From 80787aee44b7f9b5389476e45f2241073b6c89ee Mon Sep 17 00:00:00 2001 From: Sokolov Yura Date: Mon, 10 Jul 2017 12:34:48 +0000 Subject: [PATCH] Make hash table for xip in XidInMVCCSnapshot When lot of concurrent transactions attempts to update single row, then linear scan through running list in XidInMVCCSnapshot became noticebale bottleneck. With this change, hash table is built on first search of xid in snapshot->xip and snapshot->subxip arrays. If size of array is smaller than 60, then linear scan is still used, cause there is no much benefits from building hash then. (at least on Intel Xeon). --- src/backend/storage/ipc/procarray.c | 64 +++++++++++++++++++++++------ src/backend/utils/time/snapmgr.c | 24 +++++++---- src/backend/utils/time/tqual.c | 81 +++++++++++++++++++++++++++---------- src/include/utils/snapshot.h | 11 +++++ 4 files changed, 138 insertions(+), 42 deletions(-) diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index a7e8cf2..18cc233 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -1469,6 +1469,49 @@ GetMaxSnapshotSubxidCount(void) } /* + * ExtendXipSizeForHash - calculate xip array size with space for hash table + */ +int +ExtendXipSizeForHash(int xipcnt, uint8* plog) +{ + int size; + uint8 log = 0; + + size = xipcnt; + if (xipcnt >= SnapshotMinHash) + { + log = 1; + while (xipcnt) { + log++; + xipcnt >>= 1; + } + size += 1<xip = (TransactionId *) - malloc(GetMaxSnapshotXidCount() * sizeof(TransactionId)); - if (snapshot->xip == NULL) - ereport(ERROR, - (errcode(ERRCODE_OUT_OF_MEMORY), - errmsg("out of memory"))); - Assert(snapshot->subxip == NULL); - snapshot->subxip = (TransactionId *) - malloc(GetMaxSnapshotSubxidCount() * sizeof(TransactionId)); - if (snapshot->subxip == NULL) - ereport(ERROR, - (errcode(ERRCODE_OUT_OF_MEMORY), - errmsg("out of memory"))); + snapshot->xip = AllocateXip(GetMaxSnapshotXidCount(), + &snapshot->xhlog); + snapshot->subxip = AllocateXip(GetMaxSnapshotSubxidCount(), + &snapshot->subxhlog); } /* @@ -1757,6 +1791,10 @@ GetSnapshotData(Snapshot snapshot) snapshot->active_count = 0; snapshot->regd_count = 0; snapshot->copied = false; + if (snapshot->xhlog != 0) + snapshot->xhlog |= SnapshotHashNotBuilt; + if (snapshot->subxhlog != 0) + snapshot->subxhlog |= SnapshotHashNotBuilt; if (old_snapshot_threshold < 0) { diff --git a/src/backend/utils/time/snapmgr.c b/src/backend/utils/time/snapmgr.c index 08a08c8..56c1583 100644 --- a/src/backend/utils/time/snapmgr.c +++ b/src/backend/utils/time/snapmgr.c @@ -662,13 +662,16 @@ CopySnapshot(Snapshot snapshot) Snapshot newsnap; Size subxipoff; Size size; + int xcnt, subxcnt; + uint8 xhlog, subxhlog; Assert(snapshot != InvalidSnapshot); + xcnt = ExtendXipSizeForHash(snapshot->xcnt, &xhlog); + subxcnt = ExtendXipSizeForHash(snapshot->subxcnt, &subxhlog); /* We allocate any XID arrays needed in the same palloc block. */ - size = subxipoff = sizeof(SnapshotData) + - snapshot->xcnt * sizeof(TransactionId); - if (snapshot->subxcnt > 0) + size = subxipoff = sizeof(SnapshotData) + xcnt * sizeof(TransactionId); + if (subxcnt > 0) size += snapshot->subxcnt * sizeof(TransactionId); newsnap = (Snapshot) MemoryContextAlloc(TopTransactionContext, size); @@ -677,6 +680,8 @@ CopySnapshot(Snapshot snapshot) newsnap->regd_count = 0; newsnap->active_count = 0; newsnap->copied = true; + newsnap->xhlog = xhlog ? xhlog | SnapshotHashNotBuilt : 0; + newsnap->subxhlog = subxhlog ? subxhlog | SnapshotHashNotBuilt : 0; /* setup XID array */ if (snapshot->xcnt > 0) @@ -2124,16 +2129,18 @@ RestoreSnapshot(char *start_address) Size size; Snapshot snapshot; TransactionId *serialized_xids; + int xcnt, subxcnt; + uint8 xhlog, subxhlog; memcpy(&serialized_snapshot, start_address, sizeof(SerializedSnapshotData)); serialized_xids = (TransactionId *) (start_address + sizeof(SerializedSnapshotData)); + xcnt = ExtendXipSizeForHash(serialized_snapshot.xcnt, &xhlog); + subxcnt = ExtendXipSizeForHash(serialized_snapshot.subxcnt, &subxhlog); /* We allocate any XID arrays needed in the same palloc block. */ - size = sizeof(SnapshotData) - + serialized_snapshot.xcnt * sizeof(TransactionId) - + serialized_snapshot.subxcnt * sizeof(TransactionId); + size = sizeof(SnapshotData) + (xcnt + subxcnt) * sizeof(TransactionId); /* Copy all required fields */ snapshot = (Snapshot) MemoryContextAlloc(TopTransactionContext, size); @@ -2142,8 +2149,10 @@ RestoreSnapshot(char *start_address) snapshot->xmax = serialized_snapshot.xmax; snapshot->xip = NULL; snapshot->xcnt = serialized_snapshot.xcnt; + snapshot->xhlog = xhlog ? xhlog | SnapshotHashNotBuilt : 0; snapshot->subxip = NULL; snapshot->subxcnt = serialized_snapshot.subxcnt; + snapshot->subxhlog = subxhlog ? subxhlog | SnapshotHashNotBuilt : 0; snapshot->suboverflowed = serialized_snapshot.suboverflowed; snapshot->takenDuringRecovery = serialized_snapshot.takenDuringRecovery; snapshot->curcid = serialized_snapshot.curcid; @@ -2161,8 +2170,7 @@ RestoreSnapshot(char *start_address) /* Copy SubXIDs, if present. */ if (serialized_snapshot.subxcnt > 0) { - snapshot->subxip = ((TransactionId *) (snapshot + 1)) + - serialized_snapshot.xcnt; + snapshot->subxip = ((TransactionId *) (snapshot + 1)) + xcnt; memcpy(snapshot->subxip, serialized_xids + serialized_snapshot.xcnt, serialized_snapshot.subxcnt * sizeof(TransactionId)); } diff --git a/src/backend/utils/time/tqual.c b/src/backend/utils/time/tqual.c index f9da9e1..1c53790 100644 --- a/src/backend/utils/time/tqual.c +++ b/src/backend/utils/time/tqual.c @@ -1450,6 +1450,59 @@ HeapTupleIsSurelyDead(HeapTuple htup, TransactionId OldestXmin) return TransactionIdPrecedes(HeapTupleHeaderGetRawXmax(tuple), OldestXmin); } +static bool +XidInXip(TransactionId xid, TransactionId *xip, uint32 xcnt, uint8 *xhlog) +{ + uint32 i; + uint32 j, k, d, mask; + TransactionId *xiphash; + if (!TransactionIdIsNormal(xid)) + return false; + if (xcnt < SnapshotMinHash || *xhlog == 0) + { + /* full scan for small snapshots and if xiphash is not allocated */ + for (i = 0; i < xcnt; i++) + if (TransactionIdEquals(xid, xip[i])) + return true; + return false; + } + mask = ((uint32)1 << (*xhlog & ~SnapshotHashNotBuilt)) - 1; + xiphash = xip + xcnt; + if (!(*xhlog & SnapshotHashNotBuilt)) + { + d = xid >> *xhlog; + j = k = xid & mask; + while (xiphash[j] != InvalidTransactionId) + { + if (TransactionIdEquals(xiphash[j], xid)) + return true; + j = (k + d) & mask; + d = d*5 + 1; + } + return false; + } + else + { + /* build hash */ + int found = 0; + *xhlog &= ~SnapshotHashNotBuilt; + memset(xiphash, 0, (mask+1) * sizeof(TransactionId)); + for (i = 0; i < xcnt; i++) + { + found |= TransactionIdEquals(xip[i], xid); + d = xip[i] >> *xhlog; + j = k = xip[i] & mask; + while (xiphash[j] != InvalidTransactionId) + { + j = (k + d) & mask; + d = d*5 + 1; + } + xiphash[j] = xip[i]; + } + return found != 0; + } +} + /* * XidInMVCCSnapshot * Is the given XID still-in-progress according to the snapshot? @@ -1463,8 +1516,6 @@ HeapTupleIsSurelyDead(HeapTuple htup, TransactionId OldestXmin) static bool XidInMVCCSnapshot(TransactionId xid, Snapshot snapshot) { - uint32 i; - /* * Make a quick range check to eliminate most XIDs without looking at the * xip arrays. Note that this is OK even if we convert a subxact XID to @@ -1496,13 +1547,8 @@ XidInMVCCSnapshot(TransactionId xid, Snapshot snapshot) if (!snapshot->suboverflowed) { /* we have full data, so search subxip */ - int32 j; - - for (j = 0; j < snapshot->subxcnt; j++) - { - if (TransactionIdEquals(xid, snapshot->subxip[j])) - return true; - } + if (XidInXip(xid, snapshot->subxip, snapshot->subxcnt, &snapshot->subxhlog)) + return true; /* not there, fall through to search xip[] */ } @@ -1523,16 +1569,12 @@ XidInMVCCSnapshot(TransactionId xid, Snapshot snapshot) return false; } - for (i = 0; i < snapshot->xcnt; i++) - { - if (TransactionIdEquals(xid, snapshot->xip[i])) - return true; - } + + if (XidInXip(xid, snapshot->xip, snapshot->xcnt, &snapshot->xhlog)) + return true; } else { - int32 j; - /* * In recovery we store all xids in the subxact array because it is by * far the bigger array, and we mostly don't know which xids are @@ -1562,11 +1604,8 @@ XidInMVCCSnapshot(TransactionId xid, Snapshot snapshot) * indeterminate xid. We don't know whether it's top level or subxact * but it doesn't matter. If it's present, the xid is visible. */ - for (j = 0; j < snapshot->subxcnt; j++) - { - if (TransactionIdEquals(xid, snapshot->subxip[j])) - return true; - } + if (XidInXip(xid, snapshot->subxip, snapshot->subxcnt, &snapshot->subxhlog)) + return true; } return false; diff --git a/src/include/utils/snapshot.h b/src/include/utils/snapshot.h index 074cc81..9054f16 100644 --- a/src/include/utils/snapshot.h +++ b/src/include/utils/snapshot.h @@ -23,6 +23,15 @@ typedef struct SnapshotData *Snapshot; #define InvalidSnapshot ((Snapshot) NULL) +/* + * Big running list will be converted into hash table on demand. + * SnapshotMinHash is threshold between "linear scan" and "hashtable on demand". + * SnapshotHashNotBuilt indicates that hash table not built yet. + */ +#define SnapshotMinHash (60) +#define SnapshotHashNotBuilt ((uint8)0x80) +/* Calculate size to hold both array and hash table (if needed) */ +int ExtendXipSizeForHash(int xipcnt, uint8* plog); /* * We use SnapshotData structures to represent both "regular" (MVCC) @@ -76,6 +85,7 @@ typedef struct SnapshotData */ TransactionId *xip; uint32 xcnt; /* # of xact ids in xip[] */ + uint8 xhlog; /* log2 of allocated xip hash part */ /* * For non-historic MVCC snapshots, this contains subxact IDs that are in @@ -92,6 +102,7 @@ typedef struct SnapshotData bool takenDuringRecovery; /* recovery-shaped snapshot? */ bool copied; /* false if it's a static snapshot */ + uint8 subxhlog; /* log2 of allocated subxip hash part */ CommandId curcid; /* in my xact, CID < curcid are visible */ -- 1.8.3.1