : static LWLock *ScratchPartitionLock;
409 : :
410 : : /*
411 : : * The local hash table used to determine when to combine multiple fine-
412 : : * grained locks into a single courser-grained lock.
413 : : */
414 : : static HTAB *LocalPredicateLockHash = NULL;
415 : :
416 : : /*
417 : : * Keep a pointer to the currently-running serializable transaction (if any)
418 : : * for quick reference. Also, remember if we have written anything that could
419 : : * cause a rw-conflict.
420 : : */
421 : : static SERIALIZABLEXACT *MySerializableXact = InvalidSerializableXact;
422 : : static bool MyXactDidWrite = false;
423 : :
424 : : /*
425 : : * The SXACT_FLAG_RO_UNSAFE optimization might lead us to release
426 : : * MySerializableXact early. If that happens in a parallel query, the leader
427 : : * needs to defer the destruction of the SERIALIZABLEXACT until end of
428 : : * transaction, because the workers still have a reference to it. In that
429 : : * case, the leader stores it here.
430 : : */
431 : : static SERIALIZABLEXACT *SavedSerializableXact = InvalidSerializableXact;
432 : :
433 : : /* local functions */
434 : :
435 : : static SERIALIZABLEXACT *CreatePredXact(void);
436 : : static void ReleasePredXact(SERIALIZABLEXACT *sxact);
437 : :
438 : : static bool RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer);
439 : : static void SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer);
440 : : static void SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact, SERIALIZABLEXACT *activeXact);
441 : : static void ReleaseRWConflict(RWConflict conflict);
442 : : static void FlagSxactUnsafe(SERIALIZABLEXACT *sxact);
443 : :
444 : : static bool SerialPagePrecedesLogically(int64 page1, int64 page2);
445 : : static void SerialInit(void);
446 : : static void SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo);
447 : : static SerCommitSeqNo SerialGetMinConflictCommitSeqNo(TransactionId xid);
448 : : static void SerialSetActiveSerXmin(TransactionId xid);
449 : :
450 : : static uint32 predicatelock_hash(const void *key, Size keysize);
451 : : static void SummarizeOldestCommittedSxact(void);
452 : : static Snapshot GetSafeSnapshot(Snapshot origSnapshot);
453 : : static Snapshot GetSerializableTransactionSnapshotInt(Snapshot snapshot,
454 : : VirtualTransactionId *sourcevxid,
455 : : int sourcepid);
456 : : static bool PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag);
457 : : static bool GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag,
458 : : PREDICATELOCKTARGETTAG *parent);
459 : : static bool CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag);
460 : : static void RemoveScratchTarget(bool lockheld);
461 : : static void RestoreScratchTarget(bool lockheld);
462 : : static void RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target,
463 : : uint32 targettaghash);
464 : : static void DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag);
465 : : static int MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag);
466 : : static bool CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag);
467 : : static void DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag);
468 : : static void CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
469 : : uint32 targettaghash,
470 : : SERIALIZABLEXACT *sxact);
471 : : static void DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash);
472 : : static bool TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
473 : : PREDICATELOCKTARGETTAG newtargettag,
474 : : bool removeOld);
475 : : static void PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag);
476 : : static void DropAllPredicateLocksFromTable(Relation relation,
477 : : bool transfer);
478 : : static void SetNewSxactGlobalXmin(void);
479 : : static void ClearOldPredicateLocks(void);
480 : : static void ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
481 : : bool summarize);
482 : : static bool XidIsConcurrent(TransactionId xid);
483 : : static void CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag);
484 : : static void FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer);
485 : : static void OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader,
486 : : SERIALIZABLEXACT *writer);
487 : : static void CreateLocalPredicateLockHash(void);
488 : : static void ReleasePredicateLocksLocal(void);
489 : :
490 : :
491 : : /*------------------------------------------------------------------------*/
492 : :
493 : : /*
494 : : * Does this relation participate in predicate locking? Temporary and system
495 : : * relations are exempt.
496 : : */
497 : : static inline bool
4687 heikki.linnakangas@i 498 :CBC 149097 : PredicateLockingNeededForRelation(Relation relation)
499 : : {
1004 tgl@sss.pgh.pa.us 500 [ + + ]: 206221 : return !(relation->rd_id < FirstUnpinnedObjectId ||
500 michael@paquier.xyz 501 [ + + ]: 57124 : RelationUsesLocalBuffers(relation));
502 : : }
503 : :
504 : : /*
505 : : * When a public interface method is called for a read, this is the test to
506 : : * see if we should do a quick return.
507 : : *
508 : : * Note: this function has side-effects! If this transaction has been flagged
509 : : * as RO-safe since the last call, we release all predicate locks and reset
510 : : * MySerializableXact. That makes subsequent calls to return quickly.
511 : : *
512 : : * This is marked as 'inline' to eliminate the function call overhead in the
513 : : * common case that serialization is not needed.
514 : : */
515 : : static inline bool
4687 heikki.linnakangas@i 516 : 52274929 : SerializationNeededForRead(Relation relation, Snapshot snapshot)
517 : : {
518 : : /* Nothing to do if this is not a serializable transaction */
519 [ + + ]: 52274929 : if (MySerializableXact == InvalidSerializableXact)
520 : 52132695 : return false;
521 : :
522 : : /*
523 : : * Don't acquire locks or conflict when scanning with a special snapshot.
524 : : * This excludes things like CLUSTER and REINDEX. They use the wholesale
525 : : * functions TransferPredicateLocksToHeapRelation() and
526 : : * CheckTableForSerializableConflictIn() to participate in serialization,
527 : : * but the scans involved don't need serialization.
528 : : */
529 [ + + + - ]: 142234 : if (!IsMVCCSnapshot(snapshot))
530 : 1589 : return false;
531 : :
532 : : /*
533 : : * Check if we have just become "RO-safe". If we have, immediately release
534 : : * all locks as they're not needed anymore. This also resets
535 : : * MySerializableXact, so that subsequent calls to this function can exit
536 : : * quickly.
537 : : *
538 : : * A transaction is flagged as RO_SAFE if all concurrent R/W transactions
539 : : * commit without having conflicts out to an earlier snapshot, thus
540 : : * ensuring that no conflicts are possible for this transaction.
541 : : */
542 [ + + ]: 140645 : if (SxactIsROSafe(MySerializableXact))
543 : : {
1857 tmunro@postgresql.or 544 : 33 : ReleasePredicateLocks(false, true);
4687 heikki.linnakangas@i 545 : 33 : return false;
546 : : }
547 : :
548 : : /* Check if the relation doesn't participate in predicate locking */
549 [ + + ]: 140612 : if (!PredicateLockingNeededForRelation(relation))
550 : 88163 : return false;
551 : :
4686 552 : 52449 : return true; /* no excuse to skip predicate locking */
553 : : }
554 : :
555 : : /*
556 : : * Like SerializationNeededForRead(), but called on writes.
557 : : * The logic is the same, but there is no snapshot and we can't be RO-safe.
558 : : */
559 : : static inline bool
4687 560 : 15308447 : SerializationNeededForWrite(Relation relation)
561 : : {
562 : : /* Nothing to do if this is not a serializable transaction */
563 [ + + ]: 15308447 : if (MySerializableXact == InvalidSerializableXact)
564 : 15300046 : return false;
565 : :
566 : : /* Check if the relation doesn't participate in predicate locking */
567 [ + + ]: 8401 : if (!PredicateLockingNeededForRelation(relation))
568 : 3937 : return false;
569 : :
4686 570 : 4464 : return true; /* no excuse to skip predicate locking */
571 : : }
572 : :
573 : :
574 : : /*------------------------------------------------------------------------*/
575 : :
576 : : /*
577 : : * These functions are a simple implementation of a list for this specific
578 : : * type of struct. If there is ever a generalized shared memory list, we
579 : : * should probably switch to that.
580 : : */
581 : : static SERIALIZABLEXACT *
4815 582 : 2551 : CreatePredXact(void)
583 : : {
584 : : SERIALIZABLEXACT *sxact;
585 : :
451 andres@anarazel.de 586 [ - + ]: 2551 : if (dlist_is_empty(&PredXact->availableList))
4815 heikki.linnakangas@i 587 :UBC 0 : return NULL;
588 : :
451 andres@anarazel.de 589 :CBC 2551 : sxact = dlist_container(SERIALIZABLEXACT, xactLink,
590 : : dlist_pop_head_node(&PredXact->availableList));
591 : 2551 : dlist_push_tail(&PredXact->activeList, &sxact->xactLink);
592 : 2551 : return sxact;
593 : : }
594 : :
595 : : static void
4815 heikki.linnakangas@i 596 : 1653 : ReleasePredXact(SERIALIZABLEXACT *sxact)
597 : : {
598 [ - + ]: 1653 : Assert(ShmemAddrIsValid(sxact));
599 : :
451 andres@anarazel.de 600 : 1653 : dlist_delete(&sxact->xactLink);
601 : 1653 : dlist_push_tail(&PredXact->availableList, &sxact->xactLink);
4815 heikki.linnakangas@i 602 : 1653 : }
603 : :
604 : : /*------------------------------------------------------------------------*/
605 : :
606 : : /*
607 : : * These functions manage primitive access to the RWConflict pool and lists.
608 : : */
609 : : static bool
4680 tgl@sss.pgh.pa.us 610 : 2648 : RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer)
611 : : {
612 : : dlist_iter iter;
613 : :
4815 heikki.linnakangas@i 614 [ - + ]: 2648 : Assert(reader != writer);
615 : :
616 : : /* Check the ends of the purported conflict first. */
4687 617 [ + - ]: 2648 : if (SxactIsDoomed(reader)
618 [ + + ]: 2648 : || SxactIsDoomed(writer)
451 andres@anarazel.de 619 [ + + ]: 2640 : || dlist_is_empty(&reader->outConflicts)
620 [ + + ]: 607 : || dlist_is_empty(&writer->inConflicts))
4815 heikki.linnakangas@i 621 : 2113 : return false;
622 : :
623 : : /*
624 : : * A conflict is possible; walk the list to find out.
625 : : *
626 : : * The unconstify is needed as we have no const version of
627 : : * dlist_foreach().
628 : : */
451 andres@anarazel.de 629 [ + - + + ]: 559 : dlist_foreach(iter, &unconstify(SERIALIZABLEXACT *, reader)->outConflicts)
630 : : {
631 : 535 : RWConflict conflict =
331 tgl@sss.pgh.pa.us 632 : 535 : dlist_container(RWConflictData, outLink, iter.cur);
633 : :
4815 heikki.linnakangas@i 634 [ + + ]: 535 : if (conflict->sxactIn == writer)
635 : 511 : return true;
636 : : }
637 : :
638 : : /* No conflict found. */
639 : 24 : return false;
640 : : }
641 : :
642 : : static void
643 : 780 : SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
644 : : {
645 : : RWConflict conflict;
646 : :
647 [ - + ]: 780 : Assert(reader != writer);
648 [ - + ]: 780 : Assert(!RWConflictExists(reader, writer));
649 : :
451 andres@anarazel.de 650 [ - + ]: 780 : if (dlist_is_empty(&RWConflictPool->availableList))
4815 heikki.linnakangas@i 651 [ # # ]:UBC 0 : ereport(ERROR,
652 : : (errcode(ERRCODE_OUT_OF_MEMORY),
653 : : errmsg("not enough elements in RWConflictPool to record a read/write conflict"),
654 : : errhint("You might need to run fewer transactions at a time or increase max_connections.")));
655 : :
451 andres@anarazel.de 656 :CBC 780 : conflict = dlist_head_element(RWConflictData, outLink, &RWConflictPool->availableList);
657 : 780 : dlist_delete(&conflict->outLink);
658 : :
4815 heikki.linnakangas@i 659 : 780 : conflict->sxactOut = reader;
660 : 780 : conflict->sxactIn = writer;
451 andres@anarazel.de 661 : 780 : dlist_push_tail(&reader->outConflicts, &conflict->outLink);
662 : 780 : dlist_push_tail(&writer->inConflicts, &conflict->inLink);
4815 heikki.linnakangas@i 663 : 780 : }
664 : :
665 : : static void
666 : 132 : SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact,
667 : : SERIALIZABLEXACT *activeXact)
668 : : {
669 : : RWConflict conflict;
670 : :
671 [ - + ]: 132 : Assert(roXact != activeXact);
672 [ - + ]: 132 : Assert(SxactIsReadOnly(roXact));
673 [ - + ]: 132 : Assert(!SxactIsReadOnly(activeXact));
674 : :
451 andres@anarazel.de 675 [ - + ]: 132 : if (dlist_is_empty(&RWConflictPool->availableList))
4815 heikki.linnakangas@i 676 [ # # ]:UBC 0 : ereport(ERROR,
677 : : (errcode(ERRCODE_OUT_OF_MEMORY),
678 : : errmsg("not enough elements in RWConflictPool to record a potential read/write conflict"),
679 : : errhint("You might need to run fewer transactions at a time or increase max_connections.")));
680 : :
451 andres@anarazel.de 681 :CBC 132 : conflict = dlist_head_element(RWConflictData, outLink, &RWConflictPool->availableList);
682 : 132 : dlist_delete(&conflict->outLink);
683 : :
4815 heikki.linnakangas@i 684 : 132 : conflict->sxactOut = activeXact;
685 : 132 : conflict->sxactIn = roXact;
451 andres@anarazel.de 686 : 132 : dlist_push_tail(&activeXact->possibleUnsafeConflicts, &conflict->outLink);
687 : 132 : dlist_push_tail(&roXact->possibleUnsafeConflicts, &conflict->inLink);
4815 heikki.linnakangas@i 688 : 132 : }
689 : :
690 : : static void
691 : 912 : ReleaseRWConflict(RWConflict conflict)
692 : : {
451 andres@anarazel.de 693 : 912 : dlist_delete(&conflict->inLink);
694 : 912 : dlist_delete(&conflict->outLink);
695 : 912 : dlist_push_tail(&RWConflictPool->availableList, &conflict->outLink);
4815 heikki.linnakangas@i 696 : 912 : }
697 : :
698 : : static void
699 : 3 : FlagSxactUnsafe(SERIALIZABLEXACT *sxact)
700 : : {
701 : : dlist_mutable_iter iter;
702 : :
703 [ - + ]: 3 : Assert(SxactIsReadOnly(sxact));
704 [ - + ]: 3 : Assert(!SxactIsROSafe(sxact));
705 : :
706 : 3 : sxact->flags |= SXACT_FLAG_RO_UNSAFE;
707 : :
708 : : /*
709 : : * We know this isn't a safe snapshot, so we can stop looking for other
710 : : * potential conflicts.
711 : : */
451 andres@anarazel.de 712 [ + - + + ]: 6 : dlist_foreach_modify(iter, &sxact->possibleUnsafeConflicts)
713 : : {
714 : 3 : RWConflict conflict =
331 tgl@sss.pgh.pa.us 715 : 3 : dlist_container(RWConflictData, inLink, iter.cur);
716 : :
4815 heikki.linnakangas@i 717 [ - + ]: 3 : Assert(!SxactIsReadOnly(conflict->sxactOut));
718 [ - + ]: 3 : Assert(sxact == conflict->sxactIn);
719 : :
720 : 3 : ReleaseRWConflict(conflict);
721 : : }
722 : 3 : }
723 : :
724 : : /*------------------------------------------------------------------------*/
725 : :
726 : : /*
727 : : * Decide whether a Serial page number is "older" for truncation purposes.
728 : : * Analogous to CLOGPagePrecedes().
729 : : */
730 : : static bool
137 akorotkov@postgresql 731 :GNC 36818 : SerialPagePrecedesLogically(int64 page1, int64 page2)
732 : : {
733 : : TransactionId xid1;
734 : : TransactionId xid2;
735 : :
1184 noah@leadboat.com 736 :CBC 36818 : xid1 = ((TransactionId) page1) * SERIAL_ENTRIESPERPAGE;
737 : 36818 : xid1 += FirstNormalTransactionId + 1;
738 : 36818 : xid2 = ((TransactionId) page2) * SERIAL_ENTRIESPERPAGE;
739 : 36818 : xid2 += FirstNormalTransactionId + 1;
740 : :
741 [ + + + + ]: 61962 : return (TransactionIdPrecedes(xid1, xid2) &&
742 : 25144 : TransactionIdPrecedes(xid1, xid2 + SERIAL_ENTRIESPERPAGE - 1));
743 : : }
744 : :
745 : : #ifdef USE_ASSERT_CHECKING
746 : : static void
747 : 898 : SerialPagePrecedesLogicallyUnitTests(void)
748 : : {
749 : 898 : int per_page = SERIAL_ENTRIESPERPAGE,
750 : 898 : offset = per_page / 2;
751 : : int64 newestPage,
752 : : oldestPage,
753 : : headPage,
754 : : targetPage;
755 : : TransactionId newestXact,
756 : : oldestXact;
757 : :
758 : : /* GetNewTransactionId() has assigned the last XID it can safely use. */
759 : 898 : newestPage = 2 * SLRU_PAGES_PER_SEGMENT - 1; /* nothing special */
760 : 898 : newestXact = newestPage * per_page + offset;
761 [ - + ]: 898 : Assert(newestXact / per_page == newestPage);
762 : 898 : oldestXact = newestXact + 1;
763 : 898 : oldestXact -= 1U << 31;
764 : 898 : oldestPage = oldestXact / per_page;
765 : :
766 : : /*
767 : : * In this scenario, the SLRU headPage pertains to the last ~1000 XIDs
768 : : * assigned. oldestXact finishes, ~2B XIDs having elapsed since it
769 : : * started. Further transactions cause us to summarize oldestXact to
770 : : * tailPage. Function must return false so SerialAdd() doesn't zero
771 : : * tailPage (which may contain entries for other old, recently-finished
772 : : * XIDs) and half the SLRU. Reaching this requires burning ~2B XIDs in
773 : : * single-user mode, a negligible possibility.
774 : : */
775 : 898 : headPage = newestPage;
776 : 898 : targetPage = oldestPage;
777 [ - + ]: 898 : Assert(!SerialPagePrecedesLogically(headPage, targetPage));
778 : :
779 : : /*
780 : : * In this scenario, the SLRU headPage pertains to oldestXact. We're
781 : : * summarizing an XID near newestXact. (Assume few other XIDs used
782 : : * SERIALIZABLE, hence the minimal headPage advancement. Assume
783 : : * oldestXact was long-running and only recently reached the SLRU.)
784 : : * Function must return true to make SerialAdd() create targetPage.
785 : : *
786 : : * Today's implementation mishandles this case, but it doesn't matter
787 : : * enough to fix. Verify that the defect affects just one page by
788 : : * asserting correct treatment of its prior page. Reaching this case
789 : : * requires burning ~2B XIDs in single-user mode, a negligible
790 : : * possibility. Moreover, if it does happen, the consequence would be
791 : : * mild, namely a new transaction failing in SimpleLruReadPage().
792 : : */
793 : 898 : headPage = oldestPage;
794 : 898 : targetPage = newestPage;
795 [ - + ]: 898 : Assert(SerialPagePrecedesLogically(headPage, targetPage - 1));
796 : : #if 0
797 : : Assert(SerialPagePrecedesLogically(headPage, targetPage));
798 : : #endif
4815 heikki.linnakangas@i 799 : 898 : }
800 : : #endif
801 : :
802 : : /*
803 : : * Initialize for the tracking of old serializable committed xids.
804 : : */
805 : : static void
1430 tgl@sss.pgh.pa.us 806 : 898 : SerialInit(void)
807 : : {
808 : : bool found;
809 : :
810 : : /*
811 : : * Set up SLRU management of the pg_serial data.
812 : : */
813 : 898 : SerialSlruCtl->PagePrecedes = SerialPagePrecedesLogically;
46 alvherre@alvh.no-ip. 814 :GNC 898 : SimpleLruInit(SerialSlruCtl, "serializable",
815 : : serializable_buffers, 0, "pg_serial",
816 : : LWTRANCHE_SERIAL_BUFFER, LWTRANCHE_SERIAL_SLRU,
817 : : SYNC_HANDLER_NONE, false);
818 : : #ifdef USE_ASSERT_CHECKING
1184 noah@leadboat.com 819 :CBC 898 : SerialPagePrecedesLogicallyUnitTests();
820 : : #endif
821 : 898 : SlruPagePrecedesUnitTests(SerialSlruCtl, SERIAL_ENTRIESPERPAGE);
822 : :
823 : : /*
824 : : * Create or attach to the SerialControl structure.
825 : : */
1430 tgl@sss.pgh.pa.us 826 : 898 : serialControl = (SerialControl)
827 : 898 : ShmemInitStruct("SerialControlData", sizeof(SerialControlData), &found);
828 : :
2456 829 [ - + ]: 898 : Assert(found == IsUnderPostmaster);
4815 heikki.linnakangas@i 830 [ + - ]: 898 : if (!found)
831 : : {
832 : : /*
833 : : * Set control information to reflect empty SLRU.
834 : : */
75 alvherre@alvh.no-ip. 835 :GNC 898 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
1430 tgl@sss.pgh.pa.us 836 :CBC 898 : serialControl->headPage = -1;
837 : 898 : serialControl->headXid = InvalidTransactionId;
838 : 898 : serialControl->tailXid = InvalidTransactionId;
75 alvherre@alvh.no-ip. 839 :GNC 898 : LWLockRelease(SerialControlLock);
840 : : }
4815 heikki.linnakangas@i 841 :CBC 898 : }
842 : :
843 : : /*
844 : : * GUC check_hook for serializable_buffers
845 : : */
846 : : bool
46 alvherre@alvh.no-ip. 847 :GNC 928 : check_serial_buffers(int *newval, void **extra, GucSource source)
848 : : {
849 : 928 : return check_slru_buffers("serializable_buffers", newval);
850 : : }
851 : :
852 : : /*
853 : : * Record a committed read write serializable xid and the minimum
854 : : * commitSeqNo of any transactions to which this xid had a rw-conflict out.
855 : : * An invalid commitSeqNo means that there were no conflicts out from xid.
856 : : */
857 : : static void
1430 tgl@sss.pgh.pa.us 858 :UBC 0 : SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo)
859 : : {
860 : : TransactionId tailXid;
861 : : int64 targetPage;
862 : : int slotno;
863 : : int64 firstZeroPage;
864 : : bool isNewPage;
865 : : LWLock *lock;
866 : :
4815 heikki.linnakangas@i 867 [ # # ]: 0 : Assert(TransactionIdIsValid(xid));
868 : :
1430 tgl@sss.pgh.pa.us 869 : 0 : targetPage = SerialPage(xid);
46 alvherre@alvh.no-ip. 870 :UNC 0 : lock = SimpleLruGetBankLock(SerialSlruCtl, targetPage);
871 : :
872 : : /*
873 : : * In this routine, we must hold both SerialControlLock and the SLRU bank
874 : : * lock simultaneously while making the SLRU data catch up with the new
875 : : * state that we determine.
876 : : */
75 877 : 0 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
878 : :
879 : : /*
880 : : * If no serializable transactions are active, there shouldn't be anything
881 : : * to push out to the SLRU. Hitting this assert would mean there's
882 : : * something wrong with the earlier cleanup logic.
883 : : */
1430 tgl@sss.pgh.pa.us 884 :UBC 0 : tailXid = serialControl->tailXid;
4815 heikki.linnakangas@i 885 [ # # ]: 0 : Assert(TransactionIdIsValid(tailXid));
886 : :
887 : : /*
888 : : * If the SLRU is currently unused, zero out the whole active region from
889 : : * tailXid to headXid before taking it into use. Otherwise zero out only
890 : : * any new pages that enter the tailXid-headXid range as we advance
891 : : * headXid.
892 : : */
1430 tgl@sss.pgh.pa.us 893 [ # # ]: 0 : if (serialControl->headPage < 0)
894 : : {
895 : 0 : firstZeroPage = SerialPage(tailXid);
4815 heikki.linnakangas@i 896 : 0 : isNewPage = true;
897 : : }
898 : : else
899 : : {
1430 tgl@sss.pgh.pa.us 900 [ # # ]: 0 : firstZeroPage = SerialNextPage(serialControl->headPage);
901 : 0 : isNewPage = SerialPagePrecedesLogically(serialControl->headPage,
902 : : targetPage);
903 : : }
904 : :
905 [ # # ]: 0 : if (!TransactionIdIsValid(serialControl->headXid)
906 [ # # ]: 0 : || TransactionIdFollows(xid, serialControl->headXid))
907 : 0 : serialControl->headXid = xid;
4786 heikki.linnakangas@i 908 [ # # ]: 0 : if (isNewPage)
1430 tgl@sss.pgh.pa.us 909 : 0 : serialControl->headPage = targetPage;
910 : :
4815 heikki.linnakangas@i 911 [ # # ]: 0 : if (isNewPage)
912 : : {
913 : : /* Initialize intervening pages; might involve trading locks */
914 : : for (;;)
915 : : {
11 alvherre@alvh.no-ip. 916 :UNC 0 : lock = SimpleLruGetBankLock(SerialSlruCtl, firstZeroPage);
917 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
918 : 0 : slotno = SimpleLruZeroPage(SerialSlruCtl, firstZeroPage);
919 [ # # ]: 0 : if (firstZeroPage == targetPage)
920 : 0 : break;
1430 tgl@sss.pgh.pa.us 921 [ # # ]:UBC 0 : firstZeroPage = SerialNextPage(firstZeroPage);
11 alvherre@alvh.no-ip. 922 :UNC 0 : LWLockRelease(lock);
923 : : }
924 : : }
925 : : else
926 : : {
927 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
1430 tgl@sss.pgh.pa.us 928 :UBC 0 : slotno = SimpleLruReadPage(SerialSlruCtl, targetPage, true, xid);
929 : : }
930 : :
931 : 0 : SerialValue(slotno, xid) = minConflictCommitSeqNo;
932 : 0 : SerialSlruCtl->shared->page_dirty[slotno] = true;
933 : :
46 alvherre@alvh.no-ip. 934 :UNC 0 : LWLockRelease(lock);
75 935 : 0 : LWLockRelease(SerialControlLock);
4815 heikki.linnakangas@i 936 :UBC 0 : }
937 : :
938 : : /*
939 : : * Get the minimum commitSeqNo for any conflict out for the given xid. For
940 : : * a transaction which exists but has no conflict out, InvalidSerCommitSeqNo
941 : : * will be returned.
942 : : */
943 : : static SerCommitSeqNo
1430 tgl@sss.pgh.pa.us 944 :CBC 25 : SerialGetMinConflictCommitSeqNo(TransactionId xid)
945 : : {
946 : : TransactionId headXid;
947 : : TransactionId tailXid;
948 : : SerCommitSeqNo val;
949 : : int slotno;
950 : :
4815 heikki.linnakangas@i 951 [ - + ]: 25 : Assert(TransactionIdIsValid(xid));
952 : :
75 alvherre@alvh.no-ip. 953 :GNC 25 : LWLockAcquire(SerialControlLock, LW_SHARED);
1430 tgl@sss.pgh.pa.us 954 :CBC 25 : headXid = serialControl->headXid;
955 : 25 : tailXid = serialControl->tailXid;
75 alvherre@alvh.no-ip. 956 :GNC 25 : LWLockRelease(SerialControlLock);
957 : :
4815 heikki.linnakangas@i 958 [ + - ]:CBC 25 : if (!TransactionIdIsValid(headXid))
959 : 25 : return 0;
960 : :
4815 heikki.linnakangas@i 961 [ # # ]:UBC 0 : Assert(TransactionIdIsValid(tailXid));
962 : :
963 [ # # ]: 0 : if (TransactionIdPrecedes(xid, tailXid)
964 [ # # ]: 0 : || TransactionIdFollows(xid, headXid))
965 : 0 : return 0;
966 : :
967 : : /*
968 : : * The following function must be called without holding SLRU bank lock,
969 : : * but will return with that lock held, which must then be released.
970 : : */
1430 tgl@sss.pgh.pa.us 971 : 0 : slotno = SimpleLruReadPage_ReadOnly(SerialSlruCtl,
972 : : SerialPage(xid), xid);
973 : 0 : val = SerialValue(slotno, xid);
46 alvherre@alvh.no-ip. 974 :UNC 0 : LWLockRelease(SimpleLruGetBankLock(SerialSlruCtl, SerialPage(xid)));
4815 heikki.linnakangas@i 975 :UBC 0 : return val;
976 : : }
977 : :
978 : : /*
979 : : * Call this whenever there is a new xmin for active serializable
980 : : * transactions. We don't need to keep information on transactions which
981 : : * precede that. InvalidTransactionId means none active, so everything in
982 : : * the SLRU can be discarded.
983 : : */
984 : : static void
1430 tgl@sss.pgh.pa.us 985 :CBC 1690 : SerialSetActiveSerXmin(TransactionId xid)
986 : : {
75 alvherre@alvh.no-ip. 987 :GNC 1690 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
988 : :
989 : : /*
990 : : * When no sxacts are active, nothing overlaps, set the xid values to
991 : : * invalid to show that there are no valid entries. Don't clear headPage,
992 : : * though. A new xmin might still land on that page, and we don't want to
993 : : * repeatedly zero out the same page.
994 : : */
4815 heikki.linnakangas@i 995 [ + + ]:CBC 1690 : if (!TransactionIdIsValid(xid))
996 : : {
1430 tgl@sss.pgh.pa.us 997 : 836 : serialControl->tailXid = InvalidTransactionId;
998 : 836 : serialControl->headXid = InvalidTransactionId;
75 alvherre@alvh.no-ip. 999 :GNC 836 : LWLockRelease(SerialControlLock);
4815 heikki.linnakangas@i 1000 :CBC 836 : return;
1001 : : }
1002 : :
1003 : : /*
1004 : : * When we're recovering prepared transactions, the global xmin might move
1005 : : * backwards depending on the order they're recovered. Normally that's not
1006 : : * OK, but during recovery no serializable transactions will commit, so
1007 : : * the SLRU is empty and we can get away with it.
1008 : : */
1009 [ - + ]: 854 : if (RecoveryInProgress())
1010 : : {
1430 tgl@sss.pgh.pa.us 1011 [ # # ]:UBC 0 : Assert(serialControl->headPage < 0);
1012 [ # # ]: 0 : if (!TransactionIdIsValid(serialControl->tailXid)
1013 [ # # ]: 0 : || TransactionIdPrecedes(xid, serialControl->tailXid))
1014 : : {
1015 : 0 : serialControl->tailXid = xid;
1016 : : }
75 alvherre@alvh.no-ip. 1017 :UNC 0 : LWLockRelease(SerialControlLock);
4815 heikki.linnakangas@i 1018 :UBC 0 : return;
1019 : : }
1020 : :
1430 tgl@sss.pgh.pa.us 1021 [ + + - + ]:CBC 854 : Assert(!TransactionIdIsValid(serialControl->tailXid)
1022 : : || TransactionIdFollows(xid, serialControl->tailXid));
1023 : :
1024 : 854 : serialControl->tailXid = xid;
1025 : :
75 alvherre@alvh.no-ip. 1026 :GNC 854 : LWLockRelease(SerialControlLock);
1027 : : }
1028 : :
1029 : : /*
1030 : : * Perform a checkpoint --- either during shutdown, or on-the-fly
1031 : : *
1032 : : * We don't have any data that needs to survive a restart, but this is a
1033 : : * convenient place to truncate the SLRU.
1034 : : */
1035 : : void
4786 heikki.linnakangas@i 1036 :CBC 1153 : CheckPointPredicate(void)
1037 : : {
1038 : : int truncateCutoffPage;
1039 : :
75 alvherre@alvh.no-ip. 1040 :GNC 1153 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
1041 : :
1042 : : /* Exit quickly if the SLRU is currently not in use. */
1430 tgl@sss.pgh.pa.us 1043 [ + - ]:CBC 1153 : if (serialControl->headPage < 0)
1044 : : {
75 alvherre@alvh.no-ip. 1045 :GNC 1153 : LWLockRelease(SerialControlLock);
4815 heikki.linnakangas@i 1046 :CBC 1153 : return;
1047 : : }
1048 : :
1430 tgl@sss.pgh.pa.us 1049 [ # # ]:UBC 0 : if (TransactionIdIsValid(serialControl->tailXid))
1050 : : {
1051 : : int tailPage;
1052 : :
1053 : 0 : tailPage = SerialPage(serialControl->tailXid);
1054 : :
1055 : : /*
1056 : : * It is possible for the tailXid to be ahead of the headXid. This
1057 : : * occurs if we checkpoint while there are in-progress serializable
1058 : : * transaction(s) advancing the tail but we are yet to summarize the
1059 : : * transactions. In this case, we cutoff up to the headPage and the
1060 : : * next summary will advance the headXid.
1061 : : */
180 michael@paquier.xyz 1062 [ # # ]:UNC 0 : if (SerialPagePrecedesLogically(tailPage, serialControl->headPage))
1063 : : {
1064 : : /* We can truncate the SLRU up to the page containing tailXid */
1065 : 0 : truncateCutoffPage = tailPage;
1066 : : }
1067 : : else
1068 : 0 : truncateCutoffPage = serialControl->headPage;
1069 : : }
1070 : : else
1071 : : {
1072 : : /*----------
1073 : : * The SLRU is no longer needed. Truncate to head before we set head
1074 : : * invalid.
1075 : : *
1076 : : * XXX: It's possible that the SLRU is not needed again until XID
1077 : : * wrap-around has happened, so that the segment containing headPage
1078 : : * that we leave behind will appear to be new again. In that case it
1079 : : * won't be removed until XID horizon advances enough to make it
1080 : : * current again.
1081 : : *
1082 : : * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1083 : : * Consider this scenario, starting from a system with no in-progress
1084 : : * transactions and VACUUM FREEZE having maximized oldestXact:
1085 : : * - Start a SERIALIZABLE transaction.
1086 : : * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1087 : : * one SLRU page.
1088 : : * - Consume XIDs to reach xidStopLimit.
1089 : : * - Finish all transactions. Due to the long-running SERIALIZABLE
1090 : : * transaction, earlier checkpoints did not touch headPage. The
1091 : : * next checkpoint will change it, but that checkpoint happens after
1092 : : * the end of the scenario.
1093 : : * - VACUUM to advance XID limits.
1094 : : * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1095 : : * - Start, finish, and summarize a SERIALIZABLE transaction.
1096 : : * SerialAdd() declines to create the targetPage, because headPage
1097 : : * is not regarded as in the past relative to that targetPage. The
1098 : : * transaction instigating the summarize fails in
1099 : : * SimpleLruReadPage().
1100 : : */
1101 : 0 : truncateCutoffPage = serialControl->headPage;
1430 tgl@sss.pgh.pa.us 1102 :UBC 0 : serialControl->headPage = -1;
1103 : : }
1104 : :
75 alvherre@alvh.no-ip. 1105 :UNC 0 : LWLockRelease(SerialControlLock);
1106 : :
1107 : : /*
1108 : : * Truncate away pages that are no longer required. Note that no
1109 : : * additional locking is required, because this is only called as part of
1110 : : * a checkpoint, and the validity limits have already been determined.
1111 : : */
180 michael@paquier.xyz 1112 : 0 : SimpleLruTruncate(SerialSlruCtl, truncateCutoffPage);
1113 : :
1114 : : /*
1115 : : * Write dirty SLRU pages to disk
1116 : : *
1117 : : * This is not actually necessary from a correctness point of view. We do
1118 : : * it merely as a debugging aid.
1119 : : *
1120 : : * We're doing this after the truncation to avoid writing pages right
1121 : : * before deleting the file in which they sit, which would be completely
1122 : : * pointless.
1123 : : */
1297 tmunro@postgresql.or 1124 :UBC 0 : SimpleLruWriteAll(SerialSlruCtl, true);
1125 : : }
1126 : :
1127 : : /*------------------------------------------------------------------------*/
1128 : :
1129 : : /*
1130 : : * InitPredicateLocks -- Initialize the predicate locking data structures.
1131 : : *
1132 : : * This is called from CreateSharedMemoryAndSemaphores(), which see for
1133 : : * more comments. In the normal postmaster case, the shared hash tables
1134 : : * are created here. Backends inherit the pointers
1135 : : * to the shared tables via fork(). In the EXEC_BACKEND case, each
1136 : : * backend re-executes this code to obtain pointers to the already existing
1137 : : * shared hash tables.
1138 : : */
1139 : : void
4815 heikki.linnakangas@i 1140 :CBC 898 : InitPredicateLocks(void)
1141 : : {
1142 : : HASHCTL info;
1143 : : long max_table_size;
1144 : : Size requestSize;
1145 : : bool found;
1146 : :
1147 : : #ifndef EXEC_BACKEND
2456 tgl@sss.pgh.pa.us 1148 [ - + ]: 898 : Assert(!IsUnderPostmaster);
1149 : : #endif
1150 : :
1151 : : /*
1152 : : * Compute size of predicate lock target hashtable. Note these
1153 : : * calculations must agree with PredicateLockShmemSize!
1154 : : */
4815 heikki.linnakangas@i 1155 : 898 : max_table_size = NPREDICATELOCKTARGETENTS();
1156 : :
1157 : : /*
1158 : : * Allocate hash table for PREDICATELOCKTARGET structs. This stores
1159 : : * per-predicate-lock-target information.
1160 : : */
1161 : 898 : info.keysize = sizeof(PREDICATELOCKTARGETTAG);
1162 : 898 : info.entrysize = sizeof(PREDICATELOCKTARGET);
1163 : 898 : info.num_partitions = NUM_PREDICATELOCK_PARTITIONS;
1164 : :
1165 : 898 : PredicateLockTargetHash = ShmemInitHash("PREDICATELOCKTARGET hash",
1166 : : max_table_size,
1167 : : max_table_size,
1168 : : &info,
1169 : : HASH_ELEM | HASH_BLOBS |
1170 : : HASH_PARTITION | HASH_FIXED_SIZE);
1171 : :
1172 : : /*
1173 : : * Reserve a dummy entry in the hash table; we use it to make sure there's
1174 : : * always one entry available when we need to split or combine a page,
1175 : : * because running out of space there could mean aborting a
1176 : : * non-serializable transaction.
1177 : : */
2456 tgl@sss.pgh.pa.us 1178 [ + - ]: 898 : if (!IsUnderPostmaster)
1179 : : {
1180 : 898 : (void) hash_search(PredicateLockTargetHash, &ScratchTargetTag,
1181 : : HASH_ENTER, &found);
1182 [ - + ]: 898 : Assert(!found);
1183 : : }
1184 : :
1185 : : /* Pre-calculate the hash and partition lock of the scratch entry */
1186 : 898 : ScratchTargetTagHash = PredicateLockTargetTagHashCode(&ScratchTargetTag);
1187 : 898 : ScratchPartitionLock = PredicateLockHashPartitionLock(ScratchTargetTagHash);
1188 : :
1189 : : /*
1190 : : * Allocate hash table for PREDICATELOCK structs. This stores per
1191 : : * xact-lock-of-a-target information.
1192 : : */
4815 heikki.linnakangas@i 1193 : 898 : info.keysize = sizeof(PREDICATELOCKTAG);
1194 : 898 : info.entrysize = sizeof(PREDICATELOCK);
1195 : 898 : info.hash = predicatelock_hash;
1196 : 898 : info.num_partitions = NUM_PREDICATELOCK_PARTITIONS;
1197 : :
1198 : : /* Assume an average of 2 xacts per target */
2456 tgl@sss.pgh.pa.us 1199 : 898 : max_table_size *= 2;
1200 : :
4815 heikki.linnakangas@i 1201 : 898 : PredicateLockHash = ShmemInitHash("PREDICATELOCK hash",
1202 : : max_table_size,
1203 : : max_table_size,
1204 : : &info,
1205 : : HASH_ELEM | HASH_FUNCTION |
1206 : : HASH_PARTITION | HASH_FIXED_SIZE);
1207 : :
1208 : : /*
1209 : : * Compute size for serializable transaction hashtable. Note these
1210 : : * calculations must agree with PredicateLockShmemSize!
1211 : : */
733 rhaas@postgresql.org 1212 : 898 : max_table_size = (MaxBackends + max_prepared_xacts);
1213 : :
1214 : : /*
1215 : : * Allocate a list to hold information on transactions participating in
1216 : : * predicate locking.
1217 : : *
1218 : : * Assume an average of 10 predicate locking transactions per backend.
1219 : : * This allows aggressive cleanup while detail is present before data must
1220 : : * be summarized for storage in SLRU and the "dummy" transaction.
1221 : : */
4815 heikki.linnakangas@i 1222 : 898 : max_table_size *= 10;
1223 : :
1224 : 898 : PredXact = ShmemInitStruct("PredXactList",
1225 : : PredXactListDataSize,
1226 : : &found);
2456 tgl@sss.pgh.pa.us 1227 [ - + ]: 898 : Assert(found == IsUnderPostmaster);
4815 heikki.linnakangas@i 1228 [ + - ]: 898 : if (!found)
1229 : : {
1230 : : int i;
1231 : :
451 andres@anarazel.de 1232 : 898 : dlist_init(&PredXact->availableList);
1233 : 898 : dlist_init(&PredXact->activeList);
4815 heikki.linnakangas@i 1234 : 898 : PredXact->SxactGlobalXmin = InvalidTransactionId;
1235 : 898 : PredXact->SxactGlobalXminCount = 0;
1236 : 898 : PredXact->WritableSxactCount = 0;
1237 : 898 : PredXact->LastSxactCommitSeqNo = FirstNormalSerCommitSeqNo - 1;
1238 : 898 : PredXact->CanPartialClearThrough = 0;
1239 : 898 : PredXact->HavePartialClearedThrough = 0;
1240 : 898 : requestSize = mul_size((Size) max_table_size,
1241 : : sizeof(SERIALIZABLEXACT));
1242 : 898 : PredXact->element = ShmemAlloc(requestSize);
1243 : : /* Add all elements to available list, clean. */
1244 : 898 : memset(PredXact->element, 0, requestSize);
1245 [ + + ]: 753128 : for (i = 0; i < max_table_size; i++)
1246 : : {
451 andres@anarazel.de 1247 : 752230 : LWLockInitialize(&PredXact->element[i].perXactPredicateListLock,
1248 : : LWTRANCHE_PER_XACT_PREDICATE_LIST);
1249 : 752230 : dlist_push_tail(&PredXact->availableList, &PredXact->element[i].xactLink);
1250 : : }
4815 heikki.linnakangas@i 1251 : 898 : PredXact->OldCommittedSxact = CreatePredXact();
1252 : 898 : SetInvalidVirtualTransactionId(PredXact->OldCommittedSxact->vxid);
4665 1253 : 898 : PredXact->OldCommittedSxact->prepareSeqNo = 0;
4815 1254 : 898 : PredXact->OldCommittedSxact->commitSeqNo = 0;
1255 : 898 : PredXact->OldCommittedSxact->SeqNo.lastCommitBeforeSnapshot = 0;
451 andres@anarazel.de 1256 : 898 : dlist_init(&PredXact->OldCommittedSxact->outConflicts);
1257 : 898 : dlist_init(&PredXact->OldCommittedSxact->inConflicts);
1258 : 898 : dlist_init(&PredXact->OldCommittedSxact->predicateLocks);
1259 : 898 : dlist_node_init(&PredXact->OldCommittedSxact->finishedLink);
1260 : 898 : dlist_init(&PredXact->OldCommittedSxact->possibleUnsafeConflicts);
4815 heikki.linnakangas@i 1261 : 898 : PredXact->OldCommittedSxact->topXid = InvalidTransactionId;
1262 : 898 : PredXact->OldCommittedSxact->finishedBefore = InvalidTransactionId;
1263 : 898 : PredXact->OldCommittedSxact->xmin = InvalidTransactionId;
1264 : 898 : PredXact->OldCommittedSxact->flags = SXACT_FLAG_COMMITTED;
1265 : 898 : PredXact->OldCommittedSxact->pid = 0;
42 heikki.linnakangas@i 1266 :GNC 898 : PredXact->OldCommittedSxact->pgprocno = INVALID_PROC_NUMBER;
1267 : : }
1268 : : /* This never changes, so let's keep a local copy. */
4815 heikki.linnakangas@i 1269 :CBC 898 : OldCommittedSxact = PredXact->OldCommittedSxact;
1270 : :
1271 : : /*
1272 : : * Allocate hash table for SERIALIZABLEXID structs. This stores per-xid
1273 : : * information for serializable transactions which have accessed data.
1274 : : */
1275 : 898 : info.keysize = sizeof(SERIALIZABLEXIDTAG);
1276 : 898 : info.entrysize = sizeof(SERIALIZABLEXID);
1277 : :
1278 : 898 : SerializableXidHash = ShmemInitHash("SERIALIZABLEXID hash",
1279 : : max_table_size,
1280 : : max_table_size,
1281 : : &info,
1282 : : HASH_ELEM | HASH_BLOBS |
1283 : : HASH_FIXED_SIZE);
1284 : :
1285 : : /*
1286 : : * Allocate space for tracking rw-conflicts in lists attached to the
1287 : : * transactions.
1288 : : *
1289 : : * Assume an average of 5 conflicts per transaction. Calculations suggest
1290 : : * that this will prevent resource exhaustion in even the most pessimal
1291 : : * loads up to max_connections = 200 with all 200 connections pounding the
1292 : : * database with serializable transactions. Beyond that, there may be
1293 : : * occasional transactions canceled when trying to flag conflicts. That's
1294 : : * probably OK.
1295 : : */
1296 : 898 : max_table_size *= 5;
1297 : :
1298 : 898 : RWConflictPool = ShmemInitStruct("RWConflictPool",
1299 : : RWConflictPoolHeaderDataSize,
1300 : : &found);
2456 tgl@sss.pgh.pa.us 1301 [ - + ]: 898 : Assert(found == IsUnderPostmaster);
4815 heikki.linnakangas@i 1302 [ + - ]: 898 : if (!found)
1303 : : {
1304 : : int i;
1305 : :
451 andres@anarazel.de 1306 : 898 : dlist_init(&RWConflictPool->availableList);
4815 heikki.linnakangas@i 1307 : 898 : requestSize = mul_size((Size) max_table_size,
1308 : : RWConflictDataSize);
1309 : 898 : RWConflictPool->element = ShmemAlloc(requestSize);
1310 : : /* Add all elements to available list, clean. */
1311 : 898 : memset(RWConflictPool->element, 0, requestSize);
1312 [ + + ]: 3762048 : for (i = 0; i < max_table_size; i++)
1313 : : {
451 andres@anarazel.de 1314 : 3761150 : dlist_push_tail(&RWConflictPool->availableList,
1315 : 3761150 : &RWConflictPool->element[i].outLink);
1316 : : }
1317 : : }
1318 : :
1319 : : /*
1320 : : * Create or attach to the header for the list of finished serializable
1321 : : * transactions.
1322 : : */
1323 : 898 : FinishedSerializableTransactions = (dlist_head *)
4815 heikki.linnakangas@i 1324 : 898 : ShmemInitStruct("FinishedSerializableTransactions",
1325 : : sizeof(dlist_head),
1326 : : &found);
2456 tgl@sss.pgh.pa.us 1327 [ - + ]: 898 : Assert(found == IsUnderPostmaster);
4815 heikki.linnakangas@i 1328 [ + - ]: 898 : if (!found)
451 andres@anarazel.de 1329 : 898 : dlist_init(FinishedSerializableTransactions);
1330 : :
1331 : : /*
1332 : : * Initialize the SLRU storage for old committed serializable
1333 : : * transactions.
1334 : : */
1430 tgl@sss.pgh.pa.us 1335 : 898 : SerialInit();
4815 heikki.linnakangas@i 1336 : 898 : }
1337 : :
1338 : : /*
1339 : : * Estimate shared-memory space used for predicate lock table
1340 : : */
1341 : : Size
1342 : 1679 : PredicateLockShmemSize(void)
1343 : : {
1344 : 1679 : Size size = 0;
1345 : : long max_table_size;
1346 : :
1347 : : /* predicate lock target hash table */
1348 : 1679 : max_table_size = NPREDICATELOCKTARGETENTS();
1349 : 1679 : size = add_size(size, hash_estimate_size(max_table_size,
1350 : : sizeof(PREDICATELOCKTARGET)));
1351 : :
1352 : : /* predicate lock hash table */
1353 : 1679 : max_table_size *= 2;
1354 : 1679 : size = add_size(size, hash_estimate_size(max_table_size,
1355 : : sizeof(PREDICATELOCK)));
1356 : :
1357 : : /*
1358 : : * Since NPREDICATELOCKTARGETENTS is only an estimate, add 10% safety
1359 : : * margin.
1360 : : */
1361 : 1679 : size = add_size(size, size / 10);
1362 : :
1363 : : /* transaction list */
733 rhaas@postgresql.org 1364 : 1679 : max_table_size = MaxBackends + max_prepared_xacts;
4815 heikki.linnakangas@i 1365 : 1679 : max_table_size *= 10;
1366 : 1679 : size = add_size(size, PredXactListDataSize);
1367 : 1679 : size = add_size(size, mul_size((Size) max_table_size,
1368 : : sizeof(SERIALIZABLEXACT)));
1369 : :
1370 : : /* transaction xid table */
1371 : 1679 : size = add_size(size, hash_estimate_size(max_table_size,
1372 : : sizeof(SERIALIZABLEXID)));
1373 : :
1374 : : /* rw-conflict pool */
4813 1375 : 1679 : max_table_size *= 5;
1376 : 1679 : size = add_size(size, RWConflictPoolHeaderDataSize);
1377 : 1679 : size = add_size(size, mul_size((Size) max_table_size,
1378 : : RWConflictDataSize));
1379 : :
1380 : : /* Head for list of finished serializable transactions. */
451 andres@anarazel.de 1381 : 1679 : size = add_size(size, sizeof(dlist_head));
1382 : :
1383 : : /* Shared memory structures for SLRU tracking of old committed xids. */
1430 tgl@sss.pgh.pa.us 1384 : 1679 : size = add_size(size, sizeof(SerialControlData));
46 alvherre@alvh.no-ip. 1385 :GNC 1679 : size = add_size(size, SimpleLruShmemSize(serializable_buffers, 0));
1386 : :
4815 heikki.linnakangas@i 1387 :CBC 1679 : return size;
1388 : : }
1389 : :
1390 : :
1391 : : /*
1392 : : * Compute the hash code associated with a PREDICATELOCKTAG.
1393 : : *
1394 : : * Because we want to use just one set of partition locks for both the
1395 : : * PREDICATELOCKTARGET and PREDICATELOCK hash tables, we have to make sure
1396 : : * that PREDICATELOCKs fall into the same partition number as their
1397 : : * associated PREDICATELOCKTARGETs. dynahash.c expects the partition number
1398 : : * to be the low-order bits of the hash code, and therefore a
1399 : : * PREDICATELOCKTAG's hash code must have the same low-order bits as the
1400 : : * associated PREDICATELOCKTARGETTAG's hash code. We achieve this with this
1401 : : * specialized hash function.
1402 : : */
1403 : : static uint32
4815 heikki.linnakangas@i 1404 :UBC 0 : predicatelock_hash(const void *key, Size keysize)
1405 : : {
4680 tgl@sss.pgh.pa.us 1406 : 0 : const PREDICATELOCKTAG *predicatelocktag = (const PREDICATELOCKTAG *) key;
1407 : : uint32 targethash;
1408 : :
4815 heikki.linnakangas@i 1409 [ # # ]: 0 : Assert(keysize == sizeof(PREDICATELOCKTAG));
1410 : :
1411 : : /* Look into the associated target object, and compute its hash code */
1412 : 0 : targethash = PredicateLockTargetTagHashCode(&predicatelocktag->myTarget->tag);
1413 : :
1414 : 0 : return PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash);
1415 : : }
1416 : :
1417 : :
1418 : : /*
1419 : : * GetPredicateLockStatusData
1420 : : * Return a table containing the internal state of the predicate
1421 : : * lock manager for use in pg_lock_status.
1422 : : *
1423 : : * Like GetLockStatusData, this function tries to hold the partition LWLocks
1424 : : * for as short a time as possible by returning two arrays that simply
1425 : : * contain the PREDICATELOCKTARGETTAG and SERIALIZABLEXACT for each lock
1426 : : * table entry. Multiple copies of the same PREDICATELOCKTARGETTAG and
1427 : : * SERIALIZABLEXACT will likely appear.
1428 : : */
1429 : : PredicateLockData *
4815 heikki.linnakangas@i 1430 :CBC 227 : GetPredicateLockStatusData(void)
1431 : : {
1432 : : PredicateLockData *data;
1433 : : int i;
1434 : : int els,
1435 : : el;
1436 : : HASH_SEQ_STATUS seqstat;
1437 : : PREDICATELOCK *predlock;
1438 : :
1439 : 227 : data = (PredicateLockData *) palloc(sizeof(PredicateLockData));
1440 : :
1441 : : /*
1442 : : * To ensure consistency, take simultaneous locks on all partition locks
1443 : : * in ascending order, then SerializableXactHashLock.
1444 : : */
1445 [ + + ]: 3859 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
3730 rhaas@postgresql.org 1446 : 3632 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED);
4815 heikki.linnakangas@i 1447 : 227 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
1448 : :
1449 : : /* Get number of locks and allocate appropriately-sized arrays. */
1450 : 227 : els = hash_get_num_entries(PredicateLockHash);
1451 : 227 : data->nelements = els;
1452 : 227 : data->locktags = (PREDICATELOCKTARGETTAG *)
1453 : 227 : palloc(sizeof(PREDICATELOCKTARGETTAG) * els);
1454 : 227 : data->xacts = (SERIALIZABLEXACT *)
1455 : 227 : palloc(sizeof(SERIALIZABLEXACT) * els);
1456 : :
1457 : :
1458 : : /* Scan through PredicateLockHash and copy contents */
1459 : 227 : hash_seq_init(&seqstat, PredicateLockHash);
1460 : :
1461 : 227 : el = 0;
1462 : :
1463 [ + + ]: 230 : while ((predlock = (PREDICATELOCK *) hash_seq_search(&seqstat)))
1464 : : {
1465 : 3 : data->locktags[el] = predlock->tag.myTarget->tag;
1466 : 3 : data->xacts[el] = *predlock->tag.myXact;
1467 : 3 : el++;
1468 : : }
1469 : :
1470 [ - + ]: 227 : Assert(el == els);
1471 : :
1472 : : /* Release locks in reverse order */
1473 : 227 : LWLockRelease(SerializableXactHashLock);
1474 [ + + ]: 3859 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
3730 rhaas@postgresql.org 1475 : 3632 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
1476 : :
4815 heikki.linnakangas@i 1477 : 227 : return data;
1478 : : }
1479 : :
1480 : : /*
1481 : : * Free up shared memory structures by pushing the oldest sxact (the one at
1482 : : * the front of the SummarizeOldestCommittedSxact queue) into summary form.
1483 : : * Each call will free exactly one SERIALIZABLEXACT structure and may also
1484 : : * free one or more of these structures: SERIALIZABLEXID, PREDICATELOCK,
1485 : : * PREDICATELOCKTARGET, RWConflictData.
1486 : : */
1487 : : static void
4815 heikki.linnakangas@i 1488 :UBC 0 : SummarizeOldestCommittedSxact(void)
1489 : : {
1490 : : SERIALIZABLEXACT *sxact;
1491 : :
1492 : 0 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
1493 : :
1494 : : /*
1495 : : * This function is only called if there are no sxact slots available.
1496 : : * Some of them must belong to old, already-finished transactions, so
1497 : : * there should be something in FinishedSerializableTransactions list that
1498 : : * we can summarize. However, there's a race condition: while we were not
1499 : : * holding any locks, a transaction might have ended and cleaned up all
1500 : : * the finished sxact entries already, freeing up their sxact slots. In
1501 : : * that case, we have nothing to do here. The caller will find one of the
1502 : : * slots released by the other backend when it retries.
1503 : : */
451 andres@anarazel.de 1504 [ # # ]: 0 : if (dlist_is_empty(FinishedSerializableTransactions))
1505 : : {
4815 heikki.linnakangas@i 1506 : 0 : LWLockRelease(SerializableFinishedListLock);
1507 : 0 : return;
1508 : : }
1509 : :
1510 : : /*
1511 : : * Grab the first sxact off the finished list -- this will be the earliest
1512 : : * commit. Remove it from the list.
1513 : : */
451 andres@anarazel.de 1514 : 0 : sxact = dlist_head_element(SERIALIZABLEXACT, finishedLink,
1515 : : FinishedSerializableTransactions);
1516 : 0 : dlist_delete_thoroughly(&sxact->finishedLink);
1517 : :
1518 : : /* Add to SLRU summary information. */
4815 heikki.linnakangas@i 1519 [ # # # # ]: 0 : if (TransactionIdIsValid(sxact->topXid) && !SxactIsReadOnly(sxact))
1430 tgl@sss.pgh.pa.us 1520 [ # # ]: 0 : SerialAdd(sxact->topXid, SxactHasConflictOut(sxact)
1521 : : ? sxact->SeqNo.earliestOutConflictCommit : InvalidSerCommitSeqNo);
1522 : :
1523 : : /* Summarize and release the detail. */
4815 heikki.linnakangas@i 1524 : 0 : ReleaseOneSerializableXact(sxact, false, true);
1525 : :
1526 : 0 : LWLockRelease(SerializableFinishedListLock);
1527 : : }
1528 : :
1529 : : /*
1530 : : * GetSafeSnapshot
1531 : : * Obtain and register a snapshot for a READ ONLY DEFERRABLE
1532 : : * transaction. Ensures that the snapshot is "safe", i.e. a
1533 : : * read-only transaction running on it can execute serializably
1534 : : * without further checks. This requires waiting for concurrent
1535 : : * transactions to complete, and retrying with a new snapshot if
1536 : : * one of them could possibly create a conflict.
1537 : : *
1538 : : * As with GetSerializableTransactionSnapshot (which this is a subroutine
1539 : : * for), the passed-in Snapshot pointer should reference a static data
1540 : : * area that can safely be passed to GetSnapshotData.
1541 : : */
1542 : : static Snapshot
4815 heikki.linnakangas@i 1543 :CBC 4 : GetSafeSnapshot(Snapshot origSnapshot)
1544 : : {
1545 : : Snapshot snapshot;
1546 : :
1547 [ + - + - ]: 4 : Assert(XactReadOnly && XactDeferrable);
1548 : :
1549 : : while (true)
1550 : : {
1551 : : /*
1552 : : * GetSerializableTransactionSnapshotInt is going to call
1553 : : * GetSnapshotData, so we need to provide it the static snapshot area
1554 : : * our caller passed to us. The pointer returned is actually the same
1555 : : * one passed to it, but we avoid assuming that here.
1556 : : */
4558 tgl@sss.pgh.pa.us 1557 : 5 : snapshot = GetSerializableTransactionSnapshotInt(origSnapshot,
1558 : : NULL, InvalidPid);
1559 : :
4815 heikki.linnakangas@i 1560 [ + + ]: 5 : if (MySerializableXact == InvalidSerializableXact)
1561 : 4 : return snapshot; /* no concurrent r/w xacts; it's safe */
1562 : :
4692 1563 : 1 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1564 : :
1565 : : /*
1566 : : * Wait for concurrent transactions to finish. Stop early if one of
1567 : : * them marked us as conflicted.
1568 : : */
1569 : 1 : MySerializableXact->flags |= SXACT_FLAG_DEFERRABLE_WAITING;
451 andres@anarazel.de 1570 [ + + ]: 2 : while (!(dlist_is_empty(&MySerializableXact->possibleUnsafeConflicts) ||
4815 heikki.linnakangas@i 1571 [ + - ]: 1 : SxactIsROUnsafe(MySerializableXact)))
1572 : : {
4692 1573 : 1 : LWLockRelease(SerializableXactHashLock);
2749 rhaas@postgresql.org 1574 : 1 : ProcWaitForSignal(WAIT_EVENT_SAFE_SNAPSHOT);
4692 heikki.linnakangas@i 1575 : 1 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1576 : : }
4815 1577 : 1 : MySerializableXact->flags &= ~SXACT_FLAG_DEFERRABLE_WAITING;
1578 : :
1579 [ - + ]: 1 : if (!SxactIsROUnsafe(MySerializableXact))
1580 : : {
4692 heikki.linnakangas@i 1581 :LBC (2) : LWLockRelease(SerializableXactHashLock);
4815 1582 : (2) : break; /* success */
1583 : : }
1584 : :
4692 heikki.linnakangas@i 1585 :CBC 1 : LWLockRelease(SerializableXactHashLock);
1586 : :
1587 : : /* else, need to retry... */
4815 1588 [ - + ]: 1 : ereport(DEBUG2,
1589 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1590 : : errmsg_internal("deferrable snapshot was unsafe; trying a new one")));
1857 tmunro@postgresql.or 1591 : 1 : ReleasePredicateLocks(false, false);
1592 : : }
1593 : :
1594 : : /*
1595 : : * Now we have a safe snapshot, so we don't need to do any further checks.
1596 : : */
4815 heikki.linnakangas@i 1597 [ # # ]:LBC (2) : Assert(SxactIsROSafe(MySerializableXact));
1857 tmunro@postgresql.or 1598 : (2) : ReleasePredicateLocks(false, true);
1599 : :
4815 heikki.linnakangas@i 1600 : (2) : return snapshot;
1601 : : }
1602 : :
1603 : : /*
1604 : : * GetSafeSnapshotBlockingPids
1605 : : * If the specified process is currently blocked in GetSafeSnapshot,
1606 : : * write the process IDs of all processes that it is blocked by
1607 : : * into the caller-supplied buffer output[]. The list is truncated at
1608 : : * output_size, and the number of PIDs written into the buffer is
1609 : : * returned. Returns zero if the given PID is not currently blocked
1610 : : * in GetSafeSnapshot.
1611 : : */
1612 : : int
2561 tgl@sss.pgh.pa.us 1613 :CBC 796 : GetSafeSnapshotBlockingPids(int blocked_pid, int *output, int output_size)
1614 : : {
1615 : 796 : int num_written = 0;
1616 : : dlist_iter iter;
431 andres@anarazel.de 1617 : 796 : SERIALIZABLEXACT *blocking_sxact = NULL;
1618 : :
2561 tgl@sss.pgh.pa.us 1619 : 796 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
1620 : :
1621 : : /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
451 andres@anarazel.de 1622 [ + - + + ]: 1674 : dlist_foreach(iter, &PredXact->activeList)
1623 : : {
431 1624 : 954 : SERIALIZABLEXACT *sxact =
1625 : 954 : dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
1626 : :
2561 tgl@sss.pgh.pa.us 1627 [ + + ]: 954 : if (sxact->pid == blocked_pid)
1628 : : {
431 andres@anarazel.de 1629 : 76 : blocking_sxact = sxact;
2561 tgl@sss.pgh.pa.us 1630 : 76 : break;
1631 : : }
1632 : : }
1633 : :
1634 : : /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
431 andres@anarazel.de 1635 [ + + + + ]: 796 : if (blocking_sxact != NULL && SxactIsDeferrableWaiting(blocking_sxact))
1636 : : {
1637 : : /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1638 [ + - + - ]: 2 : dlist_foreach(iter, &blocking_sxact->possibleUnsafeConflicts)
1639 : : {
451 1640 : 2 : RWConflict possibleUnsafeConflict =
331 tgl@sss.pgh.pa.us 1641 : 2 : dlist_container(RWConflictData, inLink, iter.cur);
1642 : :
2561 1643 : 2 : output[num_written++] = possibleUnsafeConflict->sxactOut->pid;
1644 : :
431 andres@anarazel.de 1645 [ + - ]: 2 : if (num_written >= output_size)
1646 : 2 : break;
1647 : : }
1648 : : }
1649 : :
2561 tgl@sss.pgh.pa.us 1650 : 796 : LWLockRelease(SerializableXactHashLock);
1651 : :
1652 : 796 : return num_written;
1653 : : }
1654 : :
1655 : : /*
1656 : : * Acquire a snapshot that can be used for the current transaction.
1657 : : *
1658 : : * Make sure we have a SERIALIZABLEXACT reference in MySerializableXact.
1659 : : * It should be current for this process and be contained in PredXact.
1660 : : *
1661 : : * The passed-in Snapshot pointer should reference a static data area that
1662 : : * can safely be passed to GetSnapshotData. The return value is actually
1663 : : * always this same pointer; no new snapshot data structure is allocated
1664 : : * within this function.
1665 : : */
1666 : : Snapshot
4584 1667 : 1652 : GetSerializableTransactionSnapshot(Snapshot snapshot)
1668 : : {
4815 heikki.linnakangas@i 1669 [ - + ]: 1652 : Assert(IsolationIsSerializable());
1670 : :
1671 : : /*
1672 : : * Can't use serializable mode while recovery is still active, as it is,
1673 : : * for example, on a hot standby. We could get here despite the check in
1674 : : * check_transaction_isolation() if default_transaction_isolation is set
1675 : : * to serializable, so phrase the hint accordingly.
1676 : : */
4251 tgl@sss.pgh.pa.us 1677 [ - + ]: 1652 : if (RecoveryInProgress())
4251 tgl@sss.pgh.pa.us 1678 [ # # ]:UBC 0 : ereport(ERROR,
1679 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1680 : : errmsg("cannot use serializable mode in a hot standby"),
1681 : : errdetail("default_transaction_isolation is set to \"serializable\"."),
1682 : : errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1683 : :
1684 : : /*
1685 : : * A special optimization is available for SERIALIZABLE READ ONLY
1686 : : * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1687 : : * thereby avoid all SSI overhead once it's running.
1688 : : */
4815 heikki.linnakangas@i 1689 [ + + + + ]:CBC 1652 : if (XactReadOnly && XactDeferrable)
1690 : 4 : return GetSafeSnapshot(snapshot);
1691 : :
4558 tgl@sss.pgh.pa.us 1692 : 1648 : return GetSerializableTransactionSnapshotInt(snapshot,
1693 : : NULL, InvalidPid);
1694 : : }
1695 : :
1696 : : /*
1697 : : * Import a snapshot to be used for the current transaction.
1698 : : *
1699 : : * This is nearly the same as GetSerializableTransactionSnapshot, except that
1700 : : * we don't take a new snapshot, but rather use the data we're handed.
1701 : : *
1702 : : * The caller must have verified that the snapshot came from a serializable
1703 : : * transaction; and if we're read-write, the source transaction must not be
1704 : : * read-only.
1705 : : */
1706 : : void
1707 : 13 : SetSerializableTransactionSnapshot(Snapshot snapshot,
1708 : : VirtualTransactionId *sourcevxid,
1709 : : int sourcepid)
1710 : : {
1711 [ - + ]: 13 : Assert(IsolationIsSerializable());
1712 : :
1713 : : /*
1714 : : * If this is called by parallel.c in a parallel worker, we don't want to
1715 : : * create a SERIALIZABLEXACT just yet because the leader's
1716 : : * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1717 : : * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1718 : : * case, because the leader has already determined that the snapshot it
1719 : : * has passed us is safe. So there is nothing for us to do.
1720 : : */
1857 tmunro@postgresql.or 1721 [ + - ]: 13 : if (IsParallelWorker())
1722 : 13 : return;
1723 : :
1724 : : /*
1725 : : * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1726 : : * import snapshots, since there's no way to wait for a safe snapshot when
1727 : : * we're using the snap we're told to. (XXX instead of throwing an error,
1728 : : * we could just ignore the XactDeferrable flag?)
1729 : : */
4558 tgl@sss.pgh.pa.us 1730 [ # # # # ]:UBC 0 : if (XactReadOnly && XactDeferrable)
1731 [ # # ]: 0 : ereport(ERROR,
1732 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1733 : : errmsg("a snapshot-importing transaction must not be READ ONLY DEFERRABLE")));
1734 : :
2496 andres@anarazel.de 1735 : 0 : (void) GetSerializableTransactionSnapshotInt(snapshot, sourcevxid,
1736 : : sourcepid);
1737 : : }
1738 : :
1739 : : /*
1740 : : * Guts of GetSerializableTransactionSnapshot
1741 : : *
1742 : : * If sourcevxid is valid, this is actually an import operation and we should
1743 : : * skip calling GetSnapshotData, because the snapshot contents are already
1744 : : * loaded up. HOWEVER: to avoid race conditions, we must check that the
1745 : : * source xact is still running after we acquire SerializableXactHashLock.
1746 : : * We do that by calling ProcArrayInstallImportedXmin.
1747 : : */
1748 : : static Snapshot
4558 tgl@sss.pgh.pa.us 1749 :CBC 1653 : GetSerializableTransactionSnapshotInt(Snapshot snapshot,
1750 : : VirtualTransactionId *sourcevxid,
1751 : : int sourcepid)
1752 : : {
1753 : : PGPROC *proc;
1754 : : VirtualTransactionId vxid;
1755 : : SERIALIZABLEXACT *sxact,
1756 : : *othersxact;
1757 : :
1758 : : /* We only do this for serializable transactions. Once. */
4815 heikki.linnakangas@i 1759 [ - + ]: 1653 : Assert(MySerializableXact == InvalidSerializableXact);
1760 : :
1761 [ - + ]: 1653 : Assert(!RecoveryInProgress());
1762 : :
1763 : : /*
1764 : : * Since all parts of a serializable transaction must use the same
1765 : : * snapshot, it is too late to establish one after a parallel operation
1766 : : * has begun.
1767 : : */
3272 rhaas@postgresql.org 1768 [ - + ]: 1653 : if (IsInParallelMode())
3272 rhaas@postgresql.org 1769 [ # # ]:UBC 0 : elog(ERROR, "cannot establish serializable snapshot during a parallel operation");
1770 : :
4815 heikki.linnakangas@i 1771 :CBC 1653 : proc = MyProc;
1772 [ - + ]: 1653 : Assert(proc != NULL);
1773 : 1653 : GET_VXID_FROM_PGPROC(vxid, *proc);
1774 : :
1775 : : /*
1776 : : * First we get the sxact structure, which may involve looping and access
1777 : : * to the "finished" list to free a structure for use.
1778 : : *
1779 : : * We must hold SerializableXactHashLock when taking/checking the snapshot
1780 : : * to avoid race conditions, for much the same reasons that
1781 : : * GetSnapshotData takes the ProcArrayLock. Since we might have to
1782 : : * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1783 : : * this means we have to create the sxact first, which is a bit annoying
1784 : : * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1785 : : * the sxact). Consider refactoring to avoid this.
1786 : : */
1787 : : #ifdef TEST_SUMMARIZE_SERIAL
1788 : : SummarizeOldestCommittedSxact();
1789 : : #endif
1790 : 1653 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1791 : : do
1792 : : {
1793 : 1653 : sxact = CreatePredXact();
1794 : : /* If null, push out committed sxact to SLRU summary & retry. */
1795 [ - + ]: 1653 : if (!sxact)
1796 : : {
4815 heikki.linnakangas@i 1797 :UBC 0 : LWLockRelease(SerializableXactHashLock);
1798 : 0 : SummarizeOldestCommittedSxact();
1799 : 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1800 : : }
4815 heikki.linnakangas@i 1801 [ - + ]:CBC 1653 : } while (!sxact);
1802 : :
1803 : : /* Get the snapshot, or check that it's safe to use */
2496 andres@anarazel.de 1804 [ + - ]: 1653 : if (!sourcevxid)
4558 tgl@sss.pgh.pa.us 1805 : 1653 : snapshot = GetSnapshotData(snapshot);
2496 andres@anarazel.de 1806 [ # # ]:UBC 0 : else if (!ProcArrayInstallImportedXmin(snapshot->xmin, sourcevxid))
1807 : : {
4558 tgl@sss.pgh.pa.us 1808 : 0 : ReleasePredXact(sxact);
1809 : 0 : LWLockRelease(SerializableXactHashLock);
1810 [ # # ]: 0 : ereport(ERROR,
1811 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1812 : : errmsg("could not import the requested snapshot"),
1813 : : errdetail("The source process with PID %d is not running anymore.",
1814 : : sourcepid)));
1815 : : }
1816 : :
1817 : : /*
1818 : : * If there are no serializable transactions which are not read-only, we
1819 : : * can "opt out" of predicate locking and conflict checking for a
1820 : : * read-only transaction.
1821 : : *
1822 : : * The reason this is safe is that a read-only transaction can only become
1823 : : * part of a dangerous structure if it overlaps a writable transaction
1824 : : * which in turn overlaps a writable transaction which committed before
1825 : : * the read-only transaction started. A new writable transaction can
1826 : : * overlap this one, but it can't meet the other condition of overlapping
1827 : : * a transaction which committed before this one started.
1828 : : */
4815 heikki.linnakangas@i 1829 [ + + + + ]:CBC 1653 : if (XactReadOnly && PredXact->WritableSxactCount == 0)
1830 : : {
1831 : 112 : ReleasePredXact(sxact);
1832 : 112 : LWLockRelease(SerializableXactHashLock);
1833 : 112 : return snapshot;
1834 : : }
1835 : :
1836 : : /* Initialize the structure. */
1837 : 1541 : sxact->vxid = vxid;
1838 : 1541 : sxact->SeqNo.lastCommitBeforeSnapshot = PredXact->LastSxactCommitSeqNo;
4665 1839 : 1541 : sxact->prepareSeqNo = InvalidSerCommitSeqNo;
4815 1840 : 1541 : sxact->commitSeqNo = InvalidSerCommitSeqNo;
451 andres@anarazel.de 1841 : 1541 : dlist_init(&(sxact->outConflicts));
1842 : 1541 : dlist_init(&(sxact->inConflicts));
1843 : 1541 : dlist_init(&(sxact->possibleUnsafeConflicts));
4815 heikki.linnakangas@i 1844 : 1541 : sxact->topXid = GetTopTransactionIdIfAny();
1845 : 1541 : sxact->finishedBefore = InvalidTransactionId;
1846 : 1541 : sxact->xmin = snapshot->xmin;
1847 : 1541 : sxact->pid = MyProcPid;
52 heikki.linnakangas@i 1848 :GNC 1541 : sxact->pgprocno = MyProcNumber;
451 andres@anarazel.de 1849 :CBC 1541 : dlist_init(&sxact->predicateLocks);
1850 : 1541 : dlist_node_init(&sxact->finishedLink);
4815 heikki.linnakangas@i 1851 : 1541 : sxact->flags = 0;
1852 [ + + ]: 1541 : if (XactReadOnly)
1853 : : {
1854 : : dlist_iter iter;
1855 : :
1856 : 106 : sxact->flags |= SXACT_FLAG_READ_ONLY;
1857 : :
1858 : : /*
1859 : : * Register all concurrent r/w transactions as possible conflicts; if
1860 : : * all of them commit without any outgoing conflicts to earlier
1861 : : * transactions then this snapshot can be deemed safe (and we can run
1862 : : * without tracking predicate locks).
1863 : : */
451 andres@anarazel.de 1864 [ + - + + ]: 464 : dlist_foreach(iter, &PredXact->activeList)
1865 : : {
1866 : 358 : othersxact = dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
1867 : :
4664 heikki.linnakangas@i 1868 [ + + ]: 358 : if (!SxactIsCommitted(othersxact)
1869 [ + - ]: 239 : && !SxactIsDoomed(othersxact)
1870 [ + + ]: 239 : && !SxactIsReadOnly(othersxact))
1871 : : {
4815 1872 : 132 : SetPossibleUnsafeConflict(sxact, othersxact);
1873 : : }
1874 : : }
1875 : :
1876 : : /*
1877 : : * If we didn't find any possibly unsafe conflicts because every
1878 : : * uncommitted writable transaction turned out to be doomed, then we
1879 : : * can "opt out" immediately. See comments above the earlier check
1880 : : * for PredXact->WritableSxactCount == 0.
1881 : : */
402 tmunro@postgresql.or 1882 [ - + ]: 106 : if (dlist_is_empty(&sxact->possibleUnsafeConflicts))
1883 : : {
402 tmunro@postgresql.or 1884 :UBC 0 : ReleasePredXact(sxact);
1885 : 0 : LWLockRelease(SerializableXactHashLock);
1886 : 0 : return snapshot;
1887 : : }
1888 : : }
1889 : : else
1890 : : {
4815 heikki.linnakangas@i 1891 :CBC 1435 : ++(PredXact->WritableSxactCount);
1892 [ - + ]: 1435 : Assert(PredXact->WritableSxactCount <=
1893 : : (MaxBackends + max_prepared_xacts));
1894 : : }
1895 : :
1896 : : /* Maintain serializable global xmin info. */
402 tmunro@postgresql.or 1897 [ + + ]: 1541 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
1898 : : {
1899 [ - + ]: 836 : Assert(PredXact->SxactGlobalXminCount == 0);
1900 : 836 : PredXact->SxactGlobalXmin = snapshot->xmin;
1901 : 836 : PredXact->SxactGlobalXminCount = 1;
1902 : 836 : SerialSetActiveSerXmin(snapshot->xmin);
1903 : : }
1904 [ + + ]: 705 : else if (TransactionIdEquals(snapshot->xmin, PredXact->SxactGlobalXmin))
1905 : : {
1906 [ - + ]: 669 : Assert(PredXact->SxactGlobalXminCount > 0);
1907 : 669 : PredXact->SxactGlobalXminCount++;
1908 : : }
1909 : : else
1910 : : {
1911 [ - + ]: 36 : Assert(TransactionIdFollows(snapshot->xmin, PredXact->SxactGlobalXmin));
1912 : : }
1913 : :
4815 heikki.linnakangas@i 1914 : 1541 : MySerializableXact = sxact;
4692 1915 : 1541 : MyXactDidWrite = false; /* haven't written anything yet */
1916 : :
4815 1917 : 1541 : LWLockRelease(SerializableXactHashLock);
1918 : :
1857 tmunro@postgresql.or 1919 : 1541 : CreateLocalPredicateLockHash();
1920 : :
1921 : 1541 : return snapshot;
1922 : : }
1923 : :
1924 : : static void
1925 : 1554 : CreateLocalPredicateLockHash(void)
1926 : : {
1927 : : HASHCTL hash_ctl;
1928 : :
1929 : : /* Initialize the backend-local hash table of parent locks */
4815 heikki.linnakangas@i 1930 [ - + ]: 1554 : Assert(LocalPredicateLockHash == NULL);
1931 : 1554 : hash_ctl.keysize = sizeof(PREDICATELOCKTARGETTAG);
1932 : 1554 : hash_ctl.entrysize = sizeof(LOCALPREDICATELOCK);
1933 : 1554 : LocalPredicateLockHash = hash_create("Local predicate lock",
1934 : : max_predicate_locks_per_xact,
1935 : : &hash_ctl,
1936 : : HASH_ELEM | HASH_BLOBS);
1937 : 1554 : }
1938 : :
1939 : : /*
1940 : : * Register the top level XID in SerializableXidHash.
1941 : : * Also store it for easy reference in MySerializableXact.
1942 : : */
1943 : : void
4680 1944 : 114647 : RegisterPredicateLockingXid(TransactionId xid)
1945 : : {
1946 : : SERIALIZABLEXIDTAG sxidtag;
1947 : : SERIALIZABLEXID *sxid;
1948 : : bool found;
1949 : :
1950 : : /*
1951 : : * If we're not tracking predicate lock data for this transaction, we
1952 : : * should ignore the request and return quickly.
1953 : : */
4815 1954 [ + + ]: 114647 : if (MySerializableXact == InvalidSerializableXact)
1955 : 113372 : return;
1956 : :
1957 : : /* We should have a valid XID and be at the top level. */
1958 [ - + ]: 1275 : Assert(TransactionIdIsValid(xid));
1959 : :
4692 1960 : 1275 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1961 : :
1962 : : /* This should only be done once per transaction. */
1963 [ - + ]: 1275 : Assert(MySerializableXact->topXid == InvalidTransactionId);
1964 : :
4815 1965 : 1275 : MySerializableXact->topXid = xid;
1966 : :
1967 : 1275 : sxidtag.xid = xid;
1968 : 1275 : sxid = (SERIALIZABLEXID *) hash_search(SerializableXidHash,
1969 : : &sxidtag,
1970 : : HASH_ENTER, &found);
1971 [ - + ]: 1275 : Assert(!found);
1972 : :
1973 : : /* Initialize the structure. */
4692 1974 : 1275 : sxid->myXact = MySerializableXact;
4815 1975 : 1275 : LWLockRelease(SerializableXactHashLock);
1976 : : }
1977 : :
1978 : :
1979 : : /*
1980 : : * Check whether there are any predicate locks held by any transaction
1981 : : * for the page at the given block number.
1982 : : *
1983 : : * Note that the transaction may be completed but not yet subject to
1984 : : * cleanup due to overlapping serializable transactions. This must
1985 : : * return valid information regardless of transaction isolation level.
1986 : : *
1987 : : * Also note that this doesn't check for a conflicting relation lock,
1988 : : * just a lock specifically on the given page.
1989 : : *
1990 : : * One use is to support proper behavior during GiST index vacuum.
1991 : : */
1992 : : bool
4680 heikki.linnakangas@i 1993 :UBC 0 : PageIsPredicateLocked(Relation relation, BlockNumber blkno)
1994 : : {
1995 : : PREDICATELOCKTARGETTAG targettag;
1996 : : uint32 targettaghash;
1997 : : LWLock *partitionLock;
1998 : : PREDICATELOCKTARGET *target;
1999 : :
4815 2000 : 0 : SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
2001 : : relation->rd_locator.dbOid,
2002 : : relation->rd_id,
2003 : : blkno);
2004 : :
2005 : 0 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
2006 : 0 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
2007 : 0 : LWLockAcquire(partitionLock, LW_SHARED);
2008 : : target = (PREDICATELOCKTARGET *)
2009 : 0 : hash_search_with_hash_value(PredicateLockTargetHash,
2010 : : &targettag, targettaghash,
2011 : : HASH_FIND, NULL);
2012 : 0 : LWLockRelease(partitionLock);
2013 : :
2014 : 0 : return (target != NULL);
2015 : : }
2016 : :
2017 : :
2018 : : /*
2019 : : * Check whether a particular lock is held by this transaction.
2020 : : *
2021 : : * Important note: this function may return false even if the lock is
2022 : : * being held, because it uses the local lock table which is not
2023 : : * updated if another transaction modifies our lock list (e.g. to
2024 : : * split an index page). It can also return true when a coarser
2025 : : * granularity lock that covers this target is being held. Be careful
2026 : : * to only use this function in circumstances where such errors are
2027 : : * acceptable!
2028 : : */
2029 : : static bool
4680 tgl@sss.pgh.pa.us 2030 :CBC 77199 : PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag)
2031 : : {
2032 : : LOCALPREDICATELOCK *lock;
2033 : :
2034 : : /* check local hash table */
4815 heikki.linnakangas@i 2035 : 77199 : lock = (LOCALPREDICATELOCK *) hash_search(LocalPredicateLockHash,
2036 : : targettag,
2037 : : HASH_FIND, NULL);
2038 : :
2039 [ + + ]: 77199 : if (!lock)
2040 : 30098 : return false;
2041 : :
2042 : : /*
2043 : : * Found entry in the table, but still need to check whether it's actually
2044 : : * held -- it could just be a parent of some held lock.
2045 : : */
2046 : 47101 : return lock->held;
2047 : : }
2048 : :
2049 : : /*
2050 : : * Return the parent lock tag in the lock hierarchy: the next coarser
2051 : : * lock that covers the provided tag.
2052 : : *
2053 : : * Returns true and sets *parent to the parent tag if one exists,
2054 : : * returns false if none exists.
2055 : : */
2056 : : static bool
4680 tgl@sss.pgh.pa.us 2057 : 45167 : GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag,
2058 : : PREDICATELOCKTARGETTAG *parent)
2059 : : {
4815 heikki.linnakangas@i 2060 [ + + + + : 45167 : switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag))
+ - ]
2061 : : {
2062 : 9818 : case PREDLOCKTAG_RELATION:
2063 : : /* relation locks have no parent lock */
2064 : 9818 : return false;
2065 : :
2066 : 8416 : case PREDLOCKTAG_PAGE:
2067 : : /* parent lock is relation lock */
2068 : 8416 : SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2069 : : GET_PREDICATELOCKTARGETTAG_DB(*tag),
2070 : : GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2071 : :
2072 : 8416 : return true;
2073 : :
2074 : 26933 : case PREDLOCKTAG_TUPLE:
2075 : : /* parent lock is page lock */
2076 : 26933 : SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2077 : : GET_PREDICATELOCKTARGETTAG_DB(*tag),
2078 : : GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2079 : : GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2080 : 26933 : return true;
2081 : : }
2082 : :
2083 : : /* not reachable */
4815 heikki.linnakangas@i 2084 :UBC 0 : Assert(false);
2085 : : return false;
2086 : : }
2087 : :
2088 : : /*
2089 : : * Check whether the lock we are considering is already covered by a
2090 : : * coarser lock for our transaction.
2091 : : *
2092 : : * Like PredicateLockExists, this function might return a false
2093 : : * negative, but it will never return a false positive.
2094 : : */
2095 : : static bool
4680 tgl@sss.pgh.pa.us 2096 :CBC 26040 : CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag)
2097 : : {
2098 : : PREDICATELOCKTARGETTAG targettag,
2099 : : parenttag;
2100 : :
4815 heikki.linnakangas@i 2101 : 26040 : targettag = *newtargettag;
2102 : :
2103 : : /* check parents iteratively until no more */
2104 [ + + ]: 31424 : while (GetParentPredicateLockTag(&targettag, &parenttag))
2105 : : {
2106 : 27205 : targettag = parenttag;
2107 [ + + ]: 27205 : if (PredicateLockExists(&targettag))
2108 : 21821 : return true;
2109 : : }
2110 : :
2111 : : /* no more parents to check; lock is not covered */
2112 : 4219 : return false;
2113 : : }
2114 : :
2115 : : /*
2116 : : * Remove the dummy entry from the predicate lock target hash, to free up some
2117 : : * scratch space. The caller must be holding SerializablePredicateListLock,
2118 : : * and must restore the entry with RestoreScratchTarget() before releasing the
2119 : : * lock.
2120 : : *
2121 : : * If lockheld is true, the caller is already holding the partition lock
2122 : : * of the partition containing the scratch entry.
2123 : : */
2124 : : static void
4694 2125 : 49 : RemoveScratchTarget(bool lockheld)
2126 : : {
2127 : : bool found;
2128 : :
1430 tgl@sss.pgh.pa.us 2129 [ - + ]: 49 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2130 : :
4694 heikki.linnakangas@i 2131 [ - + ]: 49 : if (!lockheld)
4694 heikki.linnakangas@i 2132 :UBC 0 : LWLockAcquire(ScratchPartitionLock, LW_EXCLUSIVE);
4694 heikki.linnakangas@i 2133 :CBC 49 : hash_search_with_hash_value(PredicateLockTargetHash,
2134 : : &ScratchTargetTag,
2135 : : ScratchTargetTagHash,
2136 : : HASH_REMOVE, &found);
2137 [ - + ]: 49 : Assert(found);
2138 [ - + ]: 49 : if (!lockheld)
4694 heikki.linnakangas@i 2139 :UBC 0 : LWLockRelease(ScratchPartitionLock);
4694 heikki.linnakangas@i 2140 :CBC 49 : }
2141 : :
2142 : : /*
2143 : : * Re-insert the dummy entry in predicate lock target hash.
2144 : : */
2145 : : static void
2146 : 49 : RestoreScratchTarget(bool lockheld)
2147 : : {
2148 : : bool found;
2149 : :
1430 tgl@sss.pgh.pa.us 2150 [ - + ]: 49 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2151 : :
4694 heikki.linnakangas@i 2152 [ - + ]: 49 : if (!lockheld)
4694 heikki.linnakangas@i 2153 :UBC 0 : LWLockAcquire(ScratchPartitionLock, LW_EXCLUSIVE);
4694 heikki.linnakangas@i 2154 :CBC 49 : hash_search_with_hash_value(PredicateLockTargetHash,
2155 : : &ScratchTargetTag,
2156 : : ScratchTargetTagHash,
2157 : : HASH_ENTER, &found);
2158 [ - + ]: 49 : Assert(!found);
2159 [ - + ]: 49 : if (!lockheld)
4694 heikki.linnakangas@i 2160 :UBC 0 : LWLockRelease(ScratchPartitionLock);
4694 heikki.linnakangas@i 2161 :CBC 49 : }
2162 : :
2163 : : /*
2164 : : * Check whether the list of related predicate locks is empty for a
2165 : : * predicate lock target, and remove the target if it is.
2166 : : */
2167 : : static void
4815 2168 : 4213 : RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target, uint32 targettaghash)
2169 : : {
2170 : : PREDICATELOCKTARGET *rmtarget PG_USED_FOR_ASSERTS_ONLY;
2171 : :
1430 tgl@sss.pgh.pa.us 2172 [ - + ]: 4213 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2173 : :
2174 : : /* Can't remove it until no locks at this target. */
451 andres@anarazel.de 2175 [ + + ]: 4213 : if (!dlist_is_empty(&target->predicateLocks))
4815 heikki.linnakangas@i 2176 : 961 : return;
2177 : :
2178 : : /* Actually remove the target. */
2179 : 3252 : rmtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2180 : 3252 : &target->tag,
2181 : : targettaghash,
2182 : : HASH_REMOVE, NULL);
2183 [ - + ]: 3252 : Assert(rmtarget == target);
2184 : : }
2185 : :
2186 : : /*
2187 : : * Delete child target locks owned by this process.
2188 : : * This implementation is assuming that the usage of each target tag field
2189 : : * is uniform. No need to make this hard if we don't have to.
2190 : : *
2191 : : * We acquire an LWLock in the case of parallel mode, because worker
2192 : : * backends have access to the leader's SERIALIZABLEXACT. Otherwise,
2193 : : * we aren't acquiring LWLocks for the predicate lock or lock
2194 : : * target structures associated with this transaction unless we're going
2195 : : * to modify them, because no other process is permitted to modify our
2196 : : * locks.
2197 : : */
2198 : : static void
4680 tgl@sss.pgh.pa.us 2199 : 2353 : DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
2200 : : {
2201 : : SERIALIZABLEXACT *sxact;
2202 : : PREDICATELOCK *predlock;
2203 : : dlist_mutable_iter iter;
2204 : :
1430 2205 : 2353 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
4692 heikki.linnakangas@i 2206 : 2353 : sxact = MySerializableXact;
1857 tmunro@postgresql.or 2207 [ + + ]: 2353 : if (IsInParallelMode())
1430 tgl@sss.pgh.pa.us 2208 : 11 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
2209 : :
451 andres@anarazel.de 2210 [ + - + + ]: 7787 : dlist_foreach_modify(iter, &sxact->predicateLocks)
2211 : : {
2212 : : PREDICATELOCKTAG oldlocktag;
2213 : : PREDICATELOCKTARGET *oldtarget;
2214 : : PREDICATELOCKTARGETTAG oldtargettag;
2215 : :
2216 : 5434 : predlock = dlist_container(PREDICATELOCK, xactLink, iter.cur);
2217 : :
4815 heikki.linnakangas@i 2218 : 5434 : oldlocktag = predlock->tag;
2219 [ - + ]: 5434 : Assert(oldlocktag.myXact == sxact);
2220 : 5434 : oldtarget = oldlocktag.myTarget;
2221 : 5434 : oldtargettag = oldtarget->tag;
2222 : :
2223 [ + + + - : 5434 : if (TargetTagIsCoveredBy(oldtargettag, *newtargettag))
+ + + + +
+ - + +
- ]
2224 : : {
2225 : : uint32 oldtargettaghash;
2226 : : LWLock *partitionLock;
2227 : : PREDICATELOCK *rmpredlock PG_USED_FOR_ASSERTS_ONLY;
2228 : :
2229 : 999 : oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
2230 : 999 : partitionLock = PredicateLockHashPartitionLock(oldtargettaghash);
2231 : :
2232 : 999 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2233 : :
451 andres@anarazel.de 2234 : 999 : dlist_delete(&predlock->xactLink);
2235 : 999 : dlist_delete(&predlock->targetLink);
4815 heikki.linnakangas@i 2236 : 999 : rmpredlock = hash_search_with_hash_value
2237 : : (PredicateLockHash,
2238 : : &oldlocktag,
2239 : 999 : PredicateLockHashCodeFromTargetHashCode(&oldlocktag,
2240 : : oldtargettaghash),
2241 : : HASH_REMOVE, NULL);
2242 [ - + ]: 999 : Assert(rmpredlock == predlock);
2243 : :
2244 : 999 : RemoveTargetIfNoLongerUsed(oldtarget, oldtargettaghash);
2245 : :
2246 : 999 : LWLockRelease(partitionLock);
2247 : :
2248 : 999 : DecrementParentLocks(&oldtargettag);
2249 : : }
2250 : : }
1857 tmunro@postgresql.or 2251 [ + + ]: 2353 : if (IsInParallelMode())
1430 tgl@sss.pgh.pa.us 2252 : 11 : LWLockRelease(&sxact->perXactPredicateListLock);
2253 : 2353 : LWLockRelease(SerializablePredicateListLock);
4815 heikki.linnakangas@i 2254 : 2353 : }
2255 : :
2256 : : /*
2257 : : * Returns the promotion limit for a given predicate lock target. This is the
2258 : : * max number of descendant locks allowed before promoting to the specified
2259 : : * tag. Note that the limit includes non-direct descendants (e.g., both tuples
2260 : : * and pages for a relation lock).
2261 : : *
2262 : : * Currently the default limit is 2 for a page lock, and half of the value of
2263 : : * max_pred_locks_per_transaction - 1 for a relation lock, to match behavior
2264 : : * of earlier releases when upgrading.
2265 : : *
2266 : : * TODO SSI: We should probably add additional GUCs to allow a maximum ratio
2267 : : * of page and tuple locks based on the pages in a relation, and the maximum
2268 : : * ratio of tuple locks to tuples in a page. This would provide more
2269 : : * generally "balanced" allocation of locks to where they are most useful,
2270 : : * while still allowing the absolute numbers to prevent one relation from
2271 : : * tying up all predicate lock resources.
2272 : : */
2273 : : static int
2564 kgrittn@postgresql.o 2274 : 5384 : MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag)
2275 : : {
4815 heikki.linnakangas@i 2276 [ + - + + : 5384 : switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag))
- - ]
2277 : : {
2278 : 3518 : case PREDLOCKTAG_RELATION:
2564 kgrittn@postgresql.o 2279 : 3518 : return max_predicate_locks_per_relation < 0
2280 : : ? (max_predicate_locks_per_xact
2281 : 3518 : / (-max_predicate_locks_per_relation)) - 1
2282 [ + - ]: 3518 : : max_predicate_locks_per_relation;
2283 : :
4815 heikki.linnakangas@i 2284 : 1866 : case PREDLOCKTAG_PAGE:
2564 kgrittn@postgresql.o 2285 : 1866 : return max_predicate_locks_per_page;
2286 : :
4815 heikki.linnakangas@i 2287 :UBC 0 : case PREDLOCKTAG_TUPLE:
2288 : :
2289 : : /*
2290 : : * not reachable: nothing is finer-granularity than a tuple, so we
2291 : : * should never try to promote to it.
2292 : : */
2293 : 0 : Assert(false);
2294 : : return 0;
2295 : : }
2296 : :
2297 : : /* not reachable */
2298 : 0 : Assert(false);
2299 : : return 0;
2300 : : }
2301 : :
2302 : : /*
2303 : : * For all ancestors of a newly-acquired predicate lock, increment
2304 : : * their child count in the parent hash table. If any of them have
2305 : : * more descendants than their promotion threshold, acquire the
2306 : : * coarsest such lock.
2307 : : *
2308 : : * Returns true if a parent lock was acquired and false otherwise.
2309 : : */
2310 : : static bool
4680 tgl@sss.pgh.pa.us 2311 :CBC 4219 : CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag)
2312 : : {
2313 : : PREDICATELOCKTARGETTAG targettag,
2314 : : nexttag,
2315 : : promotiontag;
2316 : : LOCALPREDICATELOCK *parentlock;
2317 : : bool found,
2318 : : promote;
2319 : :
4815 heikki.linnakangas@i 2320 : 4219 : promote = false;
2321 : :
2322 : 4219 : targettag = *reqtag;
2323 : :
2324 : : /* check parents iteratively */
2325 [ + + ]: 13822 : while (GetParentPredicateLockTag(&targettag, &nexttag))
2326 : : {
2327 : 5384 : targettag = nexttag;
2328 : 5384 : parentlock = (LOCALPREDICATELOCK *) hash_search(LocalPredicateLockHash,
2329 : : &targettag,
2330 : : HASH_ENTER,
2331 : : &found);
2332 [ + + ]: 5384 : if (!found)
2333 : : {
2334 : 3323 : parentlock->held = false;
2335 : 3323 : parentlock->childLocks = 1;
2336 : : }
2337 : : else
2338 : 2061 : parentlock->childLocks++;
2339 : :
2564 kgrittn@postgresql.o 2340 [ + + ]: 5384 : if (parentlock->childLocks >
2341 : 5384 : MaxPredicateChildLocks(&targettag))
2342 : : {
2343 : : /*
2344 : : * We should promote to this parent lock. Continue to check its
2345 : : * ancestors, however, both to get their child counts right and to
2346 : : * check whether we should just go ahead and promote to one of
2347 : : * them.
2348 : : */
4815 heikki.linnakangas@i 2349 : 333 : promotiontag = targettag;
2350 : 333 : promote = true;
2351 : : }
2352 : : }
2353 : :
2354 [ + + ]: 4219 : if (promote)
2355 : : {
2356 : : /* acquire coarsest ancestor eligible for promotion */
2357 : 333 : PredicateLockAcquire(&promotiontag);
2358 : 333 : return true;
2359 : : }
2360 : : else
2361 : 3886 : return false;
2362 : : }
2363 : :
2364 : : /*
2365 : : * When releasing a lock, decrement the child count on all ancestor
2366 : : * locks.
2367 : : *
2368 : : * This is called only when releasing a lock via
2369 : : * DeleteChildTargetLocks (i.e. when a lock becomes redundant because
2370 : : * we've acquired its parent, possibly due to promotion) or when a new
2371 : : * MVCC write lock makes the predicate lock unnecessary. There's no
2372 : : * point in calling it when locks are released at transaction end, as
2373 : : * this information is no longer needed.
2374 : : */
2375 : : static void
4680 tgl@sss.pgh.pa.us 2376 : 1380 : DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag)
2377 : : {
2378 : : PREDICATELOCKTARGETTAG parenttag,
2379 : : nexttag;
2380 : :
4815 heikki.linnakangas@i 2381 : 1380 : parenttag = *targettag;
2382 : :
2383 [ + + ]: 4140 : while (GetParentPredicateLockTag(&parenttag, &nexttag))
2384 : : {
2385 : : uint32 targettaghash;
2386 : : LOCALPREDICATELOCK *parentlock,
2387 : : *rmlock PG_USED_FOR_ASSERTS_ONLY;
2388 : :
2389 : 2760 : parenttag = nexttag;
2390 : 2760 : targettaghash = PredicateLockTargetTagHashCode(&parenttag);
2391 : : parentlock = (LOCALPREDICATELOCK *)
2392 : 2760 : hash_search_with_hash_value(LocalPredicateLockHash,
2393 : : &parenttag, targettaghash,
2394 : : HASH_FIND, NULL);
2395 : :
2396 : : /*
2397 : : * There's a small chance the parent lock doesn't exist in the lock
2398 : : * table. This can happen if we prematurely removed it because an
2399 : : * index split caused the child refcount to be off.
2400 : : */
2401 [ - + ]: 2760 : if (parentlock == NULL)
4815 heikki.linnakangas@i 2402 :UBC 0 : continue;
2403 : :
4815 heikki.linnakangas@i 2404 :CBC 2760 : parentlock->childLocks--;
2405 : :
2406 : : /*
2407 : : * Under similar circumstances the parent lock's refcount might be
2408 : : * zero. This only happens if we're holding that lock (otherwise we
2409 : : * would have removed the entry).
2410 : : */
2411 [ - + ]: 2760 : if (parentlock->childLocks < 0)
2412 : : {
4815 heikki.linnakangas@i 2413 [ # # ]:UBC 0 : Assert(parentlock->held);
2414 : 0 : parentlock->childLocks = 0;
2415 : : }
2416 : :
4815 heikki.linnakangas@i 2417 [ + + + + ]:CBC 2760 : if ((parentlock->childLocks == 0) && (!parentlock->held))
2418 : : {
2419 : : rmlock = (LOCALPREDICATELOCK *)
2420 : 750 : hash_search_with_hash_value(LocalPredicateLockHash,
2421 : : &parenttag, targettaghash,
2422 : : HASH_REMOVE, NULL);
2423 [ - + ]: 750 : Assert(rmlock == parentlock);
2424 : : }
2425 : : }
2426 : 1380 : }
2427 : :
2428 : : /*
2429 : : * Indicate that a predicate lock on the given target is held by the
2430 : : * specified transaction. Has no effect if the lock is already held.
2431 : : *
2432 : : * This updates the lock table and the sxact's lock list, and creates
2433 : : * the lock target if necessary, but does *not* do anything related to
2434 : : * granularity promotion or the local lock table. See
2435 : : * PredicateLockAcquire for that.
2436 : : */
2437 : : static void
4680 tgl@sss.pgh.pa.us 2438 : 4219 : CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
2439 : : uint32 targettaghash,
2440 : : SERIALIZABLEXACT *sxact)
2441 : : {
2442 : : PREDICATELOCKTARGET *target;
2443 : : PREDICATELOCKTAG locktag;
2444 : : PREDICATELOCK *lock;
2445 : : LWLock *partitionLock;
2446 : : bool found;
2447 : :
4815 heikki.linnakangas@i 2448 : 4219 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
2449 : :
1430 tgl@sss.pgh.pa.us 2450 : 4219 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
1857 tmunro@postgresql.or 2451 [ + + ]: 4219 : if (IsInParallelMode())
1430 tgl@sss.pgh.pa.us 2452 : 16 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
4815 heikki.linnakangas@i 2453 : 4219 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2454 : :
2455 : : /* Make sure that the target is represented. */
2456 : : target = (PREDICATELOCKTARGET *)
2457 : 4219 : hash_search_with_hash_value(PredicateLockTargetHash,
2458 : : targettag, targettaghash,
2459 : : HASH_ENTER_NULL, &found);
2460 [ - + ]: 4219 : if (!target)
4815 heikki.linnakangas@i 2461 [ # # ]:UBC 0 : ereport(ERROR,
2462 : : (errcode(ERRCODE_OUT_OF_MEMORY),
2463 : : errmsg("out of shared memory"),
2464 : : errhint("You might need to increase %s.", "max_pred_locks_per_transaction")));
4815 heikki.linnakangas@i 2465 [ + + ]:CBC 4219 : if (!found)
451 andres@anarazel.de 2466 : 3252 : dlist_init(&target->predicateLocks);
2467 : :
2468 : : /* We've got the sxact and target, make sure they're joined. */
4815 heikki.linnakangas@i 2469 : 4219 : locktag.myTarget = target;
2470 : 4219 : locktag.myXact = sxact;
2471 : : lock = (PREDICATELOCK *)
2472 : 4219 : hash_search_with_hash_value(PredicateLockHash, &locktag,
2489 tgl@sss.pgh.pa.us 2473 : 4219 : PredicateLockHashCodeFromTargetHashCode(&locktag, targettaghash),
2474 : : HASH_ENTER_NULL, &found);
4815 heikki.linnakangas@i 2475 [ - + ]: 4219 : if (!lock)
4815 heikki.linnakangas@i 2476 [ # # ]:UBC 0 : ereport(ERROR,
2477 : : (errcode(ERRCODE_OUT_OF_MEMORY),
2478 : : errmsg("out of shared memory"),
2479 : : errhint("You might need to increase %s.", "max_pred_locks_per_transaction")));
2480 : :
4815 heikki.linnakangas@i 2481 [ + + ]:CBC 4219 : if (!found)
2482 : : {
451 andres@anarazel.de 2483 : 4213 : dlist_push_tail(&target->predicateLocks, &lock->targetLink);
2484 : 4213 : dlist_push_tail(&sxact->predicateLocks, &lock->xactLink);
4752 heikki.linnakangas@i 2485 : 4213 : lock->commitSeqNo = InvalidSerCommitSeqNo;
2486 : : }
2487 : :
4815 2488 : 4219 : LWLockRelease(partitionLock);
1857 tmunro@postgresql.or 2489 [ + + ]: 4219 : if (IsInParallelMode())
1430 tgl@sss.pgh.pa.us 2490 : 16 : LWLockRelease(&sxact->perXactPredicateListLock);
2491 : 4219 : LWLockRelease(SerializablePredicateListLock);
4815 heikki.linnakangas@i 2492 : 4219 : }
2493 : :
2494 : : /*
2495 : : * Acquire a predicate lock on the specified target for the current
2496 : : * connection if not already held. This updates the local lock table
2497 : : * and uses it to implement granularity promotion. It will consolidate
2498 : : * multiple locks into a coarser lock if warranted, and will release
2499 : : * any finer-grained locks covered by the new one.
2500 : : */
2501 : : static void
4680 tgl@sss.pgh.pa.us 2502 : 26240 : PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag)
2503 : : {
2504 : : uint32 targettaghash;
2505 : : bool found;
2506 : : LOCALPREDICATELOCK *locallock;
2507 : :
2508 : : /* Do we have the lock already, or a covering lock? */
4815 heikki.linnakangas@i 2509 [ + + ]: 26240 : if (PredicateLockExists(targettag))
2510 : 22021 : return;
2511 : :
2512 [ + + ]: 26040 : if (CoarserLockCovers(targettag))
2513 : 21821 : return;
2514 : :
2515 : : /* the same hash and LW lock apply to the lock target and the local lock. */
2516 : 4219 : targettaghash = PredicateLockTargetTagHashCode(targettag);
2517 : :
2518 : : /* Acquire lock in local table */
2519 : : locallock = (LOCALPREDICATELOCK *)
2520 : 4219 : hash_search_with_hash_value(LocalPredicateLockHash,
2521 : : targettag, targettaghash,
2522 : : HASH_ENTER, &found);
2523 : 4219 : locallock->held = true;
2524 [ + + ]: 4219 : if (!found)
2525 : 3886 : locallock->childLocks = 0;
2526 : :
2527 : : /* Actually create the lock */
4692 2528 : 4219 : CreatePredicateLock(targettag, targettaghash, MySerializableXact);
2529 : :
2530 : : /*
2531 : : * Lock has been acquired. Check whether it should be promoted to a
2532 : : * coarser granularity, or whether there are finer-granularity locks to
2533 : : * clean up.
2534 : : */
4815 2535 [ + + ]: 4219 : if (CheckAndPromotePredicateLockRequest(targettag))
2536 : : {
2537 : : /*
2538 : : * Lock request was promoted to a coarser-granularity lock, and that
2539 : : * lock was acquired. It will delete this lock and any of its
2540 : : * children, so we're done.
2541 : : */
2542 : : }
2543 : : else
2544 : : {
2545 : : /* Clean up any finer-granularity locks */
2546 [ + + ]: 3886 : if (GET_PREDICATELOCKTARGETTAG_TYPE(*targettag) != PREDLOCKTAG_TUPLE)
2547 : 2353 : DeleteChildTargetLocks(targettag);
2548 : : }
2549 : : }
2550 : :
2551 : :
2552 : : /*
2553 : : * PredicateLockRelation
2554 : : *
2555 : : * Gets a predicate lock at the relation level.
2556 : : * Skip if not in full serializable transaction isolation level.
2557 : : * Skip if this is a temporary table.
2558 : : * Clear any finer-grained predicate locks this session has on the relation.
2559 : : */
2560 : : void
4680 2561 : 312840 : PredicateLockRelation(Relation relation, Snapshot snapshot)
2562 : : {
2563 : : PREDICATELOCKTARGETTAG tag;
2564 : :
4687 2565 [ + + ]: 312840 : if (!SerializationNeededForRead(relation, snapshot))
4815 2566 : 312118 : return;
2567 : :
2568 : 722 : SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2569 : : relation->rd_locator.dbOid,
2570 : : relation->rd_id);
2571 : 722 : PredicateLockAcquire(&tag);
2572 : : }
2573 : :
2574 : : /*
2575 : : * PredicateLockPage
2576 : : *
2577 : : * Gets a predicate lock at the page level.
2578 : : * Skip if not in full serializable transaction isolation level.
2579 : : * Skip if this is a temporary table.
2580 : : * Skip if a coarser predicate lock already covers this page.
2581 : : * Clear any finer-grained predicate locks this session has on the relation.
2582 : : */
2583 : : void
4680 2584 : 7925426 : PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
2585 : : {
2586 : : PREDICATELOCKTARGETTAG tag;
2587 : :
4687 2588 [ + + ]: 7925426 : if (!SerializationNeededForRead(relation, snapshot))
4815 2589 : 7923995 : return;
2590 : :
2591 : 1431 : SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2592 : : relation->rd_locator.dbOid,
2593 : : relation->rd_id,
2594 : : blkno);
2595 : 1431 : PredicateLockAcquire(&tag);
2596 : : }
2597 : :
2598 : : /*
2599 : : * PredicateLockTID
2600 : : *
2601 : : * Gets a predicate lock at the tuple level.
2602 : : * Skip if not in full serializable transaction isolation level.
2603 : : * Skip if this is a temporary table.
2604 : : */
2605 : : void
1538 tmunro@postgresql.or 2606 : 13898877 : PredicateLockTID(Relation relation, ItemPointer tid, Snapshot snapshot,
2607 : : TransactionId tuple_xid)
2608 : : {
2609 : : PREDICATELOCKTARGETTAG tag;
2610 : :
4687 heikki.linnakangas@i 2611 [ + + ]: 13898877 : if (!SerializationNeededForRead(relation, snapshot))
4815 2612 : 13875123 : return;
2613 : :
2614 : : /*
2615 : : * Return if this xact wrote it.
2616 : : */
2617 [ + - ]: 23754 : if (relation->rd_index == NULL)
2618 : : {
2619 : : /* If we wrote it; we already have a write lock. */
1538 tmunro@postgresql.or 2620 [ - + ]: 23754 : if (TransactionIdIsCurrentTransactionId(tuple_xid))
1616 tmunro@postgresql.or 2621 :UBC 0 : return;
2622 : : }
2623 : :
2624 : : /*
2625 : : * Do quick-but-not-definitive test for a relation lock first. This will
2626 : : * never cause a return when the relation is *not* locked, but will
2627 : : * occasionally let the check continue when there really *is* a relation
2628 : : * level lock.
2629 : : */
4815 heikki.linnakangas@i 2630 :CBC 23754 : SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2631 : : relation->rd_locator.dbOid,
2632 : : relation->rd_id);
2633 [ - + ]: 23754 : if (PredicateLockExists(&tag))
4815 heikki.linnakangas@i 2634 :UBC 0 : return;
2635 : :
4815 heikki.linnakangas@i 2636 :CBC 23754 : SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2637 : : relation->rd_locator.dbOid,
2638 : : relation->rd_id,
2639 : : ItemPointerGetBlockNumber(tid),
2640 : : ItemPointerGetOffsetNumber(tid));
2641 : 23754 : PredicateLockAcquire(&tag);
2642 : : }
2643 : :
2644 : :
2645 : : /*
2646 : : * DeleteLockTarget
2647 : : *
2648 : : * Remove a predicate lock target along with any locks held for it.
2649 : : *
2650 : : * Caller must hold SerializablePredicateListLock and the
2651 : : * appropriate hash partition lock for the target.
2652 : : */
2653 : : static void
4815 heikki.linnakangas@i 2654 :UBC 0 : DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash)
2655 : : {
2656 : : dlist_mutable_iter iter;
2657 : :
1430 tgl@sss.pgh.pa.us 2658 [ # # ]: 0 : Assert(LWLockHeldByMeInMode(SerializablePredicateListLock,
2659 : : LW_EXCLUSIVE));
4815 heikki.linnakangas@i 2660 [ # # ]: 0 : Assert(LWLockHeldByMe(PredicateLockHashPartitionLock(targettaghash)));
2661 : :
2662 : 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2663 : :
451 andres@anarazel.de 2664 [ # # # # ]: 0 : dlist_foreach_modify(iter, &target->predicateLocks)
2665 : : {
2666 : 0 : PREDICATELOCK *predlock =
331 tgl@sss.pgh.pa.us 2667 : 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
2668 : : bool found;
2669 : :
451 andres@anarazel.de 2670 : 0 : dlist_delete(&(predlock->xactLink));
2671 : 0 : dlist_delete(&(predlock->targetLink));
2672 : :
4815 heikki.linnakangas@i 2673 : 0 : hash_search_with_hash_value
2674 : : (PredicateLockHash,
2675 : 0 : &predlock->tag,
2676 : 0 : PredicateLockHashCodeFromTargetHashCode(&predlock->tag,
2677 : : targettaghash),
2678 : : HASH_REMOVE, &found);
2679 [ # # ]: 0 : Assert(found);
2680 : : }
2681 : 0 : LWLockRelease(SerializableXactHashLock);
2682 : :
2683 : : /* Remove the target itself, if possible. */
2684 : 0 : RemoveTargetIfNoLongerUsed(target, targettaghash);
2685 : 0 : }
2686 : :
2687 : :
2688 : : /*
2689 : : * TransferPredicateLocksToNewTarget
2690 : : *
2691 : : * Move or copy all the predicate locks for a lock target, for use by
2692 : : * index page splits/combines and other things that create or replace
2693 : : * lock targets. If 'removeOld' is true, the old locks and the target
2694 : : * will be removed.
2695 : : *
2696 : : * Returns true on success, or false if we ran out of shared memory to
2697 : : * allocate the new target or locks. Guaranteed to always succeed if
2698 : : * removeOld is set (by using the scratch entry in PredicateLockTargetHash
2699 : : * for scratch space).
2700 : : *
2701 : : * Warning: the "removeOld" option should be used only with care,
2702 : : * because this function does not (indeed, can not) update other
2703 : : * backends' LocalPredicateLockHash. If we are only adding new
2704 : : * entries, this is not a problem: the local lock table is used only
2705 : : * as a hint, so missing entries for locks that are held are
2706 : : * OK. Having entries for locks that are no longer held, as can happen
2707 : : * when using "removeOld", is not in general OK. We can only use it
2708 : : * safely when replacing a lock with a coarser-granularity lock that
2709 : : * covers it, or if we are absolutely certain that no one will need to
2710 : : * refer to that lock in the future.
2711 : : *
2712 : : * Caller must hold SerializablePredicateListLock exclusively.
2713 : : */
2714 : : static bool
4680 heikki.linnakangas@i 2715 :CBC 2 : TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
2716 : : PREDICATELOCKTARGETTAG newtargettag,
2717 : : bool removeOld)
2718 : : {
2719 : : uint32 oldtargettaghash;
2720 : : LWLock *oldpartitionLock;
2721 : : PREDICATELOCKTARGET *oldtarget;
2722 : : uint32 newtargettaghash;
2723 : : LWLock *newpartitionLock;
2724 : : bool found;
4815 2725 : 2 : bool outOfShmem = false;
2726 : :
1430 tgl@sss.pgh.pa.us 2727 [ - + ]: 2 : Assert(LWLockHeldByMeInMode(SerializablePredicateListLock,
2728 : : LW_EXCLUSIVE));
2729 : :
4815 heikki.linnakangas@i 2730 : 2 : oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
2731 : 2 : newtargettaghash = PredicateLockTargetTagHashCode(&newtargettag);
2732 : 2 : oldpartitionLock = PredicateLockHashPartitionLock(oldtargettaghash);
2733 : 2 : newpartitionLock = PredicateLockHashPartitionLock(newtargettaghash);
2734 : :
2735 [ - + ]: 2 : if (removeOld)
2736 : : {
2737 : : /*
2738 : : * Remove the dummy entry to give us scratch space, so we know we'll
2739 : : * be able to create the new lock target.
2740 : : */
4694 heikki.linnakangas@i 2741 :UBC 0 : RemoveScratchTarget(false);
2742 : : }
2743 : :
2744 : : /*
2745 : : * We must get the partition locks in ascending sequence to avoid
2746 : : * deadlocks. If old and new partitions are the same, we must request the
2747 : : * lock only once.
2748 : : */
4815 heikki.linnakangas@i 2749 [ - + ]:CBC 2 : if (oldpartitionLock < newpartitionLock)
2750 : : {
4815 heikki.linnakangas@i 2751 :LBC (1) : LWLockAcquire(oldpartitionLock,
2752 : (1) : (removeOld ? LW_EXCLUSIVE : LW_SHARED));
2753 : (1) : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2754 : : }
4815 heikki.linnakangas@i 2755 [ + - ]:CBC 2 : else if (oldpartitionLock > newpartitionLock)
2756 : : {
2757 : 2 : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2758 : 2 : LWLockAcquire(oldpartitionLock,
2759 : 2 : (removeOld ? LW_EXCLUSIVE : LW_SHARED));
2760 : : }
2761 : : else
4815 heikki.linnakangas@i 2762 :UBC 0 : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2763 : :
2764 : : /*
2765 : : * Look for the old target. If not found, that's OK; no predicate locks
2766 : : * are affected, so we can just clean up and return. If it does exist,
2767 : : * walk its list of predicate locks and move or copy them to the new
2768 : : * target.
2769 : : */
4815 heikki.linnakangas@i 2770 :CBC 2 : oldtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2771 : : &oldtargettag,
2772 : : oldtargettaghash,
2773 : : HASH_FIND, NULL);
2774 : :
2775 [ + - ]: 2 : if (oldtarget)
2776 : : {
2777 : : PREDICATELOCKTARGET *newtarget;
2778 : : PREDICATELOCKTAG newpredlocktag;
2779 : : dlist_mutable_iter iter;
2780 : :
4815 heikki.linnakangas@i 2781 :UBC 0 : newtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2782 : : &newtargettag,
2783 : : newtargettaghash,
2784 : : HASH_ENTER_NULL, &found);
2785 : :
2786 [ # # ]: 0 : if (!newtarget)
2787 : : {
2788 : : /* Failed to allocate due to insufficient shmem */
2789 : 0 : outOfShmem = true;
2790 : 0 : goto exit;
2791 : : }
2792 : :
2793 : : /* If we created a new entry, initialize it */
2794 [ # # ]: 0 : if (!found)
451 andres@anarazel.de 2795 : 0 : dlist_init(&newtarget->predicateLocks);
2796 : :
4804 magnus@hagander.net 2797 : 0 : newpredlocktag.myTarget = newtarget;
2798 : :
2799 : : /*
2800 : : * Loop through all the locks on the old target, replacing them with
2801 : : * locks on the new target.
2802 : : */
4815 heikki.linnakangas@i 2803 : 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2804 : :
451 andres@anarazel.de 2805 [ # # # # ]: 0 : dlist_foreach_modify(iter, &oldtarget->predicateLocks)
2806 : : {
2807 : 0 : PREDICATELOCK *oldpredlock =
331 tgl@sss.pgh.pa.us 2808 : 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
2809 : : PREDICATELOCK *newpredlock;
4752 heikki.linnakangas@i 2810 : 0 : SerCommitSeqNo oldCommitSeqNo = oldpredlock->commitSeqNo;
2811 : :
4815 2812 : 0 : newpredlocktag.myXact = oldpredlock->tag.myXact;
2813 : :
2814 [ # # ]: 0 : if (removeOld)
2815 : : {
451 andres@anarazel.de 2816 : 0 : dlist_delete(&(oldpredlock->xactLink));
2817 : 0 : dlist_delete(&(oldpredlock->targetLink));
2818 : :
4815 heikki.linnakangas@i 2819 : 0 : hash_search_with_hash_value
2820 : : (PredicateLockHash,
2821 : 0 : &oldpredlock->tag,
2489 tgl@sss.pgh.pa.us 2822 : 0 : PredicateLockHashCodeFromTargetHashCode(&oldpredlock->tag,
2823 : : oldtargettaghash),
2824 : : HASH_REMOVE, &found);
4815 heikki.linnakangas@i 2825 [ # # ]: 0 : Assert(found);
2826 : : }
2827 : :
2828 : : newpredlock = (PREDICATELOCK *)
4680 tgl@sss.pgh.pa.us 2829 : 0 : hash_search_with_hash_value(PredicateLockHash,
2830 : : &newpredlocktag,
2489 2831 : 0 : PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
2832 : : newtargettaghash),
2833 : : HASH_ENTER_NULL,
2834 : : &found);
4815 heikki.linnakangas@i 2835 [ # # ]: 0 : if (!newpredlock)
2836 : : {
2837 : : /* Out of shared memory. Undo what we've done so far. */
2838 : 0 : LWLockRelease(SerializableXactHashLock);
2839 : 0 : DeleteLockTarget(newtarget, newtargettaghash);
2840 : 0 : outOfShmem = true;
2841 : 0 : goto exit;
2842 : : }
4804 magnus@hagander.net 2843 [ # # ]: 0 : if (!found)
2844 : : {
451 andres@anarazel.de 2845 : 0 : dlist_push_tail(&(newtarget->predicateLocks),
2846 : : &(newpredlock->targetLink));
2847 : 0 : dlist_push_tail(&(newpredlocktag.myXact->predicateLocks),
2848 : : &(newpredlock->xactLink));
4752 heikki.linnakangas@i 2849 : 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
2850 : : }
2851 : : else
2852 : : {
2853 [ # # ]: 0 : if (newpredlock->commitSeqNo < oldCommitSeqNo)
2854 : 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
2855 : : }
2856 : :
2857 [ # # ]: 0 : Assert(newpredlock->commitSeqNo != 0);
2858 [ # # # # ]: 0 : Assert((newpredlock->commitSeqNo == InvalidSerCommitSeqNo)
2859 : : || (newpredlock->tag.myXact == OldCommittedSxact));
2860 : : }
4815 2861 : 0 : LWLockRelease(SerializableXactHashLock);
2862 : :
2863 [ # # ]: 0 : if (removeOld)
2864 : : {
451 andres@anarazel.de 2865 [ # # ]: 0 : Assert(dlist_is_empty(&oldtarget->predicateLocks));
4815 heikki.linnakangas@i 2866 : 0 : RemoveTargetIfNoLongerUsed(oldtarget, oldtargettaghash);
2867 : : }
2868 : : }
2869 : :
2870 : :
4815 heikki.linnakangas@i 2871 :CBC 2 : exit:
2872 : : /* Release partition locks in reverse order of acquisition. */
2873 [ - + ]: 2 : if (oldpartitionLock < newpartitionLock)
2874 : : {
4815 heikki.linnakangas@i 2875 :LBC (1) : LWLockRelease(newpartitionLock);
2876 : (1) : LWLockRelease(oldpartitionLock);
2877 : : }
4815 heikki.linnakangas@i 2878 [ + - ]:CBC 2 : else if (oldpartitionLock > newpartitionLock)
2879 : : {
2880 : 2 : LWLockRelease(oldpartitionLock);
2881 : 2 : LWLockRelease(newpartitionLock);
2882 : : }
2883 : : else
4815 heikki.linnakangas@i 2884 :UBC 0 : LWLockRelease(newpartitionLock);
2885 : :
4815 heikki.linnakangas@i 2886 [ - + ]:CBC 2 : if (removeOld)
2887 : : {
2888 : : /* We shouldn't run out of memory if we're moving locks */
4815 heikki.linnakangas@i 2889 [ # # ]:UBC 0 : Assert(!outOfShmem);
2890 : :
2891 : : /* Put the scratch entry back */
4694 2892 : 0 : RestoreScratchTarget(false);
2893 : : }
2894 : :
4815 heikki.linnakangas@i 2895 :CBC 2 : return !outOfShmem;
2896 : : }
2897 : :
2898 : : /*
2899 : : * Drop all predicate locks of any granularity from the specified relation,
2900 : : * which can be a heap relation or an index relation. If 'transfer' is true,
2901 : : * acquire a relation lock on the heap for any transactions with any lock(s)
2902 : : * on the specified relation.
2903 : : *
2904 : : * This requires grabbing a lot of LW locks and scanning the entire lock
2905 : : * target table for matches. That makes this more expensive than most
2906 : : * predicate lock management functions, but it will only be called for DDL
2907 : : * type commands that are expensive anyway, and there are fast returns when
2908 : : * no serializable transactions are active or the relation is temporary.
2909 : : *
2910 : : * We don't use the TransferPredicateLocksToNewTarget function because it
2911 : : * acquires its own locks on the partitions of the two targets involved,
2912 : : * and we'll already be holding all partition locks.
2913 : : *
2914 : : * We can't throw an error from here, because the call could be from a
2915 : : * transaction which is not serializable.
2916 : : *
2917 : : * NOTE: This is currently only called with transfer set to true, but that may
2918 : : * change. If we decide to clean up the locks from a table on commit of a
2919 : : * transaction which executed DROP TABLE, the false condition will be useful.
2920 : : */
2921 : : static void
4680 2922 : 15884 : DropAllPredicateLocksFromTable(Relation relation, bool transfer)
2923 : : {
2924 : : HASH_SEQ_STATUS seqstat;
2925 : : PREDICATELOCKTARGET *oldtarget;
2926 : : PREDICATELOCKTARGET *heaptarget;
2927 : : Oid dbId;
2928 : : Oid relId;
2929 : : Oid heapId;
2930 : : int i;
2931 : : bool isIndex;
2932 : : bool found;
2933 : : uint32 heaptargettaghash;
2934 : :
2935 : : /*
2936 : : * Bail out quickly if there are no serializable transactions running.
2937 : : * It's safe to check this without taking locks because the caller is
2938 : : * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2939 : : * would matter here can be acquired while that is held.
2940 : : */
4694 2941 [ + + ]: 15884 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
2942 : 15835 : return;
2943 : :
4687 2944 [ + + ]: 65 : if (!PredicateLockingNeededForRelation(relation))
4694 2945 : 16 : return;
2946 : :
648 rhaas@postgresql.org 2947 : 49 : dbId = relation->rd_locator.dbOid;
4694 heikki.linnakangas@i 2948 : 49 : relId = relation->rd_id;
2949 [ + + ]: 49 : if (relation->rd_index == NULL)
2950 : : {
4694 heikki.linnakangas@i 2951 :GBC 2 : isIndex = false;
2952 : 2 : heapId = relId;
2953 : : }
2954 : : else
2955 : : {
4694 heikki.linnakangas@i 2956 :CBC 47 : isIndex = true;
2957 : 47 : heapId = relation->rd_index->indrelid;
2958 : : }
2959 [ - + ]: 49 : Assert(heapId != InvalidOid);
2489 tgl@sss.pgh.pa.us 2960 [ - + - - ]: 49 : Assert(transfer || !isIndex); /* index OID only makes sense with
2961 : : * transfer */
2962 : :
2963 : : /* Retrieve first time needed, then keep. */
4694 heikki.linnakangas@i 2964 : 49 : heaptargettaghash = 0;
2965 : 49 : heaptarget = NULL;
2966 : :
2967 : : /* Acquire locks on all lock partitions */
1430 tgl@sss.pgh.pa.us 2968 : 49 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
4694 heikki.linnakangas@i 2969 [ + + ]: 833 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
3730 rhaas@postgresql.org 2970 : 784 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_EXCLUSIVE);
4694 heikki.linnakangas@i 2971 : 49 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2972 : :
2973 : : /*
2974 : : * Remove the dummy entry to give us scratch space, so we know we'll be
2975 : : * able to create the new lock target.
2976 : : */
2977 [ + - ]: 49 : if (transfer)
2978 : 49 : RemoveScratchTarget(true);
2979 : :
2980 : : /* Scan through target map */
2981 : 49 : hash_seq_init(&seqstat, PredicateLockTargetHash);
2982 : :
2983 [ + + ]: 88 : while ((oldtarget = (PREDICATELOCKTARGET *) hash_seq_search(&seqstat)))
2984 : : {
2985 : : dlist_mutable_iter iter;
2986 : :
2987 : : /*
2988 : : * Check whether this is a target which needs attention.
2989 : : */
2990 [ + - ]: 39 : if (GET_PREDICATELOCKTARGETTAG_RELATION(oldtarget->tag) != relId)
2991 : 39 : continue; /* wrong relation id */
4694 heikki.linnakangas@i 2992 [ # # ]:UBC 0 : if (GET_PREDICATELOCKTARGETTAG_DB(oldtarget->tag) != dbId)
2993 : 0 : continue; /* wrong database id */
2994 [ # # # # ]: 0 : if (transfer && !isIndex
2995 [ # # # # ]: 0 : && GET_PREDICATELOCKTARGETTAG_TYPE(oldtarget->tag) == PREDLOCKTAG_RELATION)
2996 : 0 : continue; /* already the right lock */
2997 : :
2998 : : /*
2999 : : * If we made it here, we have work to do. We make sure the heap
3000 : : * relation lock exists, then we walk the list of predicate locks for
3001 : : * the old target we found, moving all locks to the heap relation lock
3002 : : * -- unless they already hold that.
3003 : : */
3004 : :
3005 : : /*
3006 : : * First make sure we have the heap relation target. We only need to
3007 : : * do this once.
3008 : : */
3009 [ # # # # ]: 0 : if (transfer && heaptarget == NULL)
3010 : : {
3011 : : PREDICATELOCKTARGETTAG heaptargettag;
3012 : :
3013 : 0 : SET_PREDICATELOCKTARGETTAG_RELATION(heaptargettag, dbId, heapId);
3014 : 0 : heaptargettaghash = PredicateLockTargetTagHashCode(&heaptargettag);
3015 : 0 : heaptarget = hash_search_with_hash_value(PredicateLockTargetHash,
3016 : : &heaptargettag,
3017 : : heaptargettaghash,
3018 : : HASH_ENTER, &found);
3019 [ # # ]: 0 : if (!found)
451 andres@anarazel.de 3020 : 0 : dlist_init(&heaptarget->predicateLocks);
3021 : : }
3022 : :
3023 : : /*
3024 : : * Loop through all the locks on the old target, replacing them with
3025 : : * locks on the new target.
3026 : : */
3027 [ # # # # ]: 0 : dlist_foreach_modify(iter, &oldtarget->predicateLocks)
3028 : : {
3029 : 0 : PREDICATELOCK *oldpredlock =
331 tgl@sss.pgh.pa.us 3030 : 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
3031 : : PREDICATELOCK *newpredlock;
3032 : : SerCommitSeqNo oldCommitSeqNo;
3033 : : SERIALIZABLEXACT *oldXact;
3034 : :
3035 : : /*
3036 : : * Remove the old lock first. This avoids the chance of running
3037 : : * out of lock structure entries for the hash table.
3038 : : */
4694 heikki.linnakangas@i 3039 : 0 : oldCommitSeqNo = oldpredlock->commitSeqNo;
3040 : 0 : oldXact = oldpredlock->tag.myXact;
3041 : :
451 andres@anarazel.de 3042 : 0 : dlist_delete(&(oldpredlock->xactLink));
3043 : :
3044 : : /*
3045 : : * No need for retail delete from oldtarget list, we're removing
3046 : : * the whole target anyway.
3047 : : */
4694 heikki.linnakangas@i 3048 : 0 : hash_search(PredicateLockHash,
3049 : 0 : &oldpredlock->tag,
3050 : : HASH_REMOVE, &found);
3051 [ # # ]: 0 : Assert(found);
3052 : :
3053 [ # # ]: 0 : if (transfer)
3054 : : {
3055 : : PREDICATELOCKTAG newpredlocktag;
3056 : :
3057 : 0 : newpredlocktag.myTarget = heaptarget;
3058 : 0 : newpredlocktag.myXact = oldXact;
3059 : : newpredlock = (PREDICATELOCK *)
4680 tgl@sss.pgh.pa.us 3060 : 0 : hash_search_with_hash_value(PredicateLockHash,
3061 : : &newpredlocktag,
2489 3062 : 0 : PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
3063 : : heaptargettaghash),
3064 : : HASH_ENTER,
3065 : : &found);
4694 heikki.linnakangas@i 3066 [ # # ]: 0 : if (!found)
3067 : : {
451 andres@anarazel.de 3068 : 0 : dlist_push_tail(&(heaptarget->predicateLocks),
3069 : : &(newpredlock->targetLink));
3070 : 0 : dlist_push_tail(&(newpredlocktag.myXact->predicateLocks),
3071 : : &(newpredlock->xactLink));
4694 heikki.linnakangas@i 3072 : 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
3073 : : }
3074 : : else
3075 : : {
3076 [ # # ]: 0 : if (newpredlock->commitSeqNo < oldCommitSeqNo)
3077 : 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
3078 : : }
3079 : :
3080 [ # # ]: 0 : Assert(newpredlock->commitSeqNo != 0);
3081 [ # # # # ]: 0 : Assert((newpredlock->commitSeqNo == InvalidSerCommitSeqNo)
3082 : : || (newpredlock->tag.myXact == OldCommittedSxact));
3083 : : }
3084 : : }
3085 : :
3086 : 0 : hash_search(PredicateLockTargetHash, &oldtarget->tag, HASH_REMOVE,
3087 : : &found);
3088 [ # # ]: 0 : Assert(found);
3089 : : }
3090 : :
3091 : : /* Put the scratch entry back */
4694 heikki.linnakangas@i 3092 [ + - ]:CBC 49 : if (transfer)
3093 : 49 : RestoreScratchTarget(true);
3094 : :
3095 : : /* Release locks in reverse order */
3096 : 49 : LWLockRelease(SerializableXactHashLock);
3097 [ + + ]: 833 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
3730 rhaas@postgresql.org 3098 : 784 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
1430 tgl@sss.pgh.pa.us 3099 : 49 : LWLockRelease(SerializablePredicateListLock);
3100 : : }
3101 : :
3102 : : /*
3103 : : * TransferPredicateLocksToHeapRelation
3104 : : * For all transactions, transfer all predicate locks for the given
3105 : : * relation to a single relation lock on the heap.
3106 : : */
3107 : : void
4680 heikki.linnakangas@i 3108 : 15884 : TransferPredicateLocksToHeapRelation(Relation relation)
3109 : : {
4694 3110 : 15884 : DropAllPredicateLocksFromTable(relation, true);
3111 : 15884 : }
3112 : :
3113 : :
3114 : : /*
3115 : : * PredicateLockPageSplit
3116 : : *
3117 : : * Copies any predicate locks for the old page to the new page.
3118 : : * Skip if this is a temporary table or toast table.
3119 : : *
3120 : : * NOTE: A page split (or overflow) affects all serializable transactions,
3121 : : * even if it occurs in the context of another transaction isolation level.
3122 : : *
3123 : : * NOTE: This currently leaves the local copy of the locks without
3124 : : * information on the new lock which is in shared memory. This could cause
3125 : : * problems if enough page splits occur on locked pages without the processes
3126 : : * which hold the locks getting in and noticing.
3127 : : */
3128 : : void
4680 3129 : 29811 : PredicateLockPageSplit(Relation relation, BlockNumber oldblkno,
3130 : : BlockNumber newblkno)
3131 : : {
3132 : : PREDICATELOCKTARGETTAG oldtargettag;
3133 : : PREDICATELOCKTARGETTAG newtargettag;
3134 : : bool success;
3135 : :
3136 : : /*
3137 : : * Bail out quickly if there are no serializable transactions running.
3138 : : *
3139 : : * It's safe to do this check without taking any additional locks. Even if
3140 : : * a serializable transaction starts concurrently, we know it can't take
3141 : : * any SIREAD locks on the page being split because the caller is holding
3142 : : * the associated buffer page lock. Memory reordering isn't an issue; the
3143 : : * memory barrier in the LWLock acquisition guarantees that this read
3144 : : * occurs while the buffer page lock is held.
3145 : : */
4738 rhaas@postgresql.org 3146 [ + + ]: 29811 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
3147 : 29809 : return;
3148 : :
4687 heikki.linnakangas@i 3149 [ + + ]: 19 : if (!PredicateLockingNeededForRelation(relation))
4815 3150 : 17 : return;
3151 : :
3152 [ - + ]: 2 : Assert(oldblkno != newblkno);
3153 [ - + ]: 2 : Assert(BlockNumberIsValid(oldblkno));
3154 [ - + ]: 2 : Assert(BlockNumberIsValid(newblkno));
3155 : :
3156 : 2 : SET_PREDICATELOCKTARGETTAG_PAGE(oldtargettag,
3157 : : relation->rd_locator.dbOid,
3158 : : relation->rd_id,
3159 : : oldblkno);
3160 : 2 : SET_PREDICATELOCKTARGETTAG_PAGE(newtargettag,
3161 : : relation->rd_locator.dbOid,
3162 : : relation->rd_id,
3163 : : newblkno);
3164 : :
1430 tgl@sss.pgh.pa.us 3165 : 2 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
3166 : :
3167 : : /*
3168 : : * Try copying the locks over to the new page's tag, creating it if
3169 : : * necessary.
3170 : : */
4815 heikki.linnakangas@i 3171 : 2 : success = TransferPredicateLocksToNewTarget(oldtargettag,
3172 : : newtargettag,
3173 : : false);
3174 : :
3175 [ - + ]: 2 : if (!success)
3176 : : {
3177 : : /*
3178 : : * No more predicate lock entries are available. Failure isn't an
3179 : : * option here, so promote the page lock to a relation lock.
3180 : : */
3181 : :
3182 : : /* Get the parent relation lock's lock tag */
4815 heikki.linnakangas@i 3183 :UBC 0 : success = GetParentPredicateLockTag(&oldtargettag,
3184 : : &newtargettag);
3185 [ # # ]: 0 : Assert(success);
3186 : :
3187 : : /*
3188 : : * Move the locks to the parent. This shouldn't fail.
3189 : : *
3190 : : * Note that here we are removing locks held by other backends,
3191 : : * leading to a possible inconsistency in their local lock hash table.
3192 : : * This is OK because we're replacing it with a lock that covers the
3193 : : * old one.
3194 : : */
3195 : 0 : success = TransferPredicateLocksToNewTarget(oldtargettag,
3196 : : newtargettag,
3197 : : true);
3198 [ # # ]: 0 : Assert(success);
3199 : : }
3200 : :
1430 tgl@sss.pgh.pa.us 3201 :CBC 2 : LWLockRelease(SerializablePredicateListLock);
3202 : : }
3203 : :
3204 : : /*
3205 : : * PredicateLockPageCombine
3206 : : *
3207 : : * Combines predicate locks for two existing pages.
3208 : : * Skip if this is a temporary table or toast table.
3209 : : *
3210 : : * NOTE: A page combine affects all serializable transactions, even if it
3211 : : * occurs in the context of another transaction isolation level.
3212 : : */
3213 : : void
4680 heikki.linnakangas@i 3214 : 2900 : PredicateLockPageCombine(Relation relation, BlockNumber oldblkno,
3215 : : BlockNumber newblkno)
3216 : : {
3217 : : /*
3218 : : * Page combines differ from page splits in that we ought to be able to
3219 : : * remove the locks on the old page after transferring them to the new
3220 : : * page, instead of duplicating them. However, because we can't edit other
3221 : : * backends' local lock tables, removing the old lock would leave them
3222 : : * with an entry in their LocalPredicateLockHash for a lock they're not
3223 : : * holding, which isn't acceptable. So we wind up having to do the same
3224 : : * work as a page split, acquiring a lock on the new page and keeping the
3225 : : * old page locked too. That can lead to some false positives, but should
3226 : : * be rare in practice.
3227 : : */
4793 3228 : 2900 : PredicateLockPageSplit(relation, oldblkno, newblkno);
4815 3229 : 2900 : }
3230 : :
3231 : : /*
3232 : : * Walk the list of in-progress serializable transactions and find the new
3233 : : * xmin.
3234 : : */
3235 : : static void
3236 : 854 : SetNewSxactGlobalXmin(void)
3237 : : {
3238 : : dlist_iter iter;
3239 : :
3240 [ - + ]: 854 : Assert(LWLockHeldByMe(SerializableXactHashLock));
3241 : :
3242 : 854 : PredXact->SxactGlobalXmin = InvalidTransactionId;
3243 : 854 : PredXact->SxactGlobalXminCount = 0;
3244 : :
451 andres@anarazel.de 3245 [ + - + + ]: 3264 : dlist_foreach(iter, &PredXact->activeList)
3246 : : {
3247 : 2410 : SERIALIZABLEXACT *sxact =
331 tgl@sss.pgh.pa.us 3248 : 2410 : dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
3249 : :
4681 heikki.linnakangas@i 3250 [ + + ]: 2410 : if (!SxactIsRolledBack(sxact)
4815 3251 [ + + ]: 2112 : && !SxactIsCommitted(sxact)
3252 [ + - ]: 18 : && sxact != OldCommittedSxact)
3253 : : {
3254 [ - + ]: 18 : Assert(sxact->xmin != InvalidTransactionId);
3255 [ - + ]: 18 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)
4815 heikki.linnakangas@i 3256 [ # # ]:LBC (2) : || TransactionIdPrecedes(sxact->xmin,
3257 : (2) : PredXact->SxactGlobalXmin))
3258 : : {
4815 heikki.linnakangas@i 3259 :CBC 18 : PredXact->SxactGlobalXmin = sxact->xmin;
3260 : 18 : PredXact->SxactGlobalXminCount = 1;
3261 : : }
4815 heikki.linnakangas@i 3262 [ # # ]:LBC (2) : else if (TransactionIdEquals(sxact->xmin,
3263 : : PredXact->SxactGlobalXmin))
4815 heikki.linnakangas@i 3264 :UBC 0 : PredXact->SxactGlobalXminCount++;
3265 : : }
3266 : : }
3267 : :
1430 tgl@sss.pgh.pa.us 3268 :CBC 854 : SerialSetActiveSerXmin(PredXact->SxactGlobalXmin);
4815 heikki.linnakangas@i 3269 : 854 : }
3270 : :
3271 : : /*
3272 : : * ReleasePredicateLocks
3273 : : *
3274 : : * Releases predicate locks based on completion of the current transaction,
3275 : : * whether committed or rolled back. It can also be called for a read only
3276 : : * transaction when it becomes impossible for the transaction to become
3277 : : * part of a dangerous structure.
3278 : : *
3279 : : * We do nothing unless this is a serializable transaction.
3280 : : *
3281 : : * This method must ensure that shared memory hash tables are cleaned
3282 : : * up in some relatively timely fashion.
3283 : : *
3284 : : * If this transaction is committing and is holding any predicate locks,
3285 : : * it must be added to a list of completed serializable transactions still
3286 : : * holding locks.
3287 : : *
3288 : : * If isReadOnlySafe is true, then predicate locks are being released before
3289 : : * the end of the transaction because MySerializableXact has been determined
3290 : : * to be RO_SAFE. In non-parallel mode we can release it completely, but it
3291 : : * in parallel mode we partially release the SERIALIZABLEXACT and keep it
3292 : : * around until the end of the transaction, allowing each backend to clear its
3293 : : * MySerializableXact variable and benefit from the optimization in its own
3294 : : * time.
3295 : : */
3296 : : void
1857 tmunro@postgresql.or 3297 : 432961 : ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
3298 : : {
405 3299 : 432961 : bool partiallyReleasing = false;
3300 : : bool needToClear;
3301 : : SERIALIZABLEXACT *roXact;
3302 : : dlist_mutable_iter iter;
3303 : :
3304 : : /*
3305 : : * We can't trust XactReadOnly here, because a transaction which started
3306 : : * as READ WRITE can show as READ ONLY later, e.g., within
3307 : : * subtransactions. We want to flag a transaction as READ ONLY if it
3308 : : * commits without writing so that de facto READ ONLY transactions get the
3309 : : * benefit of some RO optimizations, so we will use this local variable to
3310 : : * get some cleanup logic right which is based on whether the transaction
3311 : : * was declared READ ONLY at the top level.
3312 : : */
3313 : : bool topLevelIsDeclaredReadOnly;
3314 : :
3315 : : /* We can't be both committing and releasing early due to RO_SAFE. */
1857 3316 [ + + - + ]: 432961 : Assert(!(isCommit && isReadOnlySafe));
3317 : :
3318 : : /* Are we at the end of a transaction, that is, a commit or abort? */
3319 [ + + ]: 432961 : if (!isReadOnlySafe)
3320 : : {
3321 : : /*
3322 : : * Parallel workers mustn't release predicate locks at the end of
3323 : : * their transaction. The leader will do that at the end of its
3324 : : * transaction.
3325 : : */
3326 [ + + ]: 432928 : if (IsParallelWorker())
3327 : : {
3328 : 3966 : ReleasePredicateLocksLocal();
3329 : 431418 : return;
3330 : : }
3331 : :
3332 : : /*
3333 : : * By the time the leader in a parallel query reaches end of
3334 : : * transaction, it has waited for all workers to exit.
3335 : : */
3336 [ - + ]: 428962 : Assert(!ParallelContextActive());
3337 : :
3338 : : /*
3339 : : * If the leader in a parallel query earlier stashed a partially
3340 : : * released SERIALIZABLEXACT for final clean-up at end of transaction
3341 : : * (because workers might still have been accessing it), then it's
3342 : : * time to restore it.
3343 : : */
3344 [ + + ]: 428962 : if (SavedSerializableXact != InvalidSerializableXact)
3345 : : {
3346 [ - + ]: 1 : Assert(MySerializableXact == InvalidSerializableXact);
3347 : 1 : MySerializableXact = SavedSerializableXact;
3348 : 1 : SavedSerializableXact = InvalidSerializableXact;
3349 [ - + ]: 1 : Assert(SxactIsPartiallyReleased(MySerializableXact));
3350 : : }
3351 : : }
3352 : :
4815 heikki.linnakangas@i 3353 [ + + ]: 428995 : if (MySerializableXact == InvalidSerializableXact)
3354 : : {
3355 [ - + ]: 427449 : Assert(LocalPredicateLockHash == NULL);
3356 : 427449 : return;
3357 : : }
3358 : :
3088 kgrittn@postgresql.o 3359 : 1546 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
3360 : :
3361 : : /*
3362 : : * If the transaction is committing, but it has been partially released
3363 : : * already, then treat this as a roll back. It was marked as rolled back.
3364 : : */
1857 tmunro@postgresql.or 3365 [ + + + + ]: 1546 : if (isCommit && SxactIsPartiallyReleased(MySerializableXact))
3366 : 2 : isCommit = false;
3367 : :
3368 : : /*
3369 : : * If we're called in the middle of a transaction because we discovered
3370 : : * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3371 : : * it (that is, release the predicate locks and conflicts, but not the
3372 : : * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3373 : : */
3374 [ + + + + ]: 1546 : if (isReadOnlySafe && IsInParallelMode())
3375 : : {
3376 : : /*
3377 : : * The leader needs to stash a pointer to it, so that it can
3378 : : * completely release it at end-of-transaction.
3379 : : */
3380 [ + + ]: 5 : if (!IsParallelWorker())
3381 : 1 : SavedSerializableXact = MySerializableXact;
3382 : :
3383 : : /*
3384 : : * The first backend to reach this condition will partially release
3385 : : * the SERIALIZABLEXACT. All others will just clear their
3386 : : * backend-local state so that they stop doing SSI checks for the rest
3387 : : * of the transaction.
3388 : : */
3389 [ + + ]: 5 : if (SxactIsPartiallyReleased(MySerializableXact))
3390 : : {
3391 : 3 : LWLockRelease(SerializableXactHashLock);
3392 : 3 : ReleasePredicateLocksLocal();
3393 : 3 : return;
3394 : : }
3395 : : else
3396 : : {
3397 : 2 : MySerializableXact->flags |= SXACT_FLAG_PARTIALLY_RELEASED;
405 3398 : 2 : partiallyReleasing = true;
3399 : : /* ... and proceed to perform the partial release below. */
3400 : : }
3401 : : }
4815 heikki.linnakangas@i 3402 [ + + - + ]: 1543 : Assert(!isCommit || SxactIsPrepared(MySerializableXact));
4687 3403 [ + + - + ]: 1543 : Assert(!isCommit || !SxactIsDoomed(MySerializableXact));
4815 3404 [ - + ]: 1543 : Assert(!SxactIsCommitted(MySerializableXact));
1857 tmunro@postgresql.or 3405 [ + + - + ]: 1543 : Assert(SxactIsPartiallyReleased(MySerializableXact)
3406 : : || !SxactIsRolledBack(MySerializableXact));
3407 : :
3408 : : /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3088 kgrittn@postgresql.o 3409 [ + + - + ]: 1543 : Assert(MySerializableXact->pid == 0 || IsolationIsSerializable());
3410 : :
3411 : : /* We'd better not already be on the cleanup list. */
4692 heikki.linnakangas@i 3412 [ - + ]: 1543 : Assert(!SxactIsOnFinishedList(MySerializableXact));
3413 : :
4815 3414 : 1543 : topLevelIsDeclaredReadOnly = SxactIsReadOnly(MySerializableXact);
3415 : :
3416 : : /*
3417 : : * We don't hold XidGenLock lock here, assuming that TransactionId is
3418 : : * atomic!
3419 : : *
3420 : : * If this value is changing, we don't care that much whether we get the
3421 : : * old or new value -- it is just used to determine how far
3422 : : * SxactGlobalXmin must advance before this transaction can be fully
3423 : : * cleaned up. The worst that could happen is we wait for one more
3424 : : * transaction to complete before freeing some RAM; correctness of visible
3425 : : * behavior is not affected.
3426 : : */
128 heikki.linnakangas@i 3427 :GNC 1543 : MySerializableXact->finishedBefore = XidFromFullTransactionId(TransamVariables->nextXid);
3428 : :
3429 : : /*
3430 : : * If it's not a commit it's either a rollback or a read-only transaction
3431 : : * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3432 : : */
4815 heikki.linnakangas@i 3433 [ + + ]:CBC 1543 : if (isCommit)
3434 : : {
3435 : 1222 : MySerializableXact->flags |= SXACT_FLAG_COMMITTED;
3436 : 1222 : MySerializableXact->commitSeqNo = ++(PredXact->LastSxactCommitSeqNo);
3437 : : /* Recognize implicit read-only transaction (commit without write). */
4692 3438 [ + + ]: 1222 : if (!MyXactDidWrite)
4815 3439 : 233 : MySerializableXact->flags |= SXACT_FLAG_READ_ONLY;
3440 : : }
3441 : : else
3442 : : {
3443 : : /*
3444 : : * The DOOMED flag indicates that we intend to roll back this
3445 : : * transaction and so it should not cause serialization failures for
3446 : : * other transactions that conflict with it. Note that this flag might
3447 : : * already be set, if another backend marked this transaction for
3448 : : * abort.
3449 : : *
3450 : : * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3451 : : * has been called, and so the SerializableXact is eligible for
3452 : : * cleanup. This means it should not be considered when calculating
3453 : : * SxactGlobalXmin.
3454 : : */
1802 tmunro@postgresql.or 3455 : 321 : MySerializableXact->flags |= SXACT_FLAG_DOOMED;
4681 heikki.linnakangas@i 3456 : 321 : MySerializableXact->flags |= SXACT_FLAG_ROLLED_BACK;
3457 : :
3458 : : /*
3459 : : * If the transaction was previously prepared, but is now failing due
3460 : : * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3461 : : * prepare, clear the prepared flag. This simplifies conflict
3462 : : * checking.
3463 : : */
4665 3464 : 321 : MySerializableXact->flags &= ~SXACT_FLAG_PREPARED;
3465 : : }
3466 : :
4815 3467 [ + + ]: 1543 : if (!topLevelIsDeclaredReadOnly)
3468 : : {
3469 [ - + ]: 1435 : Assert(PredXact->WritableSxactCount > 0);
3470 [ + + ]: 1435 : if (--(PredXact->WritableSxactCount) == 0)
3471 : : {
3472 : : /*
3473 : : * Release predicate locks and rw-conflicts in for all committed
3474 : : * transactions. There are no longer any transactions which might
3475 : : * conflict with the locks and no chance for new transactions to
3476 : : * overlap. Similarly, existing conflicts in can't cause pivots,
3477 : : * and any conflicts in which could have completed a dangerous
3478 : : * structure would already have caused a rollback, so any
3479 : : * remaining ones must be benign.
3480 : : */
3481 : 846 : PredXact->CanPartialClearThrough = PredXact->LastSxactCommitSeqNo;
3482 : : }
3483 : : }
3484 : : else
3485 : : {
3486 : : /*
3487 : : * Read-only transactions: clear the list of transactions that might
3488 : : * make us unsafe. Note that we use 'inLink' for the iteration as
3489 : : * opposed to 'outLink' for the r/w xacts.
3490 : : */
451 andres@anarazel.de 3491 [ + - + + ]: 150 : dlist_foreach_modify(iter, &MySerializableXact->possibleUnsafeConflicts)
3492 : : {
3493 : 42 : RWConflict possibleUnsafeConflict =
331 tgl@sss.pgh.pa.us 3494 : 42 : dlist_container(RWConflictData, inLink, iter.cur);
3495 : :
4815 heikki.linnakangas@i 3496 [ - + ]: 42 : Assert(!SxactIsReadOnly(possibleUnsafeConflict->sxactOut));
3497 [ - + ]: 42 : Assert(MySerializableXact == possibleUnsafeConflict->sxactIn);
3498 : :
3499 : 42 : ReleaseRWConflict(possibleUnsafeConflict);
3500 : : }
3501 : : }
3502 : :
3503 : : /* Check for conflict out to old committed transactions. */
3504 [ + + ]: 1543 : if (isCommit
3505 [ + + ]: 1222 : && !SxactIsReadOnly(MySerializableXact)
3506 [ - + ]: 989 : && SxactHasSummaryConflictOut(MySerializableXact))
3507 : : {
3508 : : /*
3509 : : * we don't know which old committed transaction we conflicted with,
3510 : : * so be conservative and use FirstNormalSerCommitSeqNo here
3511 : : */
4815 heikki.linnakangas@i 3512 :UBC 0 : MySerializableXact->SeqNo.earliestOutConflictCommit =
3513 : : FirstNormalSerCommitSeqNo;
3514 : 0 : MySerializableXact->flags |= SXACT_FLAG_CONFLICT_OUT;
3515 : : }
3516 : :
3517 : : /*
3518 : : * Release all outConflicts to committed transactions. If we're rolling
3519 : : * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3520 : : * previously committed transactions.
3521 : : */
451 andres@anarazel.de 3522 [ + - + + ]:CBC 2222 : dlist_foreach_modify(iter, &MySerializableXact->outConflicts)
3523 : : {
3524 : 679 : RWConflict conflict =
331 tgl@sss.pgh.pa.us 3525 : 679 : dlist_container(RWConflictData, outLink, iter.cur);
3526 : :
4815 heikki.linnakangas@i 3527 [ + + ]: 679 : if (isCommit
3528 [ + + ]: 451 : && !SxactIsReadOnly(MySerializableXact)
3529 [ + + ]: 343 : && SxactIsCommitted(conflict->sxactIn))
3530 : : {
3531 [ - + ]: 96 : if ((MySerializableXact->flags & SXACT_FLAG_CONFLICT_OUT) == 0
4665 heikki.linnakangas@i 3532 [ # # ]:UBC 0 : || conflict->sxactIn->prepareSeqNo < MySerializableXact->SeqNo.earliestOutConflictCommit)
4665 heikki.linnakangas@i 3533 :CBC 96 : MySerializableXact->SeqNo.earliestOutConflictCommit = conflict->sxactIn->prepareSeqNo;
4815 3534 : 96 : MySerializableXact->flags |= SXACT_FLAG_CONFLICT_OUT;
3535 : : }
3536 : :
3537 [ + + ]: 679 : if (!isCommit
3538 [ + + ]: 451 : || SxactIsCommitted(conflict->sxactIn)
3539 [ - + ]: 333 : || (conflict->sxactIn->SeqNo.lastCommitBeforeSnapshot >= PredXact->LastSxactCommitSeqNo))
3540 : 346 : ReleaseRWConflict(conflict);
3541 : : }
3542 : :
3543 : : /*
3544 : : * Release all inConflicts from committed and read-only transactions. If
3545 : : * we're rolling back, clear them all.
3546 : : */
451 andres@anarazel.de 3547 [ + - + + ]: 2312 : dlist_foreach_modify(iter, &MySerializableXact->inConflicts)
3548 : : {
3549 : 769 : RWConflict conflict =
331 tgl@sss.pgh.pa.us 3550 : 769 : dlist_container(RWConflictData, inLink, iter.cur);
3551 : :
4815 heikki.linnakangas@i 3552 [ + + ]: 769 : if (!isCommit
3553 [ + + ]: 599 : || SxactIsCommitted(conflict->sxactOut)
3554 [ + + ]: 415 : || SxactIsReadOnly(conflict->sxactOut))
3555 : 434 : ReleaseRWConflict(conflict);
3556 : : }
3557 : :
3558 [ + + ]: 1543 : if (!topLevelIsDeclaredReadOnly)
3559 : : {
3560 : : /*
3561 : : * Remove ourselves from the list of possible conflicts for concurrent
3562 : : * READ ONLY transactions, flagging them as unsafe if we have a
3563 : : * conflict out. If any are waiting DEFERRABLE transactions, wake them
3564 : : * up if they are known safe or known unsafe.
3565 : : */
451 andres@anarazel.de 3566 [ + - + + ]: 1525 : dlist_foreach_modify(iter, &MySerializableXact->possibleUnsafeConflicts)
3567 : : {
3568 : 90 : RWConflict possibleUnsafeConflict =
331 tgl@sss.pgh.pa.us 3569 : 90 : dlist_container(RWConflictData, outLink, iter.cur);
3570 : :
4815 heikki.linnakangas@i 3571 : 90 : roXact = possibleUnsafeConflict->sxactIn;
3572 [ - + ]: 90 : Assert(MySerializableXact == possibleUnsafeConflict->sxactOut);
3573 [ - + ]: 90 : Assert(SxactIsReadOnly(roXact));
3574 : :
3575 : : /* Mark conflicted if necessary. */
3576 [ + + ]: 90 : if (isCommit
4692 3577 [ + + ]: 88 : && MyXactDidWrite
4815 3578 [ + + ]: 83 : && SxactHasConflictOut(MySerializableXact)
3579 : 13 : && (MySerializableXact->SeqNo.earliestOutConflictCommit
3580 [ + + ]: 13 : <= roXact->SeqNo.lastCommitBeforeSnapshot))
3581 : : {
3582 : : /*
3583 : : * This releases possibleUnsafeConflict (as well as all other
3584 : : * possible conflicts for roXact)
3585 : : */
3586 : 3 : FlagSxactUnsafe(roXact);
3587 : : }
3588 : : else
3589 : : {
3590 : 87 : ReleaseRWConflict(possibleUnsafeConflict);
3591 : :
3592 : : /*
3593 : : * If we were the last possible conflict, flag it safe. The
3594 : : * transaction can now safely release its predicate locks (but
3595 : : * that transaction's backend has to do that itself).
3596 : : */
451 andres@anarazel.de 3597 [ + + ]: 87 : if (dlist_is_empty(&roXact->possibleUnsafeConflicts))
4815 heikki.linnakangas@i 3598 : 65 : roXact->flags |= SXACT_FLAG_RO_SAFE;
3599 : : }
3600 : :
3601 : : /*
3602 : : * Wake up the process for a waiting DEFERRABLE transaction if we
3603 : : * now know it's either safe or conflicted.
3604 : : */
3605 [ + + ]: 90 : if (SxactIsDeferrableWaiting(roXact) &&
3606 [ - + - - ]: 1 : (SxactIsROUnsafe(roXact) || SxactIsROSafe(roXact)))
850 tmunro@postgresql.or 3607 : 1 : ProcSendSignal(roXact->pgprocno);
3608 : : }
3609 : : }
3610 : :
3611 : : /*
3612 : : * Check whether it's time to clean up old transactions. This can only be
3613 : : * done when the last serializable transaction with the oldest xmin among
3614 : : * serializable transactions completes. We then find the "new oldest"
3615 : : * xmin and purge any transactions which finished before this transaction
3616 : : * was launched.
3617 : : *
3618 : : * For parallel queries in read-only transactions, it might run twice. We
3619 : : * only release the reference on the first call.
3620 : : */
4815 heikki.linnakangas@i 3621 : 1543 : needToClear = false;
405 tmunro@postgresql.or 3622 [ + + ]: 1543 : if ((partiallyReleasing ||
3623 [ + + ]: 1541 : !SxactIsPartiallyReleased(MySerializableXact)) &&
3624 [ + + ]: 1541 : TransactionIdEquals(MySerializableXact->xmin,
3625 : : PredXact->SxactGlobalXmin))
3626 : : {
4815 heikki.linnakangas@i 3627 [ - + ]: 1523 : Assert(PredXact->SxactGlobalXminCount > 0);
3628 [ + + ]: 1523 : if (--(PredXact->SxactGlobalXminCount) == 0)
3629 : : {
3630 : 854 : SetNewSxactGlobalXmin();
3631 : 854 : needToClear = true;
3632 : : }
3633 : : }
3634 : :
3635 : 1543 : LWLockRelease(SerializableXactHashLock);
3636 : :
3637 : 1543 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
3638 : :
3639 : : /* Add this to the list of transactions to check for later cleanup. */
3640 [ + + ]: 1543 : if (isCommit)
451 andres@anarazel.de 3641 : 1222 : dlist_push_tail(FinishedSerializableTransactions,
3642 : 1222 : &MySerializableXact->finishedLink);
3643 : :
3644 : : /*
3645 : : * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3646 : : * partially release it. That's necessary because other backends may have
3647 : : * a reference to it. The leader will release the SERIALIZABLEXACT itself
3648 : : * at the end of the transaction after workers have stopped running.
3649 : : */
4815 heikki.linnakangas@i 3650 [ + + ]: 1543 : if (!isCommit)
1857 tmunro@postgresql.or 3651 : 321 : ReleaseOneSerializableXact(MySerializableXact,
3652 [ + + + + ]: 321 : isReadOnlySafe && IsInParallelMode(),
1857 tmunro@postgresql.or 3653 :ECB (323) : false);
3654 : :
4815 heikki.linnakangas@i 3655 :CBC 1543 : LWLockRelease(SerializableFinishedListLock);
3656 : :
3657 [ + + ]: 1543 : if (needToClear)
3658 : 854 : ClearOldPredicateLocks();
3659 : :
1857 tmunro@postgresql.or 3660 : 1543 : ReleasePredicateLocksLocal();
3661 : : }
3662 : :
3663 : : static void
3664 : 5512 : ReleasePredicateLocksLocal(void)
3665 : : {
4815 heikki.linnakangas@i 3666 : 5512 : MySerializableXact = InvalidSerializableXact;
4692 3667 : 5512 : MyXactDidWrite = false;
3668 : :
3669 : : /* Delete per-transaction lock table */
4815 3670 [ + + ]: 5512 : if (LocalPredicateLockHash != NULL)
3671 : : {
3672 : 1542 : hash_destroy(LocalPredicateLockHash);
3673 : 1542 : LocalPredicateLockHash = NULL;
3674 : : }
3675 : 5512 : }
3676 : :
3677 : : /*
3678 : : * Clear old predicate locks, belonging to committed transactions that are no
3679 : : * longer interesting to any in-progress transaction.
3680 : : */
3681 : : static void
3682 : 854 : ClearOldPredicateLocks(void)
3683 : : {
3684 : : dlist_mutable_iter iter;
3685 : :
3686 : : /*
3687 : : * Loop through finished transactions. They are in commit order, so we can
3688 : : * stop as soon as we find one that's still interesting.
3689 : : */
3690 : 854 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
3691 : 854 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
451 andres@anarazel.de 3692 [ + - + + ]: 2084 : dlist_foreach_modify(iter, FinishedSerializableTransactions)
3693 : : {
3694 : 1239 : SERIALIZABLEXACT *finishedSxact =
331 tgl@sss.pgh.pa.us 3695 : 1239 : dlist_container(SERIALIZABLEXACT, finishedLink, iter.cur);
3696 : :
4815 heikki.linnakangas@i 3697 [ + + ]: 1239 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)
3698 [ + + ]: 28 : || TransactionIdPrecedesOrEquals(finishedSxact->finishedBefore,
3699 : 28 : PredXact->SxactGlobalXmin))
3700 : : {
3701 : : /*
3702 : : * This transaction committed before any in-progress transaction
3703 : : * took its snapshot. It's no longer interesting.
3704 : : */
3705 : 1222 : LWLockRelease(SerializableXactHashLock);
451 andres@anarazel.de 3706 : 1222 : dlist_delete_thoroughly(&finishedSxact->finishedLink);
4815 heikki.linnakangas@i 3707 : 1222 : ReleaseOneSerializableXact(finishedSxact, false, false);
3708 : 1222 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
3709 : : }
3710 [ + - ]: 17 : else if (finishedSxact->commitSeqNo > PredXact->HavePartialClearedThrough
2489 tgl@sss.pgh.pa.us 3711 [ + + ]: 17 : && finishedSxact->commitSeqNo <= PredXact->CanPartialClearThrough)
3712 : : {
3713 : : /*
3714 : : * Any active transactions that took their snapshot before this
3715 : : * transaction committed are read-only, so we can clear part of
3716 : : * its state.
3717 : : */
4815 heikki.linnakangas@i 3718 : 8 : LWLockRelease(SerializableXactHashLock);
3719 : :
4470 3720 [ - + ]: 8 : if (SxactIsReadOnly(finishedSxact))
3721 : : {
3722 : : /* A read-only transaction can be removed entirely */
451 andres@anarazel.de 3723 :UBC 0 : dlist_delete_thoroughly(&(finishedSxact->finishedLink));
4470 heikki.linnakangas@i 3724 : 0 : ReleaseOneSerializableXact(finishedSxact, false, false);
3725 : : }
3726 : : else
3727 : : {
3728 : : /*
3729 : : * A read-write transaction can only be partially cleared. We
3730 : : * need to keep the SERIALIZABLEXACT but can release the
3731 : : * SIREAD locks and conflicts in.
3732 : : */
4470 heikki.linnakangas@i 3733 :CBC 8 : ReleaseOneSerializableXact(finishedSxact, true, false);
3734 : : }
3735 : :
4815 3736 : 8 : PredXact->HavePartialClearedThrough = finishedSxact->commitSeqNo;
3737 : 8 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
3738 : : }
3739 : : else
3740 : : {
3741 : : /* Still interesting. */
3742 : : break;
3743 : : }
3744 : : }
3745 : 854 : LWLockRelease(SerializableXactHashLock);
3746 : :
3747 : : /*
3748 : : * Loop through predicate locks on dummy transaction for summarized data.
3749 : : */
1430 tgl@sss.pgh.pa.us 3750 : 854 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
451 andres@anarazel.de 3751 [ + - - + ]: 854 : dlist_foreach_modify(iter, &OldCommittedSxact->predicateLocks)
3752 : : {
451 andres@anarazel.de 3753 :UBC 0 : PREDICATELOCK *predlock =
331 tgl@sss.pgh.pa.us 3754 : 0 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
3755 : : bool canDoPartialCleanup;
3756 : :
4815 heikki.linnakangas@i 3757 : 0 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4752 3758 [ # # ]: 0 : Assert(predlock->commitSeqNo != 0);
3759 [ # # ]: 0 : Assert(predlock->commitSeqNo != InvalidSerCommitSeqNo);
4815 3760 : 0 : canDoPartialCleanup = (predlock->commitSeqNo <= PredXact->CanPartialClearThrough);
3761 : 0 : LWLockRelease(SerializableXactHashLock);
3762 : :
3763 : : /*
3764 : : * If this lock originally belonged to an old enough transaction, we
3765 : : * can release it.
3766 : : */
3767 [ # # ]: 0 : if (canDoPartialCleanup)
3768 : : {
3769 : : PREDICATELOCKTAG tag;
3770 : : PREDICATELOCKTARGET *target;
3771 : : PREDICATELOCKTARGETTAG targettag;
3772 : : uint32 targettaghash;
3773 : : LWLock *partitionLock;
3774 : :
3775 : 0 : tag = predlock->tag;
3776 : 0 : target = tag.myTarget;
3777 : 0 : targettag = target->tag;
3778 : 0 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
3779 : 0 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
3780 : :
3781 : 0 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3782 : :
451 andres@anarazel.de 3783 : 0 : dlist_delete(&(predlock->targetLink));
3784 : 0 : dlist_delete(&(predlock->xactLink));
3785 : :
4815 heikki.linnakangas@i 3786 : 0 : hash_search_with_hash_value(PredicateLockHash, &tag,
2489 tgl@sss.pgh.pa.us 3787 : 0 : PredicateLockHashCodeFromTargetHashCode(&tag,
3788 : : targettaghash),
3789 : : HASH_REMOVE, NULL);
4815 heikki.linnakangas@i 3790 : 0 : RemoveTargetIfNoLongerUsed(target, targettaghash);
3791 : :
3792 : 0 : LWLockRelease(partitionLock);
3793 : : }
3794 : : }
3795 : :
1430 tgl@sss.pgh.pa.us 3796 :CBC 854 : LWLockRelease(SerializablePredicateListLock);
4815 heikki.linnakangas@i 3797 : 854 : LWLockRelease(SerializableFinishedListLock);
3798 : 854 : }
3799 : :
3800 : : /*
3801 : : * This is the normal way to delete anything from any of the predicate
3802 : : * locking hash tables. Given a transaction which we know can be deleted:
3803 : : * delete all predicate locks held by that transaction and any predicate
3804 : : * lock targets which are now unreferenced by a lock; delete all conflicts
3805 : : * for the transaction; delete all xid values for the transaction; then
3806 : : * delete the transaction.
3807 : : *
3808 : : * When the partial flag is set, we can release all predicate locks and
3809 : : * in-conflict information -- we've established that there are no longer
3810 : : * any overlapping read write transactions for which this transaction could
3811 : : * matter -- but keep the transaction entry itself and any outConflicts.
3812 : : *
3813 : : * When the summarize flag is set, we've run short of room for sxact data
3814 : : * and must summarize to the SLRU. Predicate locks are transferred to a
3815 : : * dummy "old" transaction, with duplicate locks on a single target
3816 : : * collapsing to a single lock with the "latest" commitSeqNo from among
3817 : : * the conflicting locks..
3818 : : */
3819 : : static void
3820 : 1551 : ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
3821 : : bool summarize)
3822 : : {
3823 : : SERIALIZABLEXIDTAG sxidtag;
3824 : : dlist_mutable_iter iter;
3825 : :
3826 [ - + ]: 1551 : Assert(sxact != NULL);
4681 3827 [ + + - + ]: 1551 : Assert(SxactIsRolledBack(sxact) || SxactIsCommitted(sxact));
4470 3828 [ + + - + ]: 1551 : Assert(partial || !SxactIsOnFinishedList(sxact));
4815 3829 [ - + ]: 1551 : Assert(LWLockHeldByMe(SerializableFinishedListLock));
3830 : :
3831 : : /*
3832 : : * First release all the predicate locks held by this xact (or transfer
3833 : : * them to OldCommittedSxact if summarize is true)
3834 : : */
1430 tgl@sss.pgh.pa.us 3835 : 1551 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
1857 tmunro@postgresql.or 3836 [ + + ]: 1551 : if (IsInParallelMode())
1430 tgl@sss.pgh.pa.us 3837 : 3 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
451 andres@anarazel.de 3838 [ + - + + ]: 4384 : dlist_foreach_modify(iter, &sxact->predicateLocks)
3839 : : {
3840 : 2833 : PREDICATELOCK *predlock =
331 tgl@sss.pgh.pa.us 3841 : 2833 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
3842 : : PREDICATELOCKTAG tag;
3843 : : PREDICATELOCKTARGET *target;
3844 : : PREDICATELOCKTARGETTAG targettag;
3845 : : uint32 targettaghash;
3846 : : LWLock *partitionLock;
3847 : :
4815 heikki.linnakangas@i 3848 : 2833 : tag = predlock->tag;
3849 : 2833 : target = tag.myTarget;
3850 : 2833 : targettag = target->tag;
3851 : 2833 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
3852 : 2833 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
3853 : :
3854 : 2833 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3855 : :
451 andres@anarazel.de 3856 : 2833 : dlist_delete(&predlock->targetLink);
3857 : :
4815 heikki.linnakangas@i 3858 : 2833 : hash_search_with_hash_value(PredicateLockHash, &tag,
2489 tgl@sss.pgh.pa.us 3859 : 2833 : PredicateLockHashCodeFromTargetHashCode(&tag,
3860 : : targettaghash),
3861 : : HASH_REMOVE, NULL);
4815 heikki.linnakangas@i 3862 [ - + ]: 2833 : if (summarize)
3863 : : {
3864 : : bool found;
3865 : :
3866 : : /* Fold into dummy transaction list. */
4815 heikki.linnakangas@i 3867 :UBC 0 : tag.myXact = OldCommittedSxact;
3868 : 0 : predlock = hash_search_with_hash_value(PredicateLockHash, &tag,
2489 tgl@sss.pgh.pa.us 3869 : 0 : PredicateLockHashCodeFromTargetHashCode(&tag,
3870 : : targettaghash),
3871 : : HASH_ENTER_NULL, &found);
4815 heikki.linnakangas@i 3872 [ # # ]: 0 : if (!predlock)
3873 [ # # ]: 0 : ereport(ERROR,
3874 : : (errcode(ERRCODE_OUT_OF_MEMORY),
3875 : : errmsg("out of shared memory"),
3876 : : errhint("You might need to increase %s.", "max_pred_locks_per_transaction")));
3877 [ # # ]: 0 : if (found)
3878 : : {
4752 3879 [ # # ]: 0 : Assert(predlock->commitSeqNo != 0);
3880 [ # # ]: 0 : Assert(predlock->commitSeqNo != InvalidSerCommitSeqNo);
4815 3881 [ # # ]: 0 : if (predlock->commitSeqNo < sxact->commitSeqNo)
3882 : 0 : predlock->commitSeqNo = sxact->commitSeqNo;
3883 : : }
3884 : : else
3885 : : {
451 andres@anarazel.de 3886 : 0 : dlist_push_tail(&target->predicateLocks,
3887 : : &predlock->targetLink);
3888 : 0 : dlist_push_tail(&OldCommittedSxact->predicateLocks,
3889 : : &predlock->xactLink);
4815 heikki.linnakangas@i 3890 : 0 : predlock->commitSeqNo = sxact->commitSeqNo;
3891 : : }
3892 : : }
3893 : : else
4815 heikki.linnakangas@i 3894 :CBC 2833 : RemoveTargetIfNoLongerUsed(target, targettaghash);
3895 : :
3896 : 2833 : LWLockRelease(partitionLock);
3897 : : }
3898 : :
3899 : : /*
3900 : : * Rather than retail removal, just re-init the head after we've run
3901 : : * through the list.
3902 : : */
451 andres@anarazel.de 3903 : 1551 : dlist_init(&sxact->predicateLocks);
3904 : :
1857 tmunro@postgresql.or 3905 [ + + ]: 1551 : if (IsInParallelMode())
1430 tgl@sss.pgh.pa.us 3906 : 3 : LWLockRelease(&sxact->perXactPredicateListLock);
3907 : 1551 : LWLockRelease(SerializablePredicateListLock);
3908 : :
4815 heikki.linnakangas@i 3909 : 1551 : sxidtag.xid = sxact->topXid;
3910 : 1551 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
3911 : :
3912 : : /* Release all outConflicts (unless 'partial' is true) */
3913 [ + + ]: 1551 : if (!partial)
3914 : : {
451 andres@anarazel.de 3915 [ + - - + ]: 1541 : dlist_foreach_modify(iter, &sxact->outConflicts)
3916 : : {
451 andres@anarazel.de 3917 :UBC 0 : RWConflict conflict =
331 tgl@sss.pgh.pa.us 3918 : 0 : dlist_container(RWConflictData, outLink, iter.cur);
3919 : :
4815 heikki.linnakangas@i 3920 [ # # ]: 0 : if (summarize)
3921 : 0 : conflict->sxactIn->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
3922 : 0 : ReleaseRWConflict(conflict);
3923 : : }
3924 : : }
3925 : :
3926 : : /* Release all inConflicts. */
451 andres@anarazel.de 3927 [ + - - + ]:CBC 1551 : dlist_foreach_modify(iter, &sxact->inConflicts)
3928 : : {
451 andres@anarazel.de 3929 :UBC 0 : RWConflict conflict =
331 tgl@sss.pgh.pa.us 3930 : 0 : dlist_container(RWConflictData, inLink, iter.cur);
3931 : :
4815 heikki.linnakangas@i 3932 [ # # ]: 0 : if (summarize)
3933 : 0 : conflict->sxactOut->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
3934 : 0 : ReleaseRWConflict(conflict);
3935 : : }
3936 : :
3937 : : /* Finally, get rid of the xid and the record of the transaction itself. */
4815 heikki.linnakangas@i 3938 [ + + ]:CBC 1551 : if (!partial)
3939 : : {
3940 [ + + ]: 1541 : if (sxidtag.xid != InvalidTransactionId)
3941 : 1275 : hash_search(SerializableXidHash, &sxidtag, HASH_REMOVE, NULL);
3942 : 1541 : ReleasePredXact(sxact);
3943 : : }
3944 : :
3945 : 1551 : LWLockRelease(SerializableXactHashLock);
3946 : 1551 : }
3947 : :
3948 : : /*
3949 : : * Tests whether the given top level transaction is concurrent with
3950 : : * (overlaps) our current transaction.
3951 : : *
3952 : : * We need to identify the top level transaction for SSI, anyway, so pass
3953 : : * that to this function to save the overhead of checking the snapshot's
3954 : : * subxip array.
3955 : : */
3956 : : static bool
3957 : 532 : XidIsConcurrent(TransactionId xid)
3958 : : {
3959 : : Snapshot snap;
3960 : :
3961 [ - + ]: 532 : Assert(TransactionIdIsValid(xid));
3962 [ - + ]: 532 : Assert(!TransactionIdEquals(xid, GetTopTransactionIdIfAny()));
3963 : :
3964 : 532 : snap = GetTransactionSnapshot();
3965 : :
3966 [ - + ]: 532 : if (TransactionIdPrecedes(xid, snap->xmin))
4815 heikki.linnakangas@i 3967 :UBC 0 : return false;
3968 : :
4815 heikki.linnakangas@i 3969 [ + + ]:CBC 532 : if (TransactionIdFollowsOrEquals(xid, snap->xmax))
3970 : 520 : return true;
3971 : :
570 michael@paquier.xyz 3972 : 12 : return pg_lfind32(xid, snap->xip, snap->xcnt);
3973 : : }
3974 : :
3975 : : bool
1538 tmunro@postgresql.or 3976 : 30137219 : CheckForSerializableConflictOutNeeded(Relation relation, Snapshot snapshot)
3977 : : {
3978 [ + + ]: 30137219 : if (!SerializationNeededForRead(relation, snapshot))
3979 : 30111244 : return false;
3980 : :
3981 : : /* Check if someone else has already decided that we need to die */
3982 [ - + ]: 25975 : if (SxactIsDoomed(MySerializableXact))
3983 : : {
1538 tmunro@postgresql.or 3984 [ # # ]:UBC 0 : ereport(ERROR,
3985 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
3986 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
3987 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
3988 : : errhint("The transaction might succeed if retried.")));
3989 : : }
3990 : :
1538 tmunro@postgresql.or 3991 :CBC 25975 : return true;
3992 : : }
3993 : :
3994 : : /*
3995 : : * CheckForSerializableConflictOut
3996 : : * A table AM is reading a tuple that has been modified. If it determines
3997 : : * that the tuple version it is reading is not visible to us, it should
3998 : : * pass in the top level xid of the transaction that created it.
3999 : : * Otherwise, if it determines that it is visible to us but it has been
4000 : : * deleted or there is a newer version available due to an update, it
4001 : : * should pass in the top level xid of the modifying transaction.
4002 : : *
4003 : : * This function will check for overlap with our own transaction. If the given
4004 : : * xid is also serializable and the transactions overlap (i.e., they cannot see
4005 : : * each other's writes), then we have a conflict out.
4006 : : */
4007 : : void
4008 : 567 : CheckForSerializableConflictOut(Relation relation, TransactionId xid, Snapshot snapshot)
4009 : : {
4010 : : SERIALIZABLEXIDTAG sxidtag;
4011 : : SERIALIZABLEXID *sxid;
4012 : : SERIALIZABLEXACT *sxact;
4013 : :
4687 heikki.linnakangas@i 4014 [ - + ]: 567 : if (!SerializationNeededForRead(relation, snapshot))
4815 heikki.linnakangas@i 4015 :UBC 0 : return;
4016 : :
4017 : : /* Check if someone else has already decided that we need to die */
4687 heikki.linnakangas@i 4018 [ - + ]:CBC 567 : if (SxactIsDoomed(MySerializableXact))
4019 : : {
4815 heikki.linnakangas@i 4020 [ # # ]:UBC 0 : ereport(ERROR,
4021 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4022 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4023 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4024 : : errhint("The transaction might succeed if retried.")));
4025 : : }
4815 heikki.linnakangas@i 4026 [ - + ]:CBC 567 : Assert(TransactionIdIsValid(xid));
4027 : :
4028 [ - + ]: 567 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
4815 heikki.linnakangas@i 4029 :UBC 0 : return;
4030 : :
4031 : : /*
4032 : : * Find sxact or summarized info for the top level xid.
4033 : : */
4815 heikki.linnakangas@i 4034 :CBC 567 : sxidtag.xid = xid;
4035 : 567 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4036 : : sxid = (SERIALIZABLEXID *)
4037 : 567 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
4038 [ + + ]: 567 : if (!sxid)
4039 : : {
4040 : : /*
4041 : : * Transaction not found in "normal" SSI structures. Check whether it
4042 : : * got pushed out to SLRU storage for "old committed" transactions.
4043 : : */
4044 : : SerCommitSeqNo conflictCommitSeqNo;
4045 : :
1430 tgl@sss.pgh.pa.us 4046 : 25 : conflictCommitSeqNo = SerialGetMinConflictCommitSeqNo(xid);
4815 heikki.linnakangas@i 4047 [ - + ]: 25 : if (conflictCommitSeqNo != 0)
4048 : : {
4815 heikki.linnakangas@i 4049 [ # # ]:UBC 0 : if (conflictCommitSeqNo != InvalidSerCommitSeqNo
4050 [ # # ]: 0 : && (!SxactIsReadOnly(MySerializableXact)
4051 : 0 : || conflictCommitSeqNo
4052 [ # # ]: 0 : <= MySerializableXact->SeqNo.lastCommitBeforeSnapshot))
4053 [ # # ]: 0 : ereport(ERROR,
4054 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4055 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4056 : : errdetail_internal("Reason code: Canceled on conflict out to old pivot %u.", xid),
4057 : : errhint("The transaction might succeed if retried.")));
4058 : :
4059 [ # # ]: 0 : if (SxactHasSummaryConflictIn(MySerializableXact)
451 andres@anarazel.de 4060 [ # # ]: 0 : || !dlist_is_empty(&MySerializableXact->inConflicts))
4815 heikki.linnakangas@i 4061 [ # # ]: 0 : ereport(ERROR,
4062 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4063 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4064 : : errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4065 : : errhint("The transaction might succeed if retried.")));
4066 : :
4067 : 0 : MySerializableXact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4068 : : }
4069 : :
4070 : : /* It's not serializable or otherwise not important. */
4815 heikki.linnakangas@i 4071 :CBC 25 : LWLockRelease(SerializableXactHashLock);
4072 : 25 : return;
4073 : : }
4074 : 542 : sxact = sxid->myXact;
4075 [ - + ]: 542 : Assert(TransactionIdEquals(sxact->topXid, xid));
4687 4076 [ + - + + ]: 542 : if (sxact == MySerializableXact || SxactIsDoomed(sxact))
4077 : : {
4078 : : /* Can't conflict with ourself or a transaction that will roll back. */
4815 4079 : 4 : LWLockRelease(SerializableXactHashLock);
4080 : 4 : return;
4081 : : }
4082 : :
4083 : : /*
4084 : : * We have a conflict out to a transaction which has a conflict out to a
4085 : : * summarized transaction. That summarized transaction must have
4086 : : * committed first, and we can't tell when it committed in relation to our
4087 : : * snapshot acquisition, so something needs to be canceled.
4088 : : */
4089 [ - + ]: 538 : if (SxactHasSummaryConflictOut(sxact))
4090 : : {
4815 heikki.linnakangas@i 4091 [ # # ]:UBC 0 : if (!SxactIsPrepared(sxact))
4092 : : {
4687 4093 : 0 : sxact->flags |= SXACT_FLAG_DOOMED;
4815 4094 : 0 : LWLockRelease(SerializableXactHashLock);
4095 : 0 : return;
4096 : : }
4097 : : else
4098 : : {
4099 : 0 : LWLockRelease(SerializableXactHashLock);
4100 [ # # ]: 0 : ereport(ERROR,
4101 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4102 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4103 : : errdetail_internal("Reason code: Canceled on conflict out to old pivot."),
4104 : : errhint("The transaction might succeed if retried.")));
4105 : : }
4106 : : }
4107 : :
4108 : : /*
4109 : : * If this is a read-only transaction and the writing transaction has
4110 : : * committed, and it doesn't have a rw-conflict to a transaction which
4111 : : * committed before it, no conflict.
4112 : : */
4815 heikki.linnakangas@i 4113 [ + + ]:CBC 538 : if (SxactIsReadOnly(MySerializableXact)
4114 [ + + ]: 119 : && SxactIsCommitted(sxact)
4115 [ + - ]: 8 : && !SxactHasSummaryConflictOut(sxact)
4116 [ + + ]: 8 : && (!SxactHasConflictOut(sxact)
4117 [ - + ]: 2 : || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4118 : : {
4119 : : /* Read-only transaction will appear to run first. No conflict. */
4120 : 6 : LWLockRelease(SerializableXactHashLock);
4121 : 6 : return;
4122 : : }
4123 : :
4124 [ - + ]: 532 : if (!XidIsConcurrent(xid))
4125 : : {
4126 : : /* This write was already in our snapshot; no conflict. */
4815 heikki.linnakangas@i 4127 :UBC 0 : LWLockRelease(SerializableXactHashLock);
4128 : 0 : return;
4129 : : }
4130 : :
4692 heikki.linnakangas@i 4131 [ + + ]:CBC 532 : if (RWConflictExists(MySerializableXact, sxact))
4132 : : {
4133 : : /* We don't want duplicate conflict records in the list. */
4815 4134 : 169 : LWLockRelease(SerializableXactHashLock);
4135 : 169 : return;
4136 : : }
4137 : :
4138 : : /*
4139 : : * Flag the conflict. But first, if this conflict creates a dangerous
4140 : : * structure, ereport an error.
4141 : : */
4692 4142 : 363 : FlagRWConflict(MySerializableXact, sxact);
4815 4143 : 350 : LWLockRelease(SerializableXactHashLock);
4144 : : }
4145 : :
4146 : : /*
4147 : : * Check a particular target for rw-dependency conflict in. A subroutine of
4148 : : * CheckForSerializableConflictIn().
4149 : : */
4150 : : static void
4151 : 7496 : CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
4152 : : {
4153 : : uint32 targettaghash;
4154 : : LWLock *partitionLock;
4155 : : PREDICATELOCKTARGET *target;
4714 rhaas@postgresql.org 4156 : 7496 : PREDICATELOCK *mypredlock = NULL;
4157 : : PREDICATELOCKTAG mypredlocktag;
4158 : : dlist_mutable_iter iter;
4159 : :
4815 heikki.linnakangas@i 4160 [ - + ]: 7496 : Assert(MySerializableXact != InvalidSerializableXact);
4161 : :
4162 : : /*
4163 : : * The same hash and LW lock apply to the lock target and the lock itself.
4164 : : */
4165 : 7496 : targettaghash = PredicateLockTargetTagHashCode(targettag);
4166 : 7496 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
4167 : 7496 : LWLockAcquire(partitionLock, LW_SHARED);
4168 : : target = (PREDICATELOCKTARGET *)
4169 : 7496 : hash_search_with_hash_value(PredicateLockTargetHash,
4170 : : targettag, targettaghash,
4171 : : HASH_FIND, NULL);
4172 [ + + ]: 7496 : if (!target)
4173 : : {
4174 : : /* Nothing has this target locked; we're done here. */
4175 : 5625 : LWLockRelease(partitionLock);
4793 4176 : 5625 : return;
4177 : : }
4178 : :
4179 : : /*
4180 : : * Each lock for an overlapping transaction represents a conflict: a
4181 : : * rw-dependency in to this transaction.
4182 : : */
4815 4183 : 1871 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4184 : :
451 andres@anarazel.de 4185 [ + - + + ]: 4216 : dlist_foreach_modify(iter, &target->predicateLocks)
4186 : : {
4187 : 2412 : PREDICATELOCK *predlock =
331 tgl@sss.pgh.pa.us 4188 : 2412 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
451 andres@anarazel.de 4189 : 2412 : SERIALIZABLEXACT *sxact = predlock->tag.myXact;
4190 : :
4815 heikki.linnakangas@i 4191 [ + + ]: 2412 : if (sxact == MySerializableXact)
4192 : : {
4193 : : /*
4194 : : * If we're getting a write lock on a tuple, we don't need a
4195 : : * predicate (SIREAD) lock on the same tuple. We can safely remove
4196 : : * our SIREAD lock, but we'll defer doing so until after the loop
4197 : : * because that requires upgrading to an exclusive partition lock.
4198 : : *
4199 : : * We can't use this optimization within a subtransaction because
4200 : : * the subtransaction could roll back, and we would be left
4201 : : * without any lock at the top level.
4202 : : */
4755 rhaas@postgresql.org 4203 [ + - ]: 1564 : if (!IsSubTransaction()
4204 [ + + ]: 1564 : && GET_PREDICATELOCKTARGETTAG_OFFSET(*targettag))
4205 : : {
4714 4206 : 388 : mypredlock = predlock;
4207 : 388 : mypredlocktag = predlock->tag;
4208 : : }
4209 : : }
4687 heikki.linnakangas@i 4210 [ + - ]: 848 : else if (!SxactIsDoomed(sxact)
4815 4211 [ + + ]: 848 : && (!SxactIsCommitted(sxact)
4212 [ + + ]: 83 : || TransactionIdPrecedes(GetTransactionSnapshot()->xmin,
4213 : : sxact->finishedBefore))
4692 4214 [ + + ]: 839 : && !RWConflictExists(sxact, MySerializableXact))
4215 : : {
4815 4216 : 497 : LWLockRelease(SerializableXactHashLock);
4217 : 497 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4218 : :
4219 : : /*
4220 : : * Re-check after getting exclusive lock because the other
4221 : : * transaction may have flagged a conflict.
4222 : : */
4687 4223 [ + - ]: 497 : if (!SxactIsDoomed(sxact)
4758 rhaas@postgresql.org 4224 [ + + ]: 497 : && (!SxactIsCommitted(sxact)
4225 [ + - ]: 74 : || TransactionIdPrecedes(GetTransactionSnapshot()->xmin,
4226 : : sxact->finishedBefore))
4692 heikki.linnakangas@i 4227 [ + - ]: 497 : && !RWConflictExists(sxact, MySerializableXact))
4228 : : {
4229 : 497 : FlagRWConflict(sxact, MySerializableXact);
4230 : : }
4231 : :
4815 4232 : 430 : LWLockRelease(SerializableXactHashLock);
4233 : 430 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4234 : : }
4235 : : }
4236 : 1804 : LWLockRelease(SerializableXactHashLock);
4237 : 1804 : LWLockRelease(partitionLock);
4238 : :
4239 : : /*
4240 : : * If we found one of our own SIREAD locks to remove, remove it now.
4241 : : *
4242 : : * At this point our transaction already has a RowExclusiveLock on the
4243 : : * relation, so we are OK to drop the predicate lock on the tuple, if
4244 : : * found, without fearing that another write against the tuple will occur
4245 : : * before the MVCC information makes it to the buffer.
4246 : : */
4714 rhaas@postgresql.org 4247 [ + + ]: 1804 : if (mypredlock != NULL)
4248 : : {
4249 : : uint32 predlockhashcode;
4250 : : PREDICATELOCK *rmpredlock;
4251 : :
1430 tgl@sss.pgh.pa.us 4252 : 381 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
1857 tmunro@postgresql.or 4253 [ - + ]: 381 : if (IsInParallelMode())
1430 tgl@sss.pgh.pa.us 4254 :UBC 0 : LWLockAcquire(&MySerializableXact->perXactPredicateListLock, LW_EXCLUSIVE);
4714 rhaas@postgresql.org 4255 :CBC 381 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4256 : 381 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4257 : :
4258 : : /*
4259 : : * Remove the predicate lock from shared memory, if it wasn't removed
4260 : : * while the locks were released. One way that could happen is from
4261 : : * autovacuum cleaning up an index.
4262 : : */
4263 : 381 : predlockhashcode = PredicateLockHashCodeFromTargetHashCode
4264 : : (&mypredlocktag, targettaghash);
4265 : : rmpredlock = (PREDICATELOCK *)
4266 : 381 : hash_search_with_hash_value(PredicateLockHash,
4267 : : &mypredlocktag,
4268 : : predlockhashcode,
4269 : : HASH_FIND, NULL);
4270 [ + - ]: 381 : if (rmpredlock != NULL)
4271 : : {
4272 [ - + ]: 381 : Assert(rmpredlock == mypredlock);
4273 : :
451 andres@anarazel.de 4274 : 381 : dlist_delete(&(mypredlock->targetLink));
4275 : 381 : dlist_delete(&(mypredlock->xactLink));
4276 : :
4277 : : rmpredlock = (PREDICATELOCK *)
4714 rhaas@postgresql.org 4278 : 381 : hash_search_with_hash_value(PredicateLockHash,
4279 : : &mypredlocktag,
4280 : : predlockhashcode,
4281 : : HASH_REMOVE, NULL);
4282 [ - + ]: 381 : Assert(rmpredlock == mypredlock);
4283 : :
4284 : 381 : RemoveTargetIfNoLongerUsed(target, targettaghash);
4285 : : }
4286 : :
4287 : 381 : LWLockRelease(SerializableXactHashLock);
4288 : 381 : LWLockRelease(partitionLock);
1857 tmunro@postgresql.or 4289 [ - + ]: 381 : if (IsInParallelMode())
1430 tgl@sss.pgh.pa.us 4290 :UBC 0 : LWLockRelease(&MySerializableXact->perXactPredicateListLock);
1430 tgl@sss.pgh.pa.us 4291 :CBC 381 : LWLockRelease(SerializablePredicateListLock);
4292 : :
4714 rhaas@postgresql.org 4293 [ + - ]: 381 : if (rmpredlock != NULL)
4294 : : {
4295 : : /*
4296 : : * Remove entry in local lock table if it exists. It's OK if it
4297 : : * doesn't exist; that means the lock was transferred to a new
4298 : : * target by a different backend.
4299 : : */
4300 : 381 : hash_search_with_hash_value(LocalPredicateLockHash,
4301 : : targettag, targettaghash,
4302 : : HASH_REMOVE, NULL);
4303 : :
4304 : 381 : DecrementParentLocks(targettag);
4305 : : }
4306 : : }
4307 : : }
4308 : :
4309 : : /*
4310 : : * CheckForSerializableConflictIn
4311 : : * We are writing the given tuple. If that indicates a rw-conflict
4312 : : * in from another serializable transaction, take appropriate action.
4313 : : *
4314 : : * Skip checking for any granularity for which a parameter is missing.
4315 : : *
4316 : : * A tuple update or delete is in conflict if we have a predicate lock
4317 : : * against the relation or page in which the tuple exists, or against the
4318 : : * tuple itself.
4319 : : */
4320 : : void
1538 tmunro@postgresql.or 4321 : 15308295 : CheckForSerializableConflictIn(Relation relation, ItemPointer tid, BlockNumber blkno)
4322 : : {
4323 : : PREDICATELOCKTARGETTAG targettag;
4324 : :
4687 heikki.linnakangas@i 4325 [ + + ]: 15308295 : if (!SerializationNeededForWrite(relation))
4815 4326 : 15303852 : return;
4327 : :
4328 : : /* Check if someone else has already decided that we need to die */
4687 4329 [ + + ]: 4443 : if (SxactIsDoomed(MySerializableXact))
4815 4330 [ + - ]: 1 : ereport(ERROR,
4331 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4332 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4333 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4334 : : errhint("The transaction might succeed if retried.")));
4335 : :
4336 : : /*
4337 : : * We're doing a write which might cause rw-conflicts now or later.
4338 : : * Memorize that fact.
4339 : : */
4692 4340 : 4442 : MyXactDidWrite = true;
4341 : :
4342 : : /*
4343 : : * It is important that we check for locks from the finest granularity to
4344 : : * the coarsest granularity, so that granularity promotion doesn't cause
4345 : : * us to miss a lock. The new (coarser) lock will be acquired before the
4346 : : * old (finer) locks are released.
4347 : : *
4348 : : * It is not possible to take and hold a lock across the checks for all
4349 : : * granularities because each target could be in a separate partition.
4350 : : */
1538 tmunro@postgresql.or 4351 [ + + ]: 4442 : if (tid != NULL)
4352 : : {
4815 heikki.linnakangas@i 4353 : 643 : SET_PREDICATELOCKTARGETTAG_TUPLE(targettag,
4354 : : relation->rd_locator.dbOid,
4355 : : relation->rd_id,
4356 : : ItemPointerGetBlockNumber(tid),
4357 : : ItemPointerGetOffsetNumber(tid));
4358 : 643 : CheckTargetForConflictsIn(&targettag);
4359 : : }
4360 : :
1538 tmunro@postgresql.or 4361 [ + + ]: 4419 : if (blkno != InvalidBlockNumber)
4362 : : {
4815 heikki.linnakangas@i 4363 : 2464 : SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
4364 : : relation->rd_locator.dbOid,
4365 : : relation->rd_id,
4366 : : blkno);
4367 : 2464 : CheckTargetForConflictsIn(&targettag);
4368 : : }
4369 : :
4370 : 4389 : SET_PREDICATELOCKTARGETTAG_RELATION(targettag,
4371 : : relation->rd_locator.dbOid,
4372 : : relation->rd_id);
4373 : 4389 : CheckTargetForConflictsIn(&targettag);
4374 : : }
4375 : :
4376 : : /*
4377 : : * CheckTableForSerializableConflictIn
4378 : : * The entire table is going through a DDL-style logical mass delete
4379 : : * like TRUNCATE or DROP TABLE. If that causes a rw-conflict in from
4380 : : * another serializable transaction, take appropriate action.
4381 : : *
4382 : : * While these operations do not operate entirely within the bounds of
4383 : : * snapshot isolation, they can occur inside a serializable transaction, and
4384 : : * will logically occur after any reads which saw rows which were destroyed
4385 : : * by these operations, so we do what we can to serialize properly under
4386 : : * SSI.
4387 : : *
4388 : : * The relation passed in must be a heap relation. Any predicate lock of any
4389 : : * granularity on the heap will cause a rw-conflict in to this transaction.
4390 : : * Predicate locks on indexes do not matter because they only exist to guard
4391 : : * against conflicting inserts into the index, and this is a mass *delete*.
4392 : : * When a table is truncated or dropped, the index will also be truncated
4393 : : * or dropped, and we'll deal with locks on the index when that happens.
4394 : : *
4395 : : * Dropping or truncating a table also needs to drop any existing predicate
4396 : : * locks on heap tuples or pages, because they're about to go away. This
4397 : : * should be done before altering the predicate locks because the transaction
4398 : : * could be rolled back because of a conflict, in which case the lock changes
4399 : : * are not needed. (At the moment, we don't actually bother to drop the
4400 : : * existing locks on a dropped or truncated table at the moment. That might
4401 : : * lead to some false positives, but it doesn't seem worth the trouble.)
4402 : : */
4403 : : void
4680 4404 : 22335 : CheckTableForSerializableConflictIn(Relation relation)
4405 : : {
4406 : : HASH_SEQ_STATUS seqstat;
4407 : : PREDICATELOCKTARGET *target;
4408 : : Oid dbId;
4409 : : Oid heapId;
4410 : : int i;
4411 : :
4412 : : /*
4413 : : * Bail out quickly if there are no serializable transactions running.
4414 : : * It's safe to check this without taking locks because the caller is
4415 : : * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4416 : : * would matter here can be acquired while that is held.
4417 : : */
4694 4418 [ + + ]: 22335 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
4419 : 22314 : return;
4420 : :
4687 4421 [ + + ]: 152 : if (!SerializationNeededForWrite(relation))
4694 4422 : 131 : return;
4423 : :
4424 : : /*
4425 : : * We're doing a write which might cause rw-conflicts now or later.
4426 : : * Memorize that fact.
4427 : : */
4692 4428 : 21 : MyXactDidWrite = true;
4429 : :
4694 4430 [ - + ]: 21 : Assert(relation->rd_index == NULL); /* not an index relation */
4431 : :
648 rhaas@postgresql.org 4432 : 21 : dbId = relation->rd_locator.dbOid;
4694 heikki.linnakangas@i 4433 : 21 : heapId = relation->rd_id;
4434 : :
1430 tgl@sss.pgh.pa.us 4435 : 21 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
4694 heikki.linnakangas@i 4436 [ + + ]: 357 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
3730 rhaas@postgresql.org 4437 : 336 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED);
3088 kgrittn@postgresql.o 4438 : 21 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4439 : :
4440 : : /* Scan through target list */
4694 heikki.linnakangas@i 4441 : 21 : hash_seq_init(&seqstat, PredicateLockTargetHash);
4442 : :
4443 [ + + ]: 72 : while ((target = (PREDICATELOCKTARGET *) hash_seq_search(&seqstat)))
4444 : : {
4445 : : dlist_mutable_iter iter;
4446 : :
4447 : : /*
4448 : : * Check whether this is a target which needs attention.
4449 : : */
4450 [ + + ]: 51 : if (GET_PREDICATELOCKTARGETTAG_RELATION(target->tag) != heapId)
4451 : 42 : continue; /* wrong relation id */
4694 heikki.linnakangas@i 4452 [ - + ]:GBC 9 : if (GET_PREDICATELOCKTARGETTAG_DB(target->tag) != dbId)
4694 heikki.linnakangas@i 4453 :UBC 0 : continue; /* wrong database id */
4454 : :
4455 : : /*
4456 : : * Loop through locks for this target and flag conflicts.
4457 : : */
451 andres@anarazel.de 4458 [ + - + + ]:GBC 18 : dlist_foreach_modify(iter, &target->predicateLocks)
4459 : : {
4460 : 9 : PREDICATELOCK *predlock =
331 tgl@sss.pgh.pa.us 4461 : 9 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
4462 : :
4694 heikki.linnakangas@i 4463 [ - + ]: 9 : if (predlock->tag.myXact != MySerializableXact
2489 tgl@sss.pgh.pa.us 4464 [ # # ]:UBC 0 : && !RWConflictExists(predlock->tag.myXact, MySerializableXact))
4465 : : {
4692 heikki.linnakangas@i 4466 : 0 : FlagRWConflict(predlock->tag.myXact, MySerializableXact);
4467 : : }
4468 : : }
4469 : : }
4470 : :
4471 : : /* Release locks in reverse order */
4694 heikki.linnakangas@i 4472 :CBC 21 : LWLockRelease(SerializableXactHashLock);
4473 [ + + ]: 357 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
3730 rhaas@postgresql.org 4474 : 336 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
1430 tgl@sss.pgh.pa.us 4475 : 21 : LWLockRelease(SerializablePredicateListLock);
4476 : : }
4477 : :
4478 : :
4479 : : /*
4480 : : * Flag a rw-dependency between two serializable transactions.
4481 : : *
4482 : : * The caller is responsible for ensuring that we have a LW lock on
4483 : : * the transaction hash table.
4484 : : */
4485 : : static void
4815 heikki.linnakangas@i 4486 : 860 : FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
4487 : : {
4488 [ - + ]: 860 : Assert(reader != writer);
4489 : :
4490 : : /* First, see if this conflict causes failure. */
4491 : 860 : OnConflict_CheckForSerializationFailure(reader, writer);
4492 : :
4493 : : /* Actually do the conflict flagging. */
4494 [ - + ]: 780 : if (reader == OldCommittedSxact)
4815 heikki.linnakangas@i 4495 :UBC 0 : writer->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
4815 heikki.linnakangas@i 4496 [ - + ]:CBC 780 : else if (writer == OldCommittedSxact)
4815 heikki.linnakangas@i 4497 :UBC 0 : reader->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4498 : : else
4815 heikki.linnakangas@i 4499 :CBC 780 : SetRWConflict(reader, writer);
4500 : 780 : }
4501 : :
4502 : : /*----------------------------------------------------------------------------
4503 : : * We are about to add a RW-edge to the dependency graph - check that we don't
4504 : : * introduce a dangerous structure by doing so, and abort one of the
4505 : : * transactions if so.
4506 : : *
4507 : : * A serialization failure can only occur if there is a dangerous structure
4508 : : * in the dependency graph:
4509 : : *
4510 : : * Tin ------> Tpivot ------> Tout
4511 : : * rw rw
4512 : : *
4513 : : * Furthermore, Tout must commit first.
4514 : : *
4515 : : * One more optimization is that if Tin is declared READ ONLY (or commits
4516 : : * without writing), we can only have a problem if Tout committed before Tin
4517 : : * acquired its snapshot.
4518 : : *----------------------------------------------------------------------------
4519 : : */
4520 : : static void
4680 tgl@sss.pgh.pa.us 4521 : 860 : OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader,
4522 : : SERIALIZABLEXACT *writer)
4523 : : {
4524 : : bool failure;
4525 : :
4815 heikki.linnakangas@i 4526 [ - + ]: 860 : Assert(LWLockHeldByMe(SerializableXactHashLock));
4527 : :
4528 : 860 : failure = false;
4529 : :
4530 : : /*------------------------------------------------------------------------
4531 : : * Check for already-committed writer with rw-conflict out flagged
4532 : : * (conflict-flag on W means that T2 committed before W):
4533 : : *
4534 : : * R ------> W ------> T2
4535 : : * rw rw
4536 : : *
4537 : : * That is a dangerous structure, so we must abort. (Since the writer
4538 : : * has already committed, we must be the reader)
4539 : : *------------------------------------------------------------------------
4540 : : */
4541 [ + + ]: 860 : if (SxactIsCommitted(writer)
2489 tgl@sss.pgh.pa.us 4542 [ + + - + ]: 18 : && (SxactHasConflictOut(writer) || SxactHasSummaryConflictOut(writer)))
4815 heikki.linnakangas@i 4543 : 2 : failure = true;
4544 : :
4545 : : /*------------------------------------------------------------------------
4546 : : * Check whether the writer has become a pivot with an out-conflict
4547 : : * committed transaction (T2), and T2 committed first:
4548 : : *
4549 : : * R ------> W ------> T2
4550 : : * rw rw
4551 : : *
4552 : : * Because T2 must've committed first, there is no anomaly if:
4553 : : * - the reader committed before T2
4554 : : * - the writer committed before T2
4555 : : * - the reader is a READ ONLY transaction and the reader was concurrent
4556 : : * with T2 (= reader acquired its snapshot before T2 committed)
4557 : : *
4558 : : * We also handle the case that T2 is prepared but not yet committed
4559 : : * here. In that case T2 has already checked for conflicts, so if it
4560 : : * commits first, making the above conflict real, it's too late for it
4561 : : * to abort.
4562 : : *------------------------------------------------------------------------
4563 : : */
451 andres@anarazel.de 4564 [ + + - + ]: 860 : if (!failure && SxactHasSummaryConflictOut(writer))
451 andres@anarazel.de 4565 :UBC 0 : failure = true;
451 andres@anarazel.de 4566 [ + + ]:CBC 860 : else if (!failure)
4567 : : {
4568 : : dlist_iter iter;
4569 : :
4570 [ + - + + ]: 1071 : dlist_foreach(iter, &writer->outConflicts)
4571 : : {
4572 : 288 : RWConflict conflict =
331 tgl@sss.pgh.pa.us 4573 : 288 : dlist_container(RWConflictData, outLink, iter.cur);
4703 heikki.linnakangas@i 4574 : 288 : SERIALIZABLEXACT *t2 = conflict->sxactIn;
4575 : :
4665 4576 [ + + ]: 288 : if (SxactIsPrepared(t2)
4703 4577 [ + + ]: 81 : && (!SxactIsCommitted(reader)
4665 4578 [ + - ]: 65 : || t2->prepareSeqNo <= reader->commitSeqNo)
4703 4579 [ - + ]: 81 : && (!SxactIsCommitted(writer)
4665 heikki.linnakangas@i 4580 [ # # ]:UBC 0 : || t2->prepareSeqNo <= writer->commitSeqNo)
4703 heikki.linnakangas@i 4581 [ + + ]:CBC 81 : && (!SxactIsReadOnly(reader)
2489 tgl@sss.pgh.pa.us 4582 [ + + ]: 12 : || t2->prepareSeqNo <= reader->SeqNo.lastCommitBeforeSnapshot))
4583 : : {
4815 heikki.linnakangas@i 4584 : 75 : failure = true;
4585 : 75 : break;
4586 : : }
4587 : : }
4588 : : }
4589 : :
4590 : : /*------------------------------------------------------------------------
4591 : : * Check whether the reader has become a pivot with a writer
4592 : : * that's committed (or prepared):
4593 : : *
4594 : : * T0 ------> R ------> W
4595 : : * rw rw
4596 : : *
4597 : : * Because W must've committed first for an anomaly to occur, there is no
4598 : : * anomaly if:
4599 : : * - T0 committed before the writer
4600 : : * - T0 is READ ONLY, and overlaps the writer
4601 : : *------------------------------------------------------------------------
4602 : : */
4665 4603 [ + + + + : 860 : if (!failure && SxactIsPrepared(writer) && !SxactIsReadOnly(reader))
+ - ]
4604 : : {
4703 4605 [ - + ]: 18 : if (SxactHasSummaryConflictIn(reader))
4606 : : {
4815 heikki.linnakangas@i 4607 :UBC 0 : failure = true;
4608 : : }
4609 : : else
4610 : : {
4611 : : dlist_iter iter;
4612 : :
4613 : : /*
4614 : : * The unconstify is needed as we have no const version of
4615 : : * dlist_foreach().
4616 : : */
451 andres@anarazel.de 4617 [ + - + + ]:CBC 18 : dlist_foreach(iter, &unconstify(SERIALIZABLEXACT *, reader)->inConflicts)
4618 : : {
4619 : 11 : const RWConflict conflict =
331 tgl@sss.pgh.pa.us 4620 : 11 : dlist_container(RWConflictData, inLink, iter.cur);
451 andres@anarazel.de 4621 : 11 : const SERIALIZABLEXACT *t0 = conflict->sxactOut;
4622 : :
4623 [ + - ]: 11 : if (!SxactIsDoomed(t0)
4624 [ + - ]: 11 : && (!SxactIsCommitted(t0)
4625 [ + - ]: 11 : || t0->commitSeqNo >= writer->prepareSeqNo)
4626 [ - + ]: 11 : && (!SxactIsReadOnly(t0)
451 andres@anarazel.de 4627 [ # # ]:UBC 0 : || t0->SeqNo.lastCommitBeforeSnapshot >= writer->prepareSeqNo))
4628 : : {
451 andres@anarazel.de 4629 :CBC 11 : failure = true;
4630 : 11 : break;
4631 : : }
4632 : : }
4633 : : }
4634 : : }
4635 : :
4815 heikki.linnakangas@i 4636 [ + + ]: 860 : if (failure)
4637 : : {
4638 : : /*
4639 : : * We have to kill a transaction to avoid a possible anomaly from
4640 : : * occurring. If the writer is us, we can just ereport() to cause a
4641 : : * transaction abort. Otherwise we flag the writer for termination,
4642 : : * causing it to abort when it tries to commit. However, if the writer
4643 : : * is a prepared transaction, already prepared, we can't abort it
4644 : : * anymore, so we have to kill the reader instead.
4645 : : */
4646 [ + + ]: 88 : if (MySerializableXact == writer)
4647 : : {
4648 : 67 : LWLockRelease(SerializableXactHashLock);
4649 [ + - ]: 67 : ereport(ERROR,
4650 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4651 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4652 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during write."),
4653 : : errhint("The transaction might succeed if retried.")));
4654 : : }
4655 [ + + ]: 21 : else if (SxactIsPrepared(writer))
4656 : : {
4657 : 13 : LWLockRelease(SerializableXactHashLock);
4658 : :
4659 : : /* if we're not the writer, we have to be the reader */
4703 4660 [ - + ]: 13 : Assert(MySerializableXact == reader);
4815 4661 [ + - ]: 13 : ereport(ERROR,
4662 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4663 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4664 : : errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4665 : : errhint("The transaction might succeed if retried.")));
4666 : : }
4687 4667 : 8 : writer->flags |= SXACT_FLAG_DOOMED;
4668 : : }
4815 4669 : 780 : }
4670 : :
4671 : : /*
4672 : : * PreCommit_CheckForSerializationFailure
4673 : : * Check for dangerous structures in a serializable transaction
4674 : : * at commit.
4675 : : *
4676 : : * We're checking for a dangerous structure as each conflict is recorded.
4677 : : * The only way we could have a problem at commit is if this is the "out"
4678 : : * side of a pivot, and neither the "in" side nor the pivot has yet
4679 : : * committed.
4680 : : *
4681 : : * If a dangerous structure is found, the pivot (the near conflict) is
4682 : : * marked for death, because rolling back another transaction might mean
4683 : : * that we fail without ever making progress. This transaction is
4684 : : * committing writes, so letting it commit ensures progress. If we
4685 : : * canceled the far conflict, it might immediately fail again on retry.
4686 : : */
4687 : : void
4688 : 408981 : PreCommit_CheckForSerializationFailure(void)
4689 : : {
4690 : : dlist_iter near_iter;
4691 : :
4692 [ + + ]: 408981 : if (MySerializableXact == InvalidSerializableXact)
4693 : 407587 : return;
4694 : :
4695 [ - + ]: 1394 : Assert(IsolationIsSerializable());
4696 : :
4697 : 1394 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4698 : :
4699 : : /*
4700 : : * Check if someone else has already decided that we need to die. Since
4701 : : * we set our own DOOMED flag when partially releasing, ignore in that
4702 : : * case.
4703 : : */
405 tmunro@postgresql.or 4704 [ + + ]: 1394 : if (SxactIsDoomed(MySerializableXact) &&
4705 [ + + ]: 156 : !SxactIsPartiallyReleased(MySerializableXact))
4706 : : {
4815 heikki.linnakangas@i 4707 : 155 : LWLockRelease(SerializableXactHashLock);
4708 [ + - ]: 155 : ereport(ERROR,
4709 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4710 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4711 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4712 : : errhint("The transaction might succeed if retried.")));
4713 : : }
4714 : :
451 andres@anarazel.de 4715 [ + - + + ]: 1840 : dlist_foreach(near_iter, &MySerializableXact->inConflicts)
4716 : : {
4717 : 601 : RWConflict nearConflict =
331 tgl@sss.pgh.pa.us 4718 : 601 : dlist_container(RWConflictData, inLink, near_iter.cur);
4719 : :
4815 heikki.linnakangas@i 4720 [ + + ]: 601 : if (!SxactIsCommitted(nearConflict->sxactOut)
4687 4721 [ + - ]: 417 : && !SxactIsDoomed(nearConflict->sxactOut))
4722 : : {
4723 : : dlist_iter far_iter;
4724 : :
451 andres@anarazel.de 4725 [ + - + + ]: 447 : dlist_foreach(far_iter, &nearConflict->sxactOut->inConflicts)
4726 : : {
4727 : 178 : RWConflict farConflict =
331 tgl@sss.pgh.pa.us 4728 : 178 : dlist_container(RWConflictData, inLink, far_iter.cur);
4729 : :
4815 heikki.linnakangas@i 4730 [ + + ]: 178 : if (farConflict->sxactOut == MySerializableXact
4731 [ + + ]: 42 : || (!SxactIsCommitted(farConflict->sxactOut)
4732 [ + + ]: 24 : && !SxactIsReadOnly(farConflict->sxactOut)
4687 4733 [ + - ]: 12 : && !SxactIsDoomed(farConflict->sxactOut)))
4734 : : {
4735 : : /*
4736 : : * Normally, we kill the pivot transaction to make sure we
4737 : : * make progress if the failing transaction is retried.
4738 : : * However, we can't kill it if it's already prepared, so
4739 : : * in that case we commit suicide instead.
4740 : : */
4681 4741 [ - + ]: 148 : if (SxactIsPrepared(nearConflict->sxactOut))
4742 : : {
4681 heikki.linnakangas@i 4743 :UBC 0 : LWLockRelease(SerializableXactHashLock);
4744 [ # # ]: 0 : ereport(ERROR,
4745 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4746 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4747 : : errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4748 : : errhint("The transaction might succeed if retried.")));
4749 : : }
4687 heikki.linnakangas@i 4750 :CBC 148 : nearConflict->sxactOut->flags |= SXACT_FLAG_DOOMED;
4815 4751 : 148 : break;
4752 : : }
4753 : : }
4754 : : }
4755 : : }
4756 : :
4665 4757 : 1239 : MySerializableXact->prepareSeqNo = ++(PredXact->LastSxactCommitSeqNo);
4815 4758 : 1239 : MySerializableXact->flags |= SXACT_FLAG_PREPARED;
4759 : :
4760 : 1239 : LWLockRelease(SerializableXactHashLock);
4761 : : }
4762 : :
4763 : : /*------------------------------------------------------------------------*/
4764 : :
4765 : : /*
4766 : : * Two-phase commit support
4767 : : */
4768 : :
4769 : : /*
4770 : : * AtPrepare_Locks
4771 : : * Do the preparatory work for a PREPARE: make 2PC state file
4772 : : * records for all predicate locks currently held.
4773 : : */
4774 : : void
4775 : 391 : AtPrepare_PredicateLocks(void)
4776 : : {
4777 : : SERIALIZABLEXACT *sxact;
4778 : : TwoPhasePredicateRecord record;
4779 : : TwoPhasePredicateXactRecord *xactRecord;
4780 : : TwoPhasePredicateLockRecord *lockRecord;
4781 : : dlist_iter iter;
4782 : :
4692 4783 : 391 : sxact = MySerializableXact;
4815 4784 : 391 : xactRecord = &(record.data.xactRecord);
4785 : 391 : lockRecord = &(record.data.lockRecord);
4786 : :
4787 [ + + ]: 391 : if (MySerializableXact == InvalidSerializableXact)
4788 : 379 : return;
4789 : :
4790 : : /* Generate an xact record for our SERIALIZABLEXACT */
4791 : 12 : record.type = TWOPHASEPREDICATERECORD_XACT;
4792 : 12 : xactRecord->xmin = MySerializableXact->xmin;
4793 : 12 : xactRecord->flags = MySerializableXact->flags;
4794 : :
4795 : : /*
4796 : : * Note that we don't include the list of conflicts in our out in the
4797 : : * statefile, because new conflicts can be added even after the
4798 : : * transaction prepares. We'll just make a conservative assumption during
4799 : : * recovery instead.
4800 : : */
4801 : :
4802 : 12 : RegisterTwoPhaseRecord(TWOPHASE_RM_PREDICATELOCK_ID, 0,
4803 : : &record, sizeof(record));
4804 : :
4805 : : /*
4806 : : * Generate a lock record for each lock.
4807 : : *
4808 : : * To do this, we need to walk the predicate lock list in our sxact rather
4809 : : * than using the local predicate lock table because the latter is not
4810 : : * guaranteed to be accurate.
4811 : : */
1430 tgl@sss.pgh.pa.us 4812 : 12 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
4813 : :
4814 : : /*
4815 : : * No need to take sxact->perXactPredicateListLock in parallel mode
4816 : : * because there cannot be any parallel workers running while we are
4817 : : * preparing a transaction.
4818 : : */
1857 tmunro@postgresql.or 4819 [ + - - + ]: 12 : Assert(!IsParallelWorker() && !ParallelContextActive());
4820 : :
451 andres@anarazel.de 4821 [ + - + + ]: 22 : dlist_foreach(iter, &sxact->predicateLocks)
4822 : : {
4823 : 10 : PREDICATELOCK *predlock =
331 tgl@sss.pgh.pa.us 4824 : 10 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
4825 : :
4815 heikki.linnakangas@i 4826 : 10 : record.type = TWOPHASEPREDICATERECORD_LOCK;
4827 : 10 : lockRecord->target = predlock->tag.myTarget->tag;
4828 : :
4829 : 10 : RegisterTwoPhaseRecord(TWOPHASE_RM_PREDICATELOCK_ID, 0,
4830 : : &record, sizeof(record));
4831 : : }
4832 : :
1430 tgl@sss.pgh.pa.us 4833 : 12 : LWLockRelease(SerializablePredicateListLock);
4834 : : }
4835 : :
4836 : : /*
4837 : : * PostPrepare_Locks
4838 : : * Clean up after successful PREPARE. Unlike the non-predicate
4839 : : * lock manager, we do not need to transfer locks to a dummy
4840 : : * PGPROC because our SERIALIZABLEXACT will stay around
4841 : : * anyway. We only need to clean up our local state.
4842 : : */
4843 : : void
4815 heikki.linnakangas@i 4844 : 391 : PostPrepare_PredicateLocks(TransactionId xid)
4845 : : {
4846 [ + + ]: 391 : if (MySerializableXact == InvalidSerializableXact)
4847 : 379 : return;
4848 : :
4849 [ - + ]: 12 : Assert(SxactIsPrepared(MySerializableXact));
4850 : :
4851 : 12 : MySerializableXact->pid = 0;
42 heikki.linnakangas@i 4852 :GNC 12 : MySerializableXact->pgprocno = INVALID_PROC_NUMBER;
4853 : :
4815 heikki.linnakangas@i 4854 :CBC 12 : hash_destroy(LocalPredicateLockHash);
4855 : 12 : LocalPredicateLockHash = NULL;
4856 : :
4857 : 12 : MySerializableXact = InvalidSerializableXact;
4692 4858 : 12 : MyXactDidWrite = false;
4859 : : }
4860 : :
4861 : : /*
4862 : : * PredicateLockTwoPhaseFinish
4863 : : * Release a prepared transaction's predicate locks once it
4864 : : * commits or aborts.
4865 : : */
4866 : : void
4815 4867 : 395 : PredicateLockTwoPhaseFinish(TransactionId xid, bool isCommit)
4868 : : {
4869 : : SERIALIZABLEXID *sxid;
4870 : : SERIALIZABLEXIDTAG sxidtag;
4871 : :
4872 : 395 : sxidtag.xid = xid;
4873 : :
4874 : 395 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4875 : : sxid = (SERIALIZABLEXID *)
4876 : 395 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
4877 : 395 : LWLockRelease(SerializableXactHashLock);
4878 : :
4879 : : /* xid will not be found if it wasn't a serializable transaction */
4880 [ + + ]: 395 : if (sxid == NULL)
4881 : 383 : return;
4882 : :
4883 : : /* Release its locks */
4884 : 12 : MySerializableXact = sxid->myXact;
4692 4885 : 12 : MyXactDidWrite = true; /* conservatively assume that we wrote
4886 : : * something */
1857 tmunro@postgresql.or 4887 : 12 : ReleasePredicateLocks(isCommit, false);
4888 : : }
4889 : :
4890 : : /*
4891 : : * Re-acquire a predicate lock belonging to a transaction that was prepared.
4892 : : */
4893 : : void
4815 heikki.linnakangas@i 4894 :UBC 0 : predicatelock_twophase_recover(TransactionId xid, uint16 info,
4895 : : void *recdata, uint32 len)
4896 : : {
4897 : : TwoPhasePredicateRecord *record;
4898 : :
4899 [ # # ]: 0 : Assert(len == sizeof(TwoPhasePredicateRecord));
4900 : :
4901 : 0 : record = (TwoPhasePredicateRecord *) recdata;
4902 : :
4903 [ # # # # ]: 0 : Assert((record->type == TWOPHASEPREDICATERECORD_XACT) ||
4904 : : (record->type == TWOPHASEPREDICATERECORD_LOCK));
4905 : :
4906 [ # # ]: 0 : if (record->type == TWOPHASEPREDICATERECORD_XACT)
4907 : : {
4908 : : /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4909 : : TwoPhasePredicateXactRecord *xactRecord;
4910 : : SERIALIZABLEXACT *sxact;
4911 : : SERIALIZABLEXID *sxid;
4912 : : SERIALIZABLEXIDTAG sxidtag;
4913 : : bool found;
4914 : :
4915 : 0 : xactRecord = (TwoPhasePredicateXactRecord *) &record->data.xactRecord;
4916 : :
4917 : 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4918 : 0 : sxact = CreatePredXact();
4919 [ # # ]: 0 : if (!sxact)
4920 [ # # ]: 0 : ereport(ERROR,
4921 : : (errcode(ERRCODE_OUT_OF_MEMORY),
4922 : : errmsg("out of shared memory")));
4923 : :
4924 : : /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
42 heikki.linnakangas@i 4925 :UNC 0 : sxact->vxid.procNumber = INVALID_PROC_NUMBER;
4815 heikki.linnakangas@i 4926 :UBC 0 : sxact->vxid.localTransactionId = (LocalTransactionId) xid;
4927 : 0 : sxact->pid = 0;
42 heikki.linnakangas@i 4928 :UNC 0 : sxact->pgprocno = INVALID_PROC_NUMBER;
4929 : :
4930 : : /* a prepared xact hasn't committed yet */
4665 heikki.linnakangas@i 4931 :UBC 0 : sxact->prepareSeqNo = RecoverySerCommitSeqNo;
4815 4932 : 0 : sxact->commitSeqNo = InvalidSerCommitSeqNo;
4933 : 0 : sxact->finishedBefore = InvalidTransactionId;
4934 : :
4935 : 0 : sxact->SeqNo.lastCommitBeforeSnapshot = RecoverySerCommitSeqNo;
4936 : :
4937 : : /*
4938 : : * Don't need to track this; no transactions running at the time the
4939 : : * recovered xact started are still active, except possibly other
4940 : : * prepared xacts and we don't care whether those are RO_SAFE or not.
4941 : : */
451 andres@anarazel.de 4942 : 0 : dlist_init(&(sxact->possibleUnsafeConflicts));
4943 : :
4944 : 0 : dlist_init(&(sxact->predicateLocks));
4945 : 0 : dlist_node_init(&sxact->finishedLink);
4946 : :
4815 heikki.linnakangas@i 4947 : 0 : sxact->topXid = xid;
4948 : 0 : sxact->xmin = xactRecord->xmin;
4949 : 0 : sxact->flags = xactRecord->flags;
4950 [ # # ]: 0 : Assert(SxactIsPrepared(sxact));
4951 [ # # ]: 0 : if (!SxactIsReadOnly(sxact))
4952 : : {
4953 : 0 : ++(PredXact->WritableSxactCount);
4954 [ # # ]: 0 : Assert(PredXact->WritableSxactCount <=
4955 : : (MaxBackends + max_prepared_xacts));
4956 : : }
4957 : :
4958 : : /*
4959 : : * We don't know whether the transaction had any conflicts or not, so
4960 : : * we'll conservatively assume that it had both a conflict in and a
4961 : : * conflict out, and represent that with the summary conflict flags.
4962 : : */
451 andres@anarazel.de 4963 : 0 : dlist_init(&(sxact->outConflicts));
4964 : 0 : dlist_init(&(sxact->inConflicts));
4428 heikki.linnakangas@i 4965 : 0 : sxact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
4966 : 0 : sxact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4967 : :
4968 : : /* Register the transaction's xid */
4815 4969 : 0 : sxidtag.xid = xid;
4970 : 0 : sxid = (SERIALIZABLEXID *) hash_search(SerializableXidHash,
4971 : : &sxidtag,
4972 : : HASH_ENTER, &found);
4756 rhaas@postgresql.org 4973 [ # # ]: 0 : Assert(sxid != NULL);
4815 heikki.linnakangas@i 4974 [ # # ]: 0 : Assert(!found);
4975 : 0 : sxid->myXact = (SERIALIZABLEXACT *) sxact;
4976 : :
4977 : : /*
4978 : : * Update global xmin. Note that this is a special case compared to
4979 : : * registering a normal transaction, because the global xmin might go
4980 : : * backwards. That's OK, because until recovery is over we're not
4981 : : * going to complete any transactions or create any non-prepared
4982 : : * transactions, so there's no danger of throwing away.
4983 : : */
4984 [ # # # # ]: 0 : if ((!TransactionIdIsValid(PredXact->SxactGlobalXmin)) ||
4985 : 0 : (TransactionIdFollows(PredXact->SxactGlobalXmin, sxact->xmin)))
4986 : : {
4987 : 0 : PredXact->SxactGlobalXmin = sxact->xmin;
4988 : 0 : PredXact->SxactGlobalXminCount = 1;
1430 tgl@sss.pgh.pa.us 4989 : 0 : SerialSetActiveSerXmin(sxact->xmin);
4990 : : }
4815 heikki.linnakangas@i 4991 [ # # ]: 0 : else if (TransactionIdEquals(sxact->xmin, PredXact->SxactGlobalXmin))
4992 : : {
4993 [ # # ]: 0 : Assert(PredXact->SxactGlobalXminCount > 0);
4994 : 0 : PredXact->SxactGlobalXminCount++;
4995 : : }
4996 : :
4997 : 0 : LWLockRelease(SerializableXactHashLock);
4998 : : }
4999 [ # # ]: 0 : else if (record->type == TWOPHASEPREDICATERECORD_LOCK)
5000 : : {
5001 : : /* Lock record. Recreate the PREDICATELOCK */
5002 : : TwoPhasePredicateLockRecord *lockRecord;
5003 : : SERIALIZABLEXID *sxid;
5004 : : SERIALIZABLEXACT *sxact;
5005 : : SERIALIZABLEXIDTAG sxidtag;
5006 : : uint32 targettaghash;
5007 : :
5008 : 0 : lockRecord = (TwoPhasePredicateLockRecord *) &record->data.lockRecord;
5009 : 0 : targettaghash = PredicateLockTargetTagHashCode(&lockRecord->target);
5010 : :
5011 : 0 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
5012 : 0 : sxidtag.xid = xid;
5013 : : sxid = (SERIALIZABLEXID *)
5014 : 0 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
5015 : 0 : LWLockRelease(SerializableXactHashLock);
5016 : :
5017 [ # # ]: 0 : Assert(sxid != NULL);
5018 : 0 : sxact = sxid->myXact;
5019 [ # # ]: 0 : Assert(sxact != InvalidSerializableXact);
5020 : :
5021 : 0 : CreatePredicateLock(&lockRecord->target, targettaghash, sxact);
5022 : : }
5023 : 0 : }
5024 : :
5025 : : /*
5026 : : * Prepare to share the current SERIALIZABLEXACT with parallel workers.
5027 : : * Return a handle object that can be used by AttachSerializableXact() in a
5028 : : * parallel worker.
5029 : : */
5030 : : SerializableXactHandle
1857 tmunro@postgresql.or 5031 :CBC 414 : ShareSerializableXact(void)
5032 : : {
5033 : 414 : return MySerializableXact;
5034 : : }
5035 : :
5036 : : /*
5037 : : * Allow parallel workers to import the leader's SERIALIZABLEXACT.
5038 : : */
5039 : : void
5040 : 1322 : AttachSerializableXact(SerializableXactHandle handle)
5041 : : {
5042 : :
5043 [ - + ]: 1322 : Assert(MySerializableXact == InvalidSerializableXact);
5044 : :
5045 : 1322 : MySerializableXact = (SERIALIZABLEXACT *) handle;
5046 [ + + ]: 1322 : if (MySerializableXact != InvalidSerializableXact)
5047 : 13 : CreateLocalPredicateLockHash();
5048 : 1322 : }
|
