* examining the MVCC data.)
*
* (1) Besides tuples actually read, they must cover ranges of tuples
- * which would have been read based on the predicate. This will
+ * which would have been read based on the predicate. This will
* require modelling the predicates through locks against database
* objects such as pages, index ranges, or entire tables.
*
- * (2) They must be kept in RAM for quick access. Because of this, it
+ * (2) They must be kept in RAM for quick access. Because of this, it
* isn't possible to always maintain tuple-level granularity -- when
* the space allocated to store these approaches exhaustion, a
* request for a lock may need to scan for situations where a single
*
* (4) While they are associated with a transaction, they must survive
* a successful COMMIT of that transaction, and remain until all
- * overlapping transactions complete. This even means that they
+ * overlapping transactions complete. This even means that they
* must survive termination of the transaction's process. If a
* top level transaction is rolled back, however, it is immediately
* flagged so that it can be ignored, and its SIREAD locks can be
* an existing SIREAD lock for the same transaction, the SIREAD lock
* can be deleted.
*
- * (7) A write from a serializable transaction must ensure that a xact
+ * (7) A write from a serializable transaction must ensure that an xact
* record exists for the transaction, with the same lifespan (until
* all concurrent transaction complete or the transaction is rolled
* back) so that rw-dependencies to that transaction can be
* may yet matter because they overlap still-active transactions.
*
* SerializablePredicateLockListLock
- * - Protects the linked list of locks held by a transaction. Note
+ * - Protects the linked list of locks held by a transaction. Note
* that the locks themselves are also covered by the partition
* locks of their respective lock targets; this lock only affects
* the linked list connecting the locks related to a transaction.
* - All transactions share this single lock (with no partitioning).
* - There is never a need for a process other than the one running
* an active transaction to walk the list of locks held by that
- * transaction.
+ * transaction, except parallel query workers sharing the leader's
+ * transaction. In the parallel case, an extra per-sxact lock is
+ * taken; see below.
* - It is relatively infrequent that another process needs to
* modify the list for a transaction, but it does happen for such
* things as index page splits for pages with predicate locks and
- * freeing of predicate locked pages by a vacuum process. When
+ * freeing of predicate locked pages by a vacuum process. When
* removing a lock in such cases, the lock itself contains the
* pointers needed to remove it from the list. When adding a
* lock in such cases, the lock can be added using the anchor in
- * the transaction structure. Neither requires walking the list.
+ * the transaction structure. Neither requires walking the list.
* - Cleaning up the list for a terminated transaction is sometimes
* not done on a retail basis, in which case no lock is required.
* - Due to the above, a process accessing its active transaction's
* than its own active transaction must acquire an exclusive
* lock.
*
- * FirstPredicateLockMgrLock based partition locks
+ * SERIALIZABLEXACT's member 'predicateLockListLock'
+ * - Protects the linked list of locks held by a transaction. Only
+ * needed for parallel mode, where multiple backends share the
+ * same SERIALIZABLEXACT object. Not needed if
+ * SerializablePredicateLockListLock is held exclusively.
+ *
+ * PredicateLockHashPartitionLock(hashcode)
* - The same lock protects a target, all locks on that target, and
- * the linked list of locks on the target..
- * - When more than one is needed, acquire in ascending order.
+ * the linked list of locks on the target.
+ * - When more than one is needed, acquire in ascending address order.
+ * - When all are needed (rare), acquire in ascending index order with
+ * PredicateLockHashPartitionLockByIndex(index).
*
* SerializableXactHashLock
* - Protects both PredXact and SerializableXidHash.
*
*
- * Portions Copyright (c) 1996-2014, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California
*
*
* predicate lock maintenance
* GetSerializableTransactionSnapshot(Snapshot snapshot)
* SetSerializableTransactionSnapshot(Snapshot snapshot,
- * TransactionId sourcexid)
+ * VirtualTransactionId *sourcevxid)
* RegisterPredicateLockingXid(void)
* PredicateLockRelation(Relation relation, Snapshot snapshot)
* PredicateLockPage(Relation relation, BlockNumber blkno,
* PredicateLockPageCombine(Relation relation, BlockNumber oldblkno,
* BlockNumber newblkno)
* TransferPredicateLocksToHeapRelation(Relation relation)
- * ReleasePredicateLocks(bool isCommit)
+ * ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
*
* conflict detection (may also trigger rollback)
* CheckForSerializableConflictOut(bool visible, Relation relation,
#include "postgres.h"
+#include "access/heapam.h"
#include "access/htup_details.h"
+#include "access/parallel.h"
#include "access/slru.h"
#include "access/subtrans.h"
#include "access/transam.h"
#include "access/twophase.h"
#include "access/twophase_rmgr.h"
#include "access/xact.h"
+#include "access/xlog.h"
#include "miscadmin.h"
+#include "pgstat.h"
#include "storage/bufmgr.h"
#include "storage/predicate.h"
#include "storage/predicate_internals.h"
#include "storage/procarray.h"
#include "utils/rel.h"
#include "utils/snapmgr.h"
-#include "utils/tqual.h"
/* Uncomment the next line to test the graceful degradation code. */
/* #define TEST_OLDSERXID */
#define PredicateLockHashPartition(hashcode) \
((hashcode) % NUM_PREDICATELOCK_PARTITIONS)
#define PredicateLockHashPartitionLock(hashcode) \
- ((LWLockId) (FirstPredicateLockMgrLock + PredicateLockHashPartition(hashcode)))
+ (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + \
+ PredicateLockHashPartition(hashcode)].lock)
+#define PredicateLockHashPartitionLockByIndex(i) \
+ (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + (i)].lock)
#define NPREDICATELOCKTARGETENTS() \
mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
#define SxactIsDeferrableWaiting(sxact) (((sxact)->flags & SXACT_FLAG_DEFERRABLE_WAITING) != 0)
#define SxactIsROSafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_SAFE) != 0)
#define SxactIsROUnsafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_UNSAFE) != 0)
+#define SxactIsPartiallyReleased(sxact) (((sxact)->flags & SXACT_FLAG_PARTIALLY_RELEASED) != 0)
/*
* Compute the hash code associated with a PREDICATELOCKTARGETTAG.
* the lock partition number from the hashcode.
*/
#define PredicateLockTargetTagHashCode(predicatelocktargettag) \
- (tag_hash((predicatelocktargettag), sizeof(PREDICATELOCKTARGETTAG)))
+ get_hash_value(PredicateLockTargetHash, predicatelocktargettag)
/*
* Given a predicate lock tag, and the hash for its target,
#define OLDSERXID_ENTRIESPERPAGE (OLDSERXID_PAGESIZE / OLDSERXID_ENTRYSIZE)
/*
- * Set maximum pages based on the lesser of the number needed to track all
- * transactions and the maximum that SLRU supports.
+ * Set maximum pages based on the number needed to track all transactions.
*/
-#define OLDSERXID_MAX_PAGE Min(SLRU_PAGES_PER_SEGMENT * 0x10000 - 1, \
- (MaxTransactionId) / OLDSERXID_ENTRIESPERPAGE)
+#define OLDSERXID_MAX_PAGE (MaxTransactionId / OLDSERXID_ENTRIESPERPAGE)
#define OldSerXidNextPage(page) (((page) >= OLDSERXID_MAX_PAGE) ? 0 : (page) + 1)
(OldSerXidSlruCtl->shared->page_buffer[slotno] + \
((((uint32) (xid)) % OLDSERXID_ENTRIESPERPAGE) * OLDSERXID_ENTRYSIZE))))
-#define OldSerXidPage(xid) ((((uint32) (xid)) / OLDSERXID_ENTRIESPERPAGE) % (OLDSERXID_MAX_PAGE + 1))
-#define OldSerXidSegment(page) ((page) / SLRU_PAGES_PER_SEGMENT)
+#define OldSerXidPage(xid) (((uint32) (xid)) / OLDSERXID_ENTRIESPERPAGE)
typedef struct OldSerXidControlData
{
int headPage; /* newest initialized page */
TransactionId headXid; /* newest valid Xid in the SLRU */
TransactionId tailXid; /* oldest xmin we might be interested in */
- bool warningIssued; /* have we issued SLRU wrap-around warning? */
-} OldSerXidControlData;
+} OldSerXidControlData;
typedef struct OldSerXidControlData *OldSerXidControl;
static SERIALIZABLEXACT *OldCommittedSxact;
-/* This configuration variable is used to set the predicate lock table size */
-int max_predicate_locks_per_xact; /* set by guc.c */
+/*
+ * These configuration variables are used to set the predicate lock table size
+ * and to control promotion of predicate locks to coarser granularity in an
+ * attempt to degrade performance (mostly as false positive serialization
+ * failure) gracefully in the face of memory pressurel
+ */
+int max_predicate_locks_per_xact; /* set by guc.c */
+int max_predicate_locks_per_relation; /* set by guc.c */
+int max_predicate_locks_per_page; /* set by guc.c */
/*
* This provides a list of objects in order to track transactions
- * participating in predicate locking. Entries in the list are fixed size,
+ * participating in predicate locking. Entries in the list are fixed size,
* and reside in shared memory. The memory address of an entry must remain
* fixed during its lifetime. The list will be protected from concurrent
* update externally; no provision is made in this code to manage that. The
*/
static const PREDICATELOCKTARGETTAG ScratchTargetTag = {0, 0, 0, 0};
static uint32 ScratchTargetTagHash;
-static int ScratchPartitionLock;
+static LWLock *ScratchPartitionLock;
/*
* The local hash table used to determine when to combine multiple fine-
static SERIALIZABLEXACT *MySerializableXact = InvalidSerializableXact;
static bool MyXactDidWrite = false;
+/*
+ * The SXACT_FLAG_RO_UNSAFE optimization might lead us to release
+ * MySerializableXact early. If that happens in a parallel query, the leader
+ * needs to defer the destruction of the SERIALIZABLEXACT until end of
+ * transaction, because the workers still have a reference to it. In that
+ * case, the leader stores it here.
+ */
+static SERIALIZABLEXACT *SavedSerializableXact = InvalidSerializableXact;
+
/* local functions */
static SERIALIZABLEXACT *CreatePredXact(void);
static void SummarizeOldestCommittedSxact(void);
static Snapshot GetSafeSnapshot(Snapshot snapshot);
static Snapshot GetSerializableTransactionSnapshotInt(Snapshot snapshot,
- TransactionId sourcexid);
+ VirtualTransactionId *sourcevxid,
+ int sourcepid);
static bool PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag);
static bool GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag,
- PREDICATELOCKTARGETTAG *parent);
+ PREDICATELOCKTARGETTAG *parent);
static bool CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag);
static void RemoveScratchTarget(bool lockheld);
static void RestoreScratchTarget(bool lockheld);
static void RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target,
- uint32 targettaghash);
+ uint32 targettaghash);
static void DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag);
-static int PredicateLockPromotionThreshold(const PREDICATELOCKTARGETTAG *tag);
+static int MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag);
static bool CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag);
static void DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag);
static void CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
- uint32 targettaghash,
- SERIALIZABLEXACT *sxact);
+ uint32 targettaghash,
+ SERIALIZABLEXACT *sxact);
static void DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash);
static bool TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
- PREDICATELOCKTARGETTAG newtargettag,
- bool removeOld);
+ PREDICATELOCKTARGETTAG newtargettag,
+ bool removeOld);
static void PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag);
static void DropAllPredicateLocksFromTable(Relation relation,
- bool transfer);
+ bool transfer);
static void SetNewSxactGlobalXmin(void);
static void ClearOldPredicateLocks(void);
static void ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
- bool summarize);
+ bool summarize);
static bool XidIsConcurrent(TransactionId xid);
static void CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag);
static void FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer);
static void OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader,
- SERIALIZABLEXACT *writer);
+ SERIALIZABLEXACT *writer);
+static void CreateLocalPredicateLockHash(void);
+static void ReleasePredicateLocksLocal(void);
/*------------------------------------------------------------------------*/
* as RO-safe since the last call, we release all predicate locks and reset
* MySerializableXact. That makes subsequent calls to return quickly.
*
- * This is marked as 'inline' to make to eliminate the function call overhead
- * in the common case that serialization is not needed.
+ * This is marked as 'inline' to eliminate the function call overhead in the
+ * common case that serialization is not needed.
*/
static inline bool
SerializationNeededForRead(Relation relation, Snapshot snapshot)
* Don't acquire locks or conflict when scanning with a special snapshot.
* This excludes things like CLUSTER and REINDEX. They use the wholesale
* functions TransferPredicateLocksToHeapRelation() and
- * CheckTableForSerializableConflictIn() to participate serialization, but
- * the scans involved don't need serialization.
+ * CheckTableForSerializableConflictIn() to participate in serialization,
+ * but the scans involved don't need serialization.
*/
if (!IsMVCCSnapshot(snapshot))
return false;
*/
if (SxactIsROSafe(MySerializableXact))
{
- ReleasePredicateLocks(false);
+ ReleasePredicateLocks(false, true);
return false;
}
/*
* These functions are a simple implementation of a list for this specific
- * type of struct. If there is ever a generalized shared memory list, we
+ * type of struct. If there is ever a generalized shared memory list, we
* should probably switch to that.
*/
static SERIALIZABLEXACT *
int diff;
/*
- * We have to compare modulo (OLDSERXID_MAX_PAGE+1)/2. Both inputs should
+ * We have to compare modulo (OLDSERXID_MAX_PAGE+1)/2. Both inputs should
* be in the range 0..OLDSERXID_MAX_PAGE.
*/
Assert(p >= 0 && p <= OLDSERXID_MAX_PAGE);
* Set up SLRU management of the pg_serial data.
*/
OldSerXidSlruCtl->PagePrecedes = OldSerXidPagePrecedesLogically;
- SimpleLruInit(OldSerXidSlruCtl, "OldSerXid SLRU Ctl",
- NUM_OLDSERXID_BUFFERS, 0, OldSerXidLock, "pg_serial");
+ SimpleLruInit(OldSerXidSlruCtl, "oldserxid",
+ NUM_OLDSERXID_BUFFERS, 0, OldSerXidLock, "pg_serial",
+ LWTRANCHE_OLDSERXID_BUFFERS);
/* Override default assumption that writes should be fsync'd */
OldSerXidSlruCtl->do_fsync = false;
oldSerXidControl = (OldSerXidControl)
ShmemInitStruct("OldSerXidControlData", sizeof(OldSerXidControlData), &found);
+ Assert(found == IsUnderPostmaster);
if (!found)
{
/*
oldSerXidControl->headPage = -1;
oldSerXidControl->headXid = InvalidTransactionId;
oldSerXidControl->tailXid = InvalidTransactionId;
- oldSerXidControl->warningIssued = false;
}
}
if (isNewPage)
oldSerXidControl->headPage = targetPage;
- /*
- * Give a warning if we're about to run out of SLRU pages.
- *
- * slru.c has a maximum of 64k segments, with 32 (SLRU_PAGES_PER_SEGMENT)
- * pages each. We need to store a 64-bit integer for each Xid, and with
- * default 8k block size, 65536*32 pages is only enough to cover 2^30
- * XIDs. If we're about to hit that limit and wrap around, warn the user.
- *
- * To avoid spamming the user, we only give one warning when we've used 1
- * billion XIDs, and stay silent until the situation is fixed and the
- * number of XIDs used falls below 800 million again.
- *
- * XXX: We have no safeguard to actually *prevent* the wrap-around,
- * though. All you get is a warning.
- */
- if (oldSerXidControl->warningIssued)
- {
- TransactionId lowWatermark;
-
- lowWatermark = tailXid + 800000000;
- if (lowWatermark < FirstNormalTransactionId)
- lowWatermark = FirstNormalTransactionId;
- if (TransactionIdPrecedes(xid, lowWatermark))
- oldSerXidControl->warningIssued = false;
- }
- else
- {
- TransactionId highWatermark;
-
- highWatermark = tailXid + 1000000000;
- if (highWatermark < FirstNormalTransactionId)
- highWatermark = FirstNormalTransactionId;
- if (TransactionIdFollows(xid, highWatermark))
- {
- oldSerXidControl->warningIssued = true;
- ereport(WARNING,
- (errmsg("memory for serializable conflict tracking is nearly exhausted"),
- errhint("There might be an idle transaction or a forgotten prepared transaction causing this.")));
- }
- }
-
if (isNewPage)
{
/* Initialize intervening pages. */
}
/*
- * Get the minimum commitSeqNo for any conflict out for the given xid. For
+ * Get the minimum commitSeqNo for any conflict out for the given xid. For
* a transaction which exists but has no conflict out, InvalidSerCommitSeqNo
* will be returned.
*/
/*
* When no sxacts are active, nothing overlaps, set the xid values to
* invalid to show that there are no valid entries. Don't clear headPage,
- * though. A new xmin might still land on that page, and we don't want to
+ * though. A new xmin might still land on that page, and we don't want to
* repeatedly zero out the same page.
*/
if (!TransactionIdIsValid(xid))
InitPredicateLocks(void)
{
HASHCTL info;
- int hash_flags;
long max_table_size;
Size requestSize;
bool found;
+#ifndef EXEC_BACKEND
+ Assert(!IsUnderPostmaster);
+#endif
+
/*
* Compute size of predicate lock target hashtable. Note these
* calculations must agree with PredicateLockShmemSize!
MemSet(&info, 0, sizeof(info));
info.keysize = sizeof(PREDICATELOCKTARGETTAG);
info.entrysize = sizeof(PREDICATELOCKTARGET);
- info.hash = tag_hash;
info.num_partitions = NUM_PREDICATELOCK_PARTITIONS;
- hash_flags = (HASH_ELEM | HASH_FUNCTION | HASH_PARTITION | HASH_FIXED_SIZE);
PredicateLockTargetHash = ShmemInitHash("PREDICATELOCKTARGET hash",
max_table_size,
max_table_size,
&info,
- hash_flags);
-
- /* Assume an average of 2 xacts per target */
- max_table_size *= 2;
+ HASH_ELEM | HASH_BLOBS |
+ HASH_PARTITION | HASH_FIXED_SIZE);
/*
* Reserve a dummy entry in the hash table; we use it to make sure there's
* because running out of space there could mean aborting a
* non-serializable transaction.
*/
- hash_search(PredicateLockTargetHash, &ScratchTargetTag, HASH_ENTER, NULL);
+ if (!IsUnderPostmaster)
+ {
+ (void) hash_search(PredicateLockTargetHash, &ScratchTargetTag,
+ HASH_ENTER, &found);
+ Assert(!found);
+ }
+
+ /* Pre-calculate the hash and partition lock of the scratch entry */
+ ScratchTargetTagHash = PredicateLockTargetTagHashCode(&ScratchTargetTag);
+ ScratchPartitionLock = PredicateLockHashPartitionLock(ScratchTargetTagHash);
/*
* Allocate hash table for PREDICATELOCK structs. This stores per
info.entrysize = sizeof(PREDICATELOCK);
info.hash = predicatelock_hash;
info.num_partitions = NUM_PREDICATELOCK_PARTITIONS;
- hash_flags = (HASH_ELEM | HASH_FUNCTION | HASH_PARTITION | HASH_FIXED_SIZE);
+
+ /* Assume an average of 2 xacts per target */
+ max_table_size *= 2;
PredicateLockHash = ShmemInitHash("PREDICATELOCK hash",
max_table_size,
max_table_size,
&info,
- hash_flags);
+ HASH_ELEM | HASH_FUNCTION |
+ HASH_PARTITION | HASH_FIXED_SIZE);
/*
* Compute size for serializable transaction hashtable. Note these
PredXact = ShmemInitStruct("PredXactList",
PredXactListDataSize,
&found);
+ Assert(found == IsUnderPostmaster);
if (!found)
{
int i;
requestSize = mul_size((Size) max_table_size,
PredXactListElementDataSize);
PredXact->element = ShmemAlloc(requestSize);
- if (PredXact->element == NULL)
- ereport(ERROR,
- (errcode(ERRCODE_OUT_OF_MEMORY),
- errmsg("not enough shared memory for elements of data structure"
- " \"%s\" (%zu bytes requested)",
- "PredXactList", requestSize)));
/* Add all elements to available list, clean. */
memset(PredXact->element, 0, requestSize);
for (i = 0; i < max_table_size; i++)
{
+ LWLockInitialize(&PredXact->element[i].sxact.predicateLockListLock,
+ LWTRANCHE_SXACT);
SHMQueueInsertBefore(&(PredXact->availableList),
&(PredXact->element[i].link));
}
MemSet(&info, 0, sizeof(info));
info.keysize = sizeof(SERIALIZABLEXIDTAG);
info.entrysize = sizeof(SERIALIZABLEXID);
- info.hash = tag_hash;
- hash_flags = (HASH_ELEM | HASH_FUNCTION | HASH_FIXED_SIZE);
SerializableXidHash = ShmemInitHash("SERIALIZABLEXID hash",
max_table_size,
max_table_size,
&info,
- hash_flags);
+ HASH_ELEM | HASH_BLOBS |
+ HASH_FIXED_SIZE);
/*
* Allocate space for tracking rw-conflicts in lists attached to the
RWConflictPool = ShmemInitStruct("RWConflictPool",
RWConflictPoolHeaderDataSize,
&found);
+ Assert(found == IsUnderPostmaster);
if (!found)
{
int i;
requestSize = mul_size((Size) max_table_size,
RWConflictDataSize);
RWConflictPool->element = ShmemAlloc(requestSize);
- if (RWConflictPool->element == NULL)
- ereport(ERROR,
- (errcode(ERRCODE_OUT_OF_MEMORY),
- errmsg("not enough shared memory for elements of data structure"
- " \"%s\" (%zu bytes requested)",
- "RWConflictPool", requestSize)));
/* Add all elements to available list, clean. */
memset(RWConflictPool->element, 0, requestSize);
for (i = 0; i < max_table_size; i++)
ShmemInitStruct("FinishedSerializableTransactions",
sizeof(SHM_QUEUE),
&found);
+ Assert(found == IsUnderPostmaster);
if (!found)
SHMQueueInit(FinishedSerializableTransactions);
* transactions.
*/
OldSerXidInit();
-
- /* Pre-calculate the hash and partition lock of the scratch entry */
- ScratchTargetTagHash = PredicateLockTargetTagHashCode(&ScratchTargetTag);
- ScratchPartitionLock = PredicateLockHashPartitionLock(ScratchTargetTagHash);
}
/*
* in ascending order, then SerializableXactHashLock.
*/
for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
- LWLockAcquire(FirstPredicateLockMgrLock + i, LW_SHARED);
+ LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED);
LWLockAcquire(SerializableXactHashLock, LW_SHARED);
/* Get number of locks and allocate appropriately-sized arrays. */
/* Release locks in reverse order */
LWLockRelease(SerializableXactHashLock);
for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
- LWLockRelease(FirstPredicateLockMgrLock + i);
+ LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
return data;
}
/*
* Grab the first sxact off the finished list -- this will be the earliest
- * commit. Remove it from the list.
+ * commit. Remove it from the list.
*/
sxact = (SERIALIZABLEXACT *)
SHMQueueNext(FinishedSerializableTransactions,
/* Add to SLRU summary information. */
if (TransactionIdIsValid(sxact->topXid) && !SxactIsReadOnly(sxact))
OldSerXidAdd(sxact->topXid, SxactHasConflictOut(sxact)
- ? sxact->SeqNo.earliestOutConflictCommit : InvalidSerCommitSeqNo);
+ ? sxact->SeqNo.earliestOutConflictCommit : InvalidSerCommitSeqNo);
/* Summarize and release the detail. */
ReleaseOneSerializableXact(sxact, false, true);
* one passed to it, but we avoid assuming that here.
*/
snapshot = GetSerializableTransactionSnapshotInt(origSnapshot,
- InvalidTransactionId);
+ NULL, InvalidPid);
if (MySerializableXact == InvalidSerializableXact)
return snapshot; /* no concurrent r/w xacts; it's safe */
SxactIsROUnsafe(MySerializableXact)))
{
LWLockRelease(SerializableXactHashLock);
- ProcWaitForSignal();
+ ProcWaitForSignal(WAIT_EVENT_SAFE_SNAPSHOT);
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
}
MySerializableXact->flags &= ~SXACT_FLAG_DEFERRABLE_WAITING;
ereport(DEBUG2,
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
errmsg("deferrable snapshot was unsafe; trying a new one")));
- ReleasePredicateLocks(false);
+ ReleasePredicateLocks(false, false);
}
/*
* Now we have a safe snapshot, so we don't need to do any further checks.
*/
Assert(SxactIsROSafe(MySerializableXact));
- ReleasePredicateLocks(false);
+ ReleasePredicateLocks(false, true);
return snapshot;
}
+/*
+ * GetSafeSnapshotBlockingPids
+ * If the specified process is currently blocked in GetSafeSnapshot,
+ * write the process IDs of all processes that it is blocked by
+ * into the caller-supplied buffer output[]. The list is truncated at
+ * output_size, and the number of PIDs written into the buffer is
+ * returned. Returns zero if the given PID is not currently blocked
+ * in GetSafeSnapshot.
+ */
+int
+GetSafeSnapshotBlockingPids(int blocked_pid, int *output, int output_size)
+{
+ int num_written = 0;
+ SERIALIZABLEXACT *sxact;
+
+ LWLockAcquire(SerializableXactHashLock, LW_SHARED);
+
+ /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
+ for (sxact = FirstPredXact(); sxact != NULL; sxact = NextPredXact(sxact))
+ {
+ if (sxact->pid == blocked_pid)
+ break;
+ }
+
+ /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
+ if (sxact != NULL && SxactIsDeferrableWaiting(sxact))
+ {
+ RWConflict possibleUnsafeConflict;
+
+ /* Traverse the list of possible unsafe conflicts collecting PIDs. */
+ possibleUnsafeConflict = (RWConflict)
+ SHMQueueNext(&sxact->possibleUnsafeConflicts,
+ &sxact->possibleUnsafeConflicts,
+ offsetof(RWConflictData, inLink));
+
+ while (possibleUnsafeConflict != NULL && num_written < output_size)
+ {
+ output[num_written++] = possibleUnsafeConflict->sxactOut->pid;
+ possibleUnsafeConflict = (RWConflict)
+ SHMQueueNext(&sxact->possibleUnsafeConflicts,
+ &possibleUnsafeConflict->inLink,
+ offsetof(RWConflictData, inLink));
+ }
+ }
+
+ LWLockRelease(SerializableXactHashLock);
+
+ return num_written;
+}
+
/*
* Acquire a snapshot that can be used for the current transaction.
*
return GetSafeSnapshot(snapshot);
return GetSerializableTransactionSnapshotInt(snapshot,
- InvalidTransactionId);
+ NULL, InvalidPid);
}
/*
*/
void
SetSerializableTransactionSnapshot(Snapshot snapshot,
- TransactionId sourcexid)
+ VirtualTransactionId *sourcevxid,
+ int sourcepid)
{
Assert(IsolationIsSerializable());
+ /*
+ * If this is called by parallel.c in a parallel worker, we don't want to
+ * create a SERIALIZABLEXACT just yet because the leader's
+ * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
+ * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
+ * case, because the leader has already determined that the snapshot it
+ * has passed us is safe. So there is nothing for us to do.
+ */
+ if (IsParallelWorker())
+ return;
+
/*
* We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
* import snapshots, since there's no way to wait for a safe snapshot when
- * we're using the snap we're told to. (XXX instead of throwing an error,
+ * we're using the snap we're told to. (XXX instead of throwing an error,
* we could just ignore the XactDeferrable flag?)
*/
if (XactReadOnly && XactDeferrable)
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("a snapshot-importing transaction must not be READ ONLY DEFERRABLE")));
- (void) GetSerializableTransactionSnapshotInt(snapshot, sourcexid);
+ (void) GetSerializableTransactionSnapshotInt(snapshot, sourcevxid,
+ sourcepid);
}
/*
*/
static Snapshot
GetSerializableTransactionSnapshotInt(Snapshot snapshot,
- TransactionId sourcexid)
+ VirtualTransactionId *sourcevxid,
+ int sourcepid)
{
PGPROC *proc;
VirtualTransactionId vxid;
SERIALIZABLEXACT *sxact,
*othersxact;
- HASHCTL hash_ctl;
/* We only do this for serializable transactions. Once. */
Assert(MySerializableXact == InvalidSerializableXact);
Assert(!RecoveryInProgress());
+ /*
+ * Since all parts of a serializable transaction must use the same
+ * snapshot, it is too late to establish one after a parallel operation
+ * has begun.
+ */
+ if (IsInParallelMode())
+ elog(ERROR, "cannot establish serializable snapshot during a parallel operation");
+
proc = MyProc;
Assert(proc != NULL);
GET_VXID_FROM_PGPROC(vxid, *proc);
* release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
* this means we have to create the sxact first, which is a bit annoying
* (in particular, an elog(ERROR) in procarray.c would cause us to leak
- * the sxact). Consider refactoring to avoid this.
+ * the sxact). Consider refactoring to avoid this.
*/
#ifdef TEST_OLDSERXID
SummarizeOldestCommittedSxact();
} while (!sxact);
/* Get the snapshot, or check that it's safe to use */
- if (!TransactionIdIsValid(sourcexid))
+ if (!sourcevxid)
snapshot = GetSnapshotData(snapshot);
- else if (!ProcArrayInstallImportedXmin(snapshot->xmin, sourcexid))
+ else if (!ProcArrayInstallImportedXmin(snapshot->xmin, sourcevxid))
{
ReleasePredXact(sxact);
LWLockRelease(SerializableXactHashLock);
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("could not import the requested snapshot"),
- errdetail("The source transaction %u is not running anymore.",
- sourcexid)));
+ errdetail("The source process with PID %d is not running anymore.",
+ sourcepid)));
}
/*
LWLockRelease(SerializableXactHashLock);
+ CreateLocalPredicateLockHash();
+
+ return snapshot;
+}
+
+static void
+CreateLocalPredicateLockHash(void)
+{
+ HASHCTL hash_ctl;
+
/* Initialize the backend-local hash table of parent locks */
Assert(LocalPredicateLockHash == NULL);
MemSet(&hash_ctl, 0, sizeof(hash_ctl));
hash_ctl.keysize = sizeof(PREDICATELOCKTARGETTAG);
hash_ctl.entrysize = sizeof(LOCALPREDICATELOCK);
- hash_ctl.hash = tag_hash;
LocalPredicateLockHash = hash_create("Local predicate lock",
max_predicate_locks_per_xact,
&hash_ctl,
- HASH_ELEM | HASH_FUNCTION);
-
- return snapshot;
+ HASH_ELEM | HASH_BLOBS);
}
/*
{
PREDICATELOCKTARGETTAG targettag;
uint32 targettaghash;
- LWLockId partitionLock;
+ LWLock *partitionLock;
PREDICATELOCKTARGET *target;
SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
case PREDLOCKTAG_PAGE:
/* parent lock is relation lock */
SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
- GET_PREDICATELOCKTARGETTAG_DB(*tag),
- GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
+ GET_PREDICATELOCKTARGETTAG_DB(*tag),
+ GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
return true;
case PREDLOCKTAG_TUPLE:
/* parent lock is page lock */
SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
- GET_PREDICATELOCKTARGETTAG_DB(*tag),
- GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
- GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
+ GET_PREDICATELOCKTARGETTAG_DB(*tag),
+ GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
+ GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
return true;
}
/*
* Delete child target locks owned by this process.
* This implementation is assuming that the usage of each target tag field
- * is uniform. No need to make this hard if we don't have to.
+ * is uniform. No need to make this hard if we don't have to.
*
- * We aren't acquiring lightweight locks for the predicate lock or lock
+ * We acquire an LWLock in the case of parallel mode, because worker
+ * backends have access to the leader's SERIALIZABLEXACT. Otherwise,
+ * we aren't acquiring LWLocks for the predicate lock or lock
* target structures associated with this transaction unless we're going
* to modify them, because no other process is permitted to modify our
* locks.
LWLockAcquire(SerializablePredicateLockListLock, LW_SHARED);
sxact = MySerializableXact;
+ if (IsInParallelMode())
+ LWLockAcquire(&sxact->predicateLockListLock, LW_EXCLUSIVE);
predlock = (PREDICATELOCK *)
SHMQueueNext(&(sxact->predicateLocks),
&(sxact->predicateLocks),
if (TargetTagIsCoveredBy(oldtargettag, *newtargettag))
{
uint32 oldtargettaghash;
- LWLockId partitionLock;
+ LWLock *partitionLock;
PREDICATELOCK *rmpredlock PG_USED_FOR_ASSERTS_ONLY;
oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
predlock = nextpredlock;
}
+ if (IsInParallelMode())
+ LWLockRelease(&sxact->predicateLockListLock);
LWLockRelease(SerializablePredicateLockListLock);
}
/*
- * Returns the promotion threshold for a given predicate lock
- * target. This is the number of descendant locks required to promote
- * to the specified tag. Note that the threshold includes non-direct
- * descendants, e.g. both tuples and pages for a relation lock.
+ * Returns the promotion limit for a given predicate lock target. This is the
+ * max number of descendant locks allowed before promoting to the specified
+ * tag. Note that the limit includes non-direct descendants (e.g., both tuples
+ * and pages for a relation lock).
+ *
+ * Currently the default limit is 2 for a page lock, and half of the value of
+ * max_pred_locks_per_transaction - 1 for a relation lock, to match behavior
+ * of earlier releases when upgrading.
*
- * TODO SSI: We should do something more intelligent about what the
- * thresholds are, either making it proportional to the number of
- * tuples in a page & pages in a relation, or at least making it a
- * GUC. Currently the threshold is 3 for a page lock, and
- * max_pred_locks_per_transaction/2 for a relation lock, chosen
- * entirely arbitrarily (and without benchmarking).
+ * TODO SSI: We should probably add additional GUCs to allow a maximum ratio
+ * of page and tuple locks based on the pages in a relation, and the maximum
+ * ratio of tuple locks to tuples in a page. This would provide more
+ * generally "balanced" allocation of locks to where they are most useful,
+ * while still allowing the absolute numbers to prevent one relation from
+ * tying up all predicate lock resources.
*/
static int
-PredicateLockPromotionThreshold(const PREDICATELOCKTARGETTAG *tag)
+MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag)
{
switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag))
{
case PREDLOCKTAG_RELATION:
- return max_predicate_locks_per_xact / 2;
+ return max_predicate_locks_per_relation < 0
+ ? (max_predicate_locks_per_xact
+ / (-max_predicate_locks_per_relation)) - 1
+ : max_predicate_locks_per_relation;
case PREDLOCKTAG_PAGE:
- return 3;
+ return max_predicate_locks_per_page;
case PREDLOCKTAG_TUPLE:
else
parentlock->childLocks++;
- if (parentlock->childLocks >=
- PredicateLockPromotionThreshold(&targettag))
+ if (parentlock->childLocks >
+ MaxPredicateChildLocks(&targettag))
{
/*
* We should promote to this parent lock. Continue to check its
PREDICATELOCKTARGET *target;
PREDICATELOCKTAG locktag;
PREDICATELOCK *lock;
- LWLockId partitionLock;
+ LWLock *partitionLock;
bool found;
partitionLock = PredicateLockHashPartitionLock(targettaghash);
LWLockAcquire(SerializablePredicateLockListLock, LW_SHARED);
+ if (IsInParallelMode())
+ LWLockAcquire(&sxact->predicateLockListLock, LW_EXCLUSIVE);
LWLockAcquire(partitionLock, LW_EXCLUSIVE);
/* Make sure that the target is represented. */
locktag.myXact = sxact;
lock = (PREDICATELOCK *)
hash_search_with_hash_value(PredicateLockHash, &locktag,
- PredicateLockHashCodeFromTargetHashCode(&locktag, targettaghash),
+ PredicateLockHashCodeFromTargetHashCode(&locktag, targettaghash),
HASH_ENTER_NULL, &found);
if (!lock)
ereport(ERROR,
}
LWLockRelease(partitionLock);
+ if (IsInParallelMode())
+ LWLockRelease(&sxact->predicateLockListLock);
LWLockRelease(SerializablePredicateLockListLock);
}
}
/*
- * Do quick-but-not-definitive test for a relation lock first. This will
+ * Do quick-but-not-definitive test for a relation lock first. This will
* never cause a return when the relation is *not* locked, but will
* occasionally let the check continue when there really *is* a relation
* level lock.
PREDICATELOCK *nextpredlock;
bool found;
- Assert(LWLockHeldByMe(SerializablePredicateLockListLock));
+ Assert(LWLockHeldByMeInMode(SerializablePredicateLockListLock,
+ LW_EXCLUSIVE));
Assert(LWLockHeldByMe(PredicateLockHashPartitionLock(targettaghash)));
predlock = (PREDICATELOCK *)
* covers it, or if we are absolutely certain that no one will need to
* refer to that lock in the future.
*
- * Caller must hold SerializablePredicateLockListLock.
+ * Caller must hold SerializablePredicateLockListLock exclusively.
*/
static bool
TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
bool removeOld)
{
uint32 oldtargettaghash;
- LWLockId oldpartitionLock;
+ LWLock *oldpartitionLock;
PREDICATELOCKTARGET *oldtarget;
uint32 newtargettaghash;
- LWLockId newpartitionLock;
+ LWLock *newpartitionLock;
bool found;
bool outOfShmem = false;
- Assert(LWLockHeldByMe(SerializablePredicateLockListLock));
+ Assert(LWLockHeldByMeInMode(SerializablePredicateLockListLock,
+ LW_EXCLUSIVE));
oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
newtargettaghash = PredicateLockTargetTagHashCode(&newtargettag);
hash_search_with_hash_value
(PredicateLockHash,
&oldpredlock->tag,
- PredicateLockHashCodeFromTargetHashCode(&oldpredlock->tag,
- oldtargettaghash),
+ PredicateLockHashCodeFromTargetHashCode(&oldpredlock->tag,
+ oldtargettaghash),
HASH_REMOVE, &found);
Assert(found);
}
newpredlock = (PREDICATELOCK *)
hash_search_with_hash_value(PredicateLockHash,
&newpredlocktag,
- PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
- newtargettaghash),
+ PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
+ newtargettaghash),
HASH_ENTER_NULL,
&found);
if (!newpredlock)
/* We shouldn't run out of memory if we're moving locks */
Assert(!outOfShmem);
- /* Put the scrach entry back */
+ /* Put the scratch entry back */
RestoreScratchTarget(false);
}
* transaction which is not serializable.
*
* NOTE: This is currently only called with transfer set to true, but that may
- * change. If we decide to clean up the locks from a table on commit of a
+ * change. If we decide to clean up the locks from a table on commit of a
* transaction which executed DROP TABLE, the false condition will be useful.
*/
static void
heapId = relation->rd_index->indrelid;
}
Assert(heapId != InvalidOid);
- Assert(transfer || !isIndex); /* index OID only makes sense with
- * transfer */
+ Assert(transfer || !isIndex); /* index OID only makes sense with
+ * transfer */
/* Retrieve first time needed, then keep. */
heaptargettaghash = 0;
/* Acquire locks on all lock partitions */
LWLockAcquire(SerializablePredicateLockListLock, LW_EXCLUSIVE);
for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
- LWLockAcquire(FirstPredicateLockMgrLock + i, LW_EXCLUSIVE);
+ LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_EXCLUSIVE);
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
/*
continue; /* already the right lock */
/*
- * If we made it here, we have work to do. We make sure the heap
+ * If we made it here, we have work to do. We make sure the heap
* relation lock exists, then we walk the list of predicate locks for
* the old target we found, moving all locks to the heap relation lock
* -- unless they already hold that.
newpredlock = (PREDICATELOCK *)
hash_search_with_hash_value(PredicateLockHash,
&newpredlocktag,
- PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
- heaptargettaghash),
+ PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
+ heaptargettaghash),
HASH_ENTER,
&found);
if (!found)
/* Release locks in reverse order */
LWLockRelease(SerializableXactHashLock);
for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
- LWLockRelease(FirstPredicateLockMgrLock + i);
+ LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
LWLockRelease(SerializablePredicateLockListLock);
}
* If this transaction is committing and is holding any predicate locks,
* it must be added to a list of completed serializable transactions still
* holding locks.
+ *
+ * If isReadOnlySafe is true, then predicate locks are being released before
+ * the end of the transaction because MySerializableXact has been determined
+ * to be RO_SAFE. In non-parallel mode we can release it completely, but it
+ * in parallel mode we partially release the SERIALIZABLEXACT and keep it
+ * around until the end of the transaction, allowing each backend to clear its
+ * MySerializableXact variable and benefit from the optimization in its own
+ * time.
*/
void
-ReleasePredicateLocks(bool isCommit)
+ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
{
bool needToClear;
RWConflict conflict,
/*
* We can't trust XactReadOnly here, because a transaction which started
* as READ WRITE can show as READ ONLY later, e.g., within
- * substransactions. We want to flag a transaction as READ ONLY if it
+ * subtransactions. We want to flag a transaction as READ ONLY if it
* commits without writing so that de facto READ ONLY transactions get the
* benefit of some RO optimizations, so we will use this local variable to
* get some cleanup logic right which is based on whether the transaction
*/
bool topLevelIsDeclaredReadOnly;
+ /* We can't be both committing and releasing early due to RO_SAFE. */
+ Assert(!(isCommit && isReadOnlySafe));
+
+ /* Are we at the end of a transaction, that is, a commit or abort? */
+ if (!isReadOnlySafe)
+ {
+ /*
+ * Parallel workers mustn't release predicate locks at the end of
+ * their transaction. The leader will do that at the end of its
+ * transaction.
+ */
+ if (IsParallelWorker())
+ {
+ ReleasePredicateLocksLocal();
+ return;
+ }
+
+ /*
+ * By the time the leader in a parallel query reaches end of
+ * transaction, it has waited for all workers to exit.
+ */
+ Assert(!ParallelContextActive());
+
+ /*
+ * If the leader in a parallel query earlier stashed a partially
+ * released SERIALIZABLEXACT for final clean-up at end of transaction
+ * (because workers might still have been accessing it), then it's
+ * time to restore it.
+ */
+ if (SavedSerializableXact != InvalidSerializableXact)
+ {
+ Assert(MySerializableXact == InvalidSerializableXact);
+ MySerializableXact = SavedSerializableXact;
+ SavedSerializableXact = InvalidSerializableXact;
+ Assert(SxactIsPartiallyReleased(MySerializableXact));
+ }
+ }
+
if (MySerializableXact == InvalidSerializableXact)
{
Assert(LocalPredicateLockHash == NULL);
return;
}
+ LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
+
+ /*
+ * If the transaction is committing, but it has been partially released
+ * already, then treat this as a roll back. It was marked as rolled back.
+ */
+ if (isCommit && SxactIsPartiallyReleased(MySerializableXact))
+ isCommit = false;
+
+ /*
+ * If we're called in the middle of a transaction because we discovered
+ * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
+ * it (that is, release the predicate locks and conflicts, but not the
+ * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
+ */
+ if (isReadOnlySafe && IsInParallelMode())
+ {
+ /*
+ * The leader needs to stash a pointer to it, so that it can
+ * completely release it at end-of-transaction.
+ */
+ if (!IsParallelWorker())
+ SavedSerializableXact = MySerializableXact;
+
+ /*
+ * The first backend to reach this condition will partially release
+ * the SERIALIZABLEXACT. All others will just clear their
+ * backend-local state so that they stop doing SSI checks for the rest
+ * of the transaction.
+ */
+ if (SxactIsPartiallyReleased(MySerializableXact))
+ {
+ LWLockRelease(SerializableXactHashLock);
+ ReleasePredicateLocksLocal();
+ return;
+ }
+ else
+ {
+ MySerializableXact->flags |= SXACT_FLAG_PARTIALLY_RELEASED;
+ /* ... and proceed to perform the partial release below. */
+ }
+ }
Assert(!isCommit || SxactIsPrepared(MySerializableXact));
Assert(!isCommit || !SxactIsDoomed(MySerializableXact));
Assert(!SxactIsCommitted(MySerializableXact));
- Assert(!SxactIsRolledBack(MySerializableXact));
+ Assert(SxactIsPartiallyReleased(MySerializableXact)
+ || !SxactIsRolledBack(MySerializableXact));
/* may not be serializable during COMMIT/ROLLBACK PREPARED */
- if (MySerializableXact->pid != 0)
- Assert(IsolationIsSerializable());
+ Assert(MySerializableXact->pid == 0 || IsolationIsSerializable());
/* We'd better not already be on the cleanup list. */
Assert(!SxactIsOnFinishedList(MySerializableXact));
topLevelIsDeclaredReadOnly = SxactIsReadOnly(MySerializableXact);
- LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
-
/*
* We don't hold XidGenLock lock here, assuming that TransactionId is
* atomic!
*
* If this value is changing, we don't care that much whether we get the
* old or new value -- it is just used to determine how far
- * GlobalSerizableXmin must advance before this transaction can be fully
- * cleaned up. The worst that could happen is we wait for one more
+ * GlobalSerializableXmin must advance before this transaction can be
+ * fully cleaned up. The worst that could happen is we wait for one more
* transaction to complete before freeing some RAM; correctness of visible
* behavior is not affected.
*/
- MySerializableXact->finishedBefore = ShmemVariableCache->nextXid;
+ MySerializableXact->finishedBefore = XidFromFullTransactionId(ShmemVariableCache->nextFullXid);
/*
- * If it's not a commit it's a rollback, and we can clear our locks
- * immediately.
+ * If it's not a commit it's either a rollback or a read-only transaction
+ * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
*/
if (isCommit)
{
}
/*
- * Release all outConflicts to committed transactions. If we're rolling
+ * Release all outConflicts to committed transactions. If we're rolling
* back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
* previously committed transactions.
*/
SHMQueueInsertBefore(FinishedSerializableTransactions,
&MySerializableXact->finishedLink);
+ /*
+ * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
+ * partially release it. That's necessary because other backends may have
+ * a reference to it. The leader will release the SERIALIZABLEXACT itself
+ * at the end of the transaction after workers have stopped running.
+ */
if (!isCommit)
- ReleaseOneSerializableXact(MySerializableXact, false, false);
+ ReleaseOneSerializableXact(MySerializableXact,
+ isReadOnlySafe && IsInParallelMode(),
+ false);
LWLockRelease(SerializableFinishedListLock);
if (needToClear)
ClearOldPredicateLocks();
+ ReleasePredicateLocksLocal();
+}
+
+static void
+ReleasePredicateLocksLocal(void)
+{
MySerializableXact = InvalidSerializableXact;
MyXactDidWrite = false;
LWLockAcquire(SerializableXactHashLock, LW_SHARED);
}
else if (finishedSxact->commitSeqNo > PredXact->HavePartialClearedThrough
- && finishedSxact->commitSeqNo <= PredXact->CanPartialClearThrough)
+ && finishedSxact->commitSeqNo <= PredXact->CanPartialClearThrough)
{
/*
* Any active transactions that took their snapshot before this
PREDICATELOCKTARGET *target;
PREDICATELOCKTARGETTAG targettag;
uint32 targettaghash;
- LWLockId partitionLock;
+ LWLock *partitionLock;
tag = predlock->tag;
target = tag.myTarget;
SHMQueueDelete(&(predlock->xactLink));
hash_search_with_hash_value(PredicateLockHash, &tag,
- PredicateLockHashCodeFromTargetHashCode(&tag,
- targettaghash),
+ PredicateLockHashCodeFromTargetHashCode(&tag,
+ targettaghash),
HASH_REMOVE, NULL);
RemoveTargetIfNoLongerUsed(target, targettaghash);
* matter -- but keep the transaction entry itself and any outConflicts.
*
* When the summarize flag is set, we've run short of room for sxact data
- * and must summarize to the SLRU. Predicate locks are transferred to a
+ * and must summarize to the SLRU. Predicate locks are transferred to a
* dummy "old" transaction, with duplicate locks on a single target
* collapsing to a single lock with the "latest" commitSeqNo from among
* the conflicting locks..
* them to OldCommittedSxact if summarize is true)
*/
LWLockAcquire(SerializablePredicateLockListLock, LW_SHARED);
+ if (IsInParallelMode())
+ LWLockAcquire(&sxact->predicateLockListLock, LW_EXCLUSIVE);
predlock = (PREDICATELOCK *)
SHMQueueNext(&(sxact->predicateLocks),
&(sxact->predicateLocks),
PREDICATELOCKTARGET *target;
PREDICATELOCKTARGETTAG targettag;
uint32 targettaghash;
- LWLockId partitionLock;
+ LWLock *partitionLock;
nextpredlock = (PREDICATELOCK *)
SHMQueueNext(&(sxact->predicateLocks),
SHMQueueDelete(targetLink);
hash_search_with_hash_value(PredicateLockHash, &tag,
- PredicateLockHashCodeFromTargetHashCode(&tag,
- targettaghash),
+ PredicateLockHashCodeFromTargetHashCode(&tag,
+ targettaghash),
HASH_REMOVE, NULL);
if (summarize)
{
/* Fold into dummy transaction list. */
tag.myXact = OldCommittedSxact;
predlock = hash_search_with_hash_value(PredicateLockHash, &tag,
- PredicateLockHashCodeFromTargetHashCode(&tag,
- targettaghash),
+ PredicateLockHashCodeFromTargetHashCode(&tag,
+ targettaghash),
HASH_ENTER_NULL, &found);
if (!predlock)
ereport(ERROR,
*/
SHMQueueInit(&sxact->predicateLocks);
+ if (IsInParallelMode())
+ LWLockRelease(&sxact->predicateLockListLock);
LWLockRelease(SerializablePredicateLockListLock);
sxidtag.xid = sxact->topXid;
/*
* CheckForSerializableConflictOut
* We are reading a tuple which has been modified. If it is visible to
- * us but has been deleted, that indicates a rw-conflict out. If it's
+ * us but has been deleted, that indicates a rw-conflict out. If it's
* not visible and was created by a concurrent (overlapping)
* serializable transaction, that is also a rw-conflict out,
*
Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
/*
- * Find top level xid. Bail out if xid is too early to be a conflict, or
+ * Find top level xid. Bail out if xid is too early to be a conflict, or
* if it's our own xid.
*/
if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
errmsg("could not serialize access due to read/write dependencies among transactions"),
errdetail_internal("Reason code: Canceled on conflict out to old pivot %u.", xid),
- errhint("The transaction might succeed if retried.")));
+ errhint("The transaction might succeed if retried.")));
if (SxactHasSummaryConflictIn(MySerializableXact)
|| !SHMQueueEmpty(&MySerializableXact->inConflicts))
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
errmsg("could not serialize access due to read/write dependencies among transactions"),
errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
- errhint("The transaction might succeed if retried.")));
+ errhint("The transaction might succeed if retried.")));
MySerializableXact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
}
/*
* We have a conflict out to a transaction which has a conflict out to a
- * summarized transaction. That summarized transaction must have
+ * summarized transaction. That summarized transaction must have
* committed first, and we can't tell when it committed in relation to our
* snapshot acquisition, so something needs to be canceled.
*/
&& (!SxactHasConflictOut(sxact)
|| MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
{
- /* Read-only transaction will appear to run first. No conflict. */
+ /* Read-only transaction will appear to run first. No conflict. */
LWLockRelease(SerializableXactHashLock);
return;
}
CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
{
uint32 targettaghash;
- LWLockId partitionLock;
+ LWLock *partitionLock;
PREDICATELOCKTARGET *target;
PREDICATELOCK *predlock;
PREDICATELOCK *mypredlock = NULL;
/*
* If we found one of our own SIREAD locks to remove, remove it now.
*
- * At this point our transaction already has an ExclusiveRowLock on the
+ * At this point our transaction already has a RowExclusiveLock on the
* relation, so we are OK to drop the predicate lock on the tuple, if
* found, without fearing that another write against the tuple will occur
* before the MVCC information makes it to the buffer.
PREDICATELOCK *rmpredlock;
LWLockAcquire(SerializablePredicateLockListLock, LW_SHARED);
+ if (IsInParallelMode())
+ LWLockAcquire(&MySerializableXact->predicateLockListLock, LW_EXCLUSIVE);
LWLockAcquire(partitionLock, LW_EXCLUSIVE);
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
LWLockRelease(SerializableXactHashLock);
LWLockRelease(partitionLock);
+ if (IsInParallelMode())
+ LWLockRelease(&MySerializableXact->predicateLockListLock);
LWLockRelease(SerializablePredicateLockListLock);
if (rmpredlock != NULL)
SET_PREDICATELOCKTARGETTAG_TUPLE(targettag,
relation->rd_node.dbNode,
relation->rd_id,
- ItemPointerGetBlockNumber(&(tuple->t_self)),
- ItemPointerGetOffsetNumber(&(tuple->t_self)));
+ ItemPointerGetBlockNumber(&(tuple->t_self)),
+ ItemPointerGetOffsetNumber(&(tuple->t_self)));
CheckTargetForConflictsIn(&targettag);
}
LWLockAcquire(SerializablePredicateLockListLock, LW_EXCLUSIVE);
for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
- LWLockAcquire(FirstPredicateLockMgrLock + i, LW_SHARED);
- LWLockAcquire(SerializableXactHashLock, LW_SHARED);
+ LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED);
+ LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
/* Scan through target list */
hash_seq_init(&seqstat, PredicateLockTargetHash);
offsetof(PREDICATELOCK, targetLink));
if (predlock->tag.myXact != MySerializableXact
- && !RWConflictExists(predlock->tag.myXact, MySerializableXact))
+ && !RWConflictExists(predlock->tag.myXact, MySerializableXact))
{
FlagRWConflict(predlock->tag.myXact, MySerializableXact);
}
/* Release locks in reverse order */
LWLockRelease(SerializableXactHashLock);
for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
- LWLockRelease(FirstPredicateLockMgrLock + i);
+ LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
LWLockRelease(SerializablePredicateLockListLock);
}
*------------------------------------------------------------------------
*/
if (SxactIsCommitted(writer)
- && (SxactHasConflictOut(writer) || SxactHasSummaryConflictOut(writer)))
+ && (SxactHasConflictOut(writer) || SxactHasSummaryConflictOut(writer)))
failure = true;
/*------------------------------------------------------------------------
&& (!SxactIsCommitted(writer)
|| t2->prepareSeqNo <= writer->commitSeqNo)
&& (!SxactIsReadOnly(reader)
- || t2->prepareSeqNo <= reader->SeqNo.lastCommitBeforeSnapshot))
+ || t2->prepareSeqNo <= reader->SeqNo.lastCommitBeforeSnapshot))
{
failure = true;
break;
&& (!SxactIsCommitted(t0)
|| t0->commitSeqNo >= writer->prepareSeqNo)
&& (!SxactIsReadOnly(t0)
- || t0->SeqNo.lastCommitBeforeSnapshot >= writer->prepareSeqNo))
+ || t0->SeqNo.lastCommitBeforeSnapshot >= writer->prepareSeqNo))
{
failure = true;
break;
*
* If a dangerous structure is found, the pivot (the near conflict) is
* marked for death, because rolling back another transaction might mean
- * that we flail without ever making progress. This transaction is
+ * that we flail without ever making progress. This transaction is
* committing writes, so letting it commit ensures progress. If we
* canceled the far conflict, it might immediately fail again on retry.
*/
/* Check if someone else has already decided that we need to die */
if (SxactIsDoomed(MySerializableXact))
{
+ Assert(!SxactIsPartiallyReleased(MySerializableXact));
LWLockRelease(SerializableXactHashLock);
ereport(ERROR,
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
if (MySerializableXact == InvalidSerializableXact)
return;
- /* Generate a xact record for our SERIALIZABLEXACT */
+ /* Generate an xact record for our SERIALIZABLEXACT */
record.type = TWOPHASEPREDICATERECORD_XACT;
xactRecord->xmin = MySerializableXact->xmin;
xactRecord->flags = MySerializableXact->flags;
*/
LWLockAcquire(SerializablePredicateLockListLock, LW_SHARED);
+ /*
+ * No need to take sxact->predicateLockListLock in parallel mode because
+ * there cannot be any parallel workers running while we are preparing a
+ * transaction.
+ */
+ Assert(!IsParallelWorker() && !ParallelContextActive());
+
predlock = (PREDICATELOCK *)
SHMQueueNext(&(sxact->predicateLocks),
&(sxact->predicateLocks),
MySerializableXact = sxid->myXact;
MyXactDidWrite = true; /* conservatively assume that we wrote
* something */
- ReleasePredicateLocks(isCommit);
+ ReleasePredicateLocks(isCommit, false);
}
/*
CreatePredicateLock(&lockRecord->target, targettaghash, sxact);
}
}
+
+/*
+ * Prepare to share the current SERIALIZABLEXACT with parallel workers.
+ * Return a handle object that can be used by AttachSerializableXact() in a
+ * parallel worker.
+ */
+SerializableXactHandle
+ShareSerializableXact(void)
+{
+ return MySerializableXact;
+}
+
+/*
+ * Allow parallel workers to import the leader's SERIALIZABLEXACT.
+ */
+void
+AttachSerializableXact(SerializableXactHandle handle)
+{
+
+ Assert(MySerializableXact == InvalidSerializableXact);
+
+ MySerializableXact = (SERIALIZABLEXACT *) handle;
+ if (MySerializableXact != InvalidSerializableXact)
+ CreateLocalPredicateLockHash();
+}
projects / postgresql / blobdiff