* varsup.c
* postgres OID & XID variables support routines
*
- * Copyright (c) 2000-2018, PostgreSQL Global Development Group
+ * Copyright (c) 2000-2019, PostgreSQL Global Development Group
*
* IDENTIFICATION
* src/backend/access/transam/varsup.c
/*
- * Allocate the next XID for a new transaction or subtransaction.
+ * Allocate the next FullTransactionId for a new transaction or
+ * subtransaction.
*
* The new XID is also stored into MyPgXact before returning.
*
* does something. So it is safe to do a database lookup if we want to
* issue a warning about XID wrap.
*/
-TransactionId
+FullTransactionId
GetNewTransactionId(bool isSubXact)
{
+ FullTransactionId full_xid;
TransactionId xid;
/*
{
Assert(!isSubXact);
MyPgXact->xid = BootstrapTransactionId;
- return BootstrapTransactionId;
+ return FullTransactionIdFromEpochAndXid(0, BootstrapTransactionId);
}
/* safety check, we should never get this far in a HS standby */
LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
- xid = ShmemVariableCache->nextXid;
+ full_xid = ShmemVariableCache->nextFullXid;
+ xid = XidFromFullTransactionId(full_xid);
/*----------
* Check to see if it's safe to assign another XID. This protects against
/* Re-acquire lock and start over */
LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
- xid = ShmemVariableCache->nextXid;
+ full_xid = ShmemVariableCache->nextFullXid;
+ xid = XidFromFullTransactionId(full_xid);
}
/*
ExtendSUBTRANS(xid);
/*
- * Now advance the nextXid counter. This must not happen until after we
- * have successfully completed ExtendCLOG() --- if that routine fails, we
- * want the next incoming transaction to try it again. We cannot assign
- * more XIDs until there is CLOG space for them.
+ * Now advance the nextFullXid counter. This must not happen until after
+ * we have successfully completed ExtendCLOG() --- if that routine fails,
+ * we want the next incoming transaction to try it again. We cannot
+ * assign more XIDs until there is CLOG space for them.
*/
- TransactionIdAdvance(ShmemVariableCache->nextXid);
+ FullTransactionIdAdvance(&ShmemVariableCache->nextFullXid);
/*
* We must store the new XID into the shared ProcArray before releasing
* latestCompletedXid is present in the ProcArray, which is essential for
* correct OldestXmin tracking; see src/backend/access/transam/README.
*
- * XXX by storing xid into MyPgXact without acquiring ProcArrayLock, we
- * are relying on fetch/store of an xid to be atomic, else other backends
- * might see a partially-set xid here. But holding both locks at once
- * would be a nasty concurrency hit. So for now, assume atomicity.
- *
* Note that readers of PGXACT xid fields should be careful to fetch the
* value only once, rather than assume they can read a value multiple
- * times and get the same answer each time.
+ * times and get the same answer each time. Note we are assuming that
+ * TransactionId and int fetch/store are atomic.
*
* The same comments apply to the subxact xid count and overflow fields.
*
- * A solution to the atomic-store problem would be to give each PGXACT its
- * own spinlock used only for fetching/storing that PGXACT's xid and
- * related fields.
+ * Use of a write barrier prevents dangerous code rearrangement in this
+ * function; other backends could otherwise e.g. be examining my subxids
+ * info concurrently, and we don't want them to see an invalid
+ * intermediate state, such as an incremented nxids before the array entry
+ * is filled.
+ *
+ * Other processes that read nxids should do so before reading xids
+ * elements with a pg_read_barrier() in between, so that they can be sure
+ * not to read an uninitialized array element; see
+ * src/backend/storage/lmgr/README.barrier.
*
* If there's no room to fit a subtransaction XID into PGPROC, set the
* cache-overflowed flag instead. This forces readers to look in
* window *will* include the parent XID, so they will deliver the correct
* answer later on when someone does have a reason to inquire.)
*/
+ if (!isSubXact)
+ MyPgXact->xid = xid; /* LWLockRelease acts as barrier */
+ else
{
- /*
- * Use volatile pointer to prevent code rearrangement; other backends
- * could be examining my subxids info concurrently, and we don't want
- * them to see an invalid intermediate state, such as incrementing
- * nxids before filling the array entry. Note we are assuming that
- * TransactionId and int fetch/store are atomic.
- */
- volatile PGPROC *myproc = MyProc;
- volatile PGXACT *mypgxact = MyPgXact;
+ int nxids = MyPgXact->nxids;
- if (!isSubXact)
- mypgxact->xid = xid;
- else
+ if (nxids < PGPROC_MAX_CACHED_SUBXIDS)
{
- int nxids = mypgxact->nxids;
-
- if (nxids < PGPROC_MAX_CACHED_SUBXIDS)
- {
- myproc->subxids.xids[nxids] = xid;
- mypgxact->nxids = nxids + 1;
- }
- else
- mypgxact->overflowed = true;
+ MyProc->subxids.xids[nxids] = xid;
+ pg_write_barrier();
+ MyPgXact->nxids = nxids + 1;
}
+ else
+ MyPgXact->overflowed = true;
}
LWLockRelease(XidGenLock);
- return xid;
+ return full_xid;
}
/*
- * Read nextXid but don't allocate it.
+ * Read nextFullXid but don't allocate it.
*/
-TransactionId
-ReadNewTransactionId(void)
+FullTransactionId
+ReadNextFullTransactionId(void)
{
- TransactionId xid;
+ FullTransactionId fullXid;
LWLockAcquire(XidGenLock, LW_SHARED);
- xid = ShmemVariableCache->nextXid;
+ fullXid = ShmemVariableCache->nextFullXid;
LWLockRelease(XidGenLock);
- return xid;
+ return fullXid;
+}
+
+/*
+ * Advance nextFullXid to the value after a given xid. The epoch is inferred.
+ * This must only be called during recovery or from two-phase start-up code.
+ */
+void
+AdvanceNextFullTransactionIdPastXid(TransactionId xid)
+{
+ FullTransactionId newNextFullXid;
+ TransactionId next_xid;
+ uint32 epoch;
+
+ /*
+ * It is safe to read nextFullXid without a lock, because this is only
+ * called from the startup process or single-process mode, meaning that no
+ * other process can modify it.
+ */
+ Assert(AmStartupProcess() || !IsUnderPostmaster);
+
+ /* Fast return if this isn't an xid high enough to move the needle. */
+ next_xid = XidFromFullTransactionId(ShmemVariableCache->nextFullXid);
+ if (!TransactionIdFollowsOrEquals(xid, next_xid))
+ return;
+
+ /*
+ * Compute the FullTransactionId that comes after the given xid. To do
+ * this, we preserve the existing epoch, but detect when we've wrapped
+ * into a new epoch. This is necessary because WAL records and 2PC state
+ * currently contain 32 bit xids. The wrap logic is safe in those cases
+ * because the span of active xids cannot exceed one epoch at any given
+ * point in the WAL stream.
+ */
+ TransactionIdAdvance(xid);
+ epoch = EpochFromFullTransactionId(ShmemVariableCache->nextFullXid);
+ if (unlikely(xid < next_xid))
+ ++epoch;
+ newNextFullXid = FullTransactionIdFromEpochAndXid(epoch, xid);
+
+ /*
+ * We still need to take a lock to modify the value when there are
+ * concurrent readers.
+ */
+ LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
+ ShmemVariableCache->nextFullXid = newNextFullXid;
+ LWLockRelease(XidGenLock);
}
/*
ShmemVariableCache->xidStopLimit = xidStopLimit;
ShmemVariableCache->xidWrapLimit = xidWrapLimit;
ShmemVariableCache->oldestXidDB = oldest_datoid;
- curXid = ShmemVariableCache->nextXid;
+ curXid = XidFromFullTransactionId(ShmemVariableCache->nextFullXid);
LWLockRelease(XidGenLock);
/* Log the info */
/* Locking is probably not really necessary, but let's be careful */
LWLockAcquire(XidGenLock, LW_SHARED);
- nextXid = ShmemVariableCache->nextXid;
+ nextXid = XidFromFullTransactionId(ShmemVariableCache->nextFullXid);
xidVacLimit = ShmemVariableCache->xidVacLimit;
oldestXid = ShmemVariableCache->oldestXid;
oldestXidDB = ShmemVariableCache->oldestXidDB;
if (!TransactionIdIsValid(xidVacLimit))
return true; /* this shouldn't happen anymore either */
if (TransactionIdFollowsOrEquals(nextXid, xidVacLimit))
- return true; /* past VacLimit, don't delay updating */
+ return true; /* past xidVacLimit, don't delay updating */
if (!SearchSysCacheExists1(DATABASEOID, ObjectIdGetDatum(oldestXidDB)))
return true; /* could happen, per comments above */
return false;
* OIDs are generated by a cluster-wide counter. Since they are only 32 bits
* wide, counter wraparound will occur eventually, and therefore it is unwise
* to assume they are unique unless precautions are taken to make them so.
- * Hence, this routine should generally not be used directly. The only
- * direct callers should be GetNewOid() and GetNewRelFileNode() in
+ * Hence, this routine should generally not be used directly. The only direct
+ * callers should be GetNewOidWithIndex() and GetNewRelFileNode() in
* catalog/catalog.c.
*/
Oid
* Check for wraparound of the OID counter. We *must* not return 0
* (InvalidOid), and in normal operation we mustn't return anything below
* FirstNormalObjectId since that range is reserved for initdb (see
- * IsCatalogClass()). Note we are relying on unsigned comparison.
+ * IsCatalogRelationOid()). Note we are relying on unsigned comparison.
*
* During initdb, we start the OID generator at FirstBootstrapObjectId, so
* we only wrap if before that point when in bootstrap or standalone mode.