1 背景

本文通过简单修改开源Postgresql源码，实现批量获取事务ID的功能，对比前后性能差异。

周末实验项目for fun，代码可以随意使用。

！！！注意：修改会带来的并发问题会造成数据不一致，ProcArray和快照的逻辑很多都是在XID严格递增的情况下设计的，修改后的xid空洞、跳变需要很大的修改量来适配。

2 改造前

（性能数据没有太大参考意义，只用于前后对比）

16C小规格测试机128并发压测，PG参数全部异步写，瓶颈来到事务ID生成

128并发压测只写120秒XidGen锁每秒的出现数量：均值在60左右，QPS = 80589

-- 参数
fsync = off
synchronous_commit = off
autovacuum = offcreate table testbl1(c1 int, c2 int, c3 int, c4 text, c5 text);
-- in.sql
insert into testbl1 values (12,123,456,'avzdsqerqwadsf','asdfgerrerg');pgbench -c 128 -j 128 -n -r -P 1 -T 120 -f ./in.sql
for i in {1..60};do psql -c "select count(*) from pg_stat_activity where wait_event='XidGen'" -A -t; sleep 1;done;0
12
100
41
0
50
45
64
94
98
97
27
...
...

3 改造方案

由于是实验项目，改造会造成逻辑复制等代码crash，这里不关注。

3.1 改造方案一

【本地进程】拿事务ID从一次拿一个变成一次拿N个，其他不变。

关键改造点：

• GetNewTransactionId：预存本地N个事务ID，取的时候先取本地，再去共享的。
• ExtendClog：clog页面的原生扩展机制是严格按顺序递增的，需要改造。
• GetSnapshotData：要求事务ID必须严格递增，这里可能会有空洞触发assert。
• ProcArrayEndTransactionInternal：并发问题，PGPROC的xids数组数据错乱。

3.2 改造方案二（较复杂不做测试）

拿事务ID由每个进程自己拿，变成由一个进程统一分配。

4 最终效果（一批拿5个xid、一批拿64个xid）

结论：QPS有略微提升（和环境关系比较大，CPU共享性能很差）

QPS对比

• 优化前：80589
• 优化后：84923

【一批拿5个xid】 vs 【一次拿1个xid】xidgen锁事件对比

xidgen明显下降，瓶颈点打散到ProcArrayGroupUpdate、XactGroupUpdate等

【一批拿64个xid】 vs 【一次拿1个xid】xidgen锁事件对比

观测不到xidgen，瓶颈点打散到ProcArrayGroupUpdate、XactGroupUpdate等

部分代码

FullTransactionId localTransactionId = {0};
int localTransactionIdCnt = 0;FullTransactionId
GetNewTransactionId(bool isSubXact)
{FullTransactionId full_xid;TransactionId xid;/** Workers synchronize transaction state at the beginning of each parallel* operation, so we can't account for new XIDs after that point.*/if (IsInParallelMode())elog(ERROR, "cannot assign TransactionIds during a parallel operation");/** During bootstrap initialization, we return the special bootstrap* transaction id.*/if (IsBootstrapProcessingMode()){Assert(!isSubXact);MyProc->xid = BootstrapTransactionId;ProcGlobal->xids[MyProc->pgxactoff] = BootstrapTransactionId;return FullTransactionIdFromEpochAndXid(0, BootstrapTransactionId);}/* safety check, we should never get this far in a HS standby */if (RecoveryInProgress())elog(ERROR, "cannot assign TransactionIds during recovery");bool needlock = false;if (localTransactionIdCnt > 0){// LWLockAcquire(XidGenLock, LW_EXCLUSIVE);Assert(localTransactionId.value > 0);full_xid = localTransactionId;xid = XidFromFullTransactionId(full_xid);FullTransactionIdAdvance(&localTransactionId);localTransactionIdCnt--;needlock = false;}else{FullTransactionId prevTransactionId = {0};TransactionId prevXid;LWLockAcquire(XidGenLock, LW_EXCLUSIVE);needlock = true;// [1] get 1000, use 1000localTransactionId = full_xid = ShmemVariableCache->nextXid;xid = XidFromFullTransactionId(full_xid);// [2] move local to 1001FullTransactionIdAdvance(&localTransactionId);// [3] move share to 1001FullTransactionIdAdvance(&ShmemVariableCache->nextXid);for (int i = 0; i < 5; i++){prevTransactionId = ShmemVariableCache->nextXid;// [4] move share to 1006 (1006 for others!)FullTransactionIdAdvance(&ShmemVariableCache->nextXid);// [5] cnt == 5 (local: 1001 1002 1003 1004 1005)localTransactionIdCnt++;}// [6] extend once to 1005prevXid = XidFromFullTransactionId(prevTransactionId);ExtendCLOG(prevXid);ExtendCommitTs(prevXid);ExtendSUBTRANS(prevXid);}Assert(localTransactionIdCnt >= 0);if (!isSubXact){Assert(ProcGlobal->subxidStates[MyProc->pgxactoff].count == 0);Assert(!ProcGlobal->subxidStates[MyProc->pgxactoff].overflowed);Assert(MyProc->subxidStatus.count == 0);Assert(!MyProc->subxidStatus.overflowed);/* LWLockRelease acts as barrier */MyProc->xid = xid;ProcGlobal->xids[MyProc->pgxactoff] = xid;}else{XidCacheStatus *substat = &ProcGlobal->subxidStates[MyProc->pgxactoff];int			nxids = MyProc->subxidStatus.count;Assert(substat->count == MyProc->subxidStatus.count);Assert(substat->overflowed == MyProc->subxidStatus.overflowed);if (nxids < PGPROC_MAX_CACHED_SUBXIDS){MyProc->subxids.xids[nxids] = xid;pg_write_barrier();MyProc->subxidStatus.count = substat->count = nxids + 1;}elseMyProc->subxidStatus.overflowed = substat->overflowed = true;}if (needlock)LWLockRelease(XidGenLock);// elog(WARNING, "[%ld](%d)->[%ld]", localTransactionId.value, localTransactionIdCnt, full_xid.value);return full_xid;
}#define CLOG_MAX_PAGES (UINT_MAX / CLOG_XACTS_PER_PAGE) // 131071
bool ClogPageMark[CLOG_MAX_PAGES] = {false};void
ExtendCLOG(TransactionId newestXact)
{int			pageno;/** No work except at first XID of a page.  But beware: just after* wraparound, the first XID of page zero is FirstNormalTransactionId.*/// if (TransactionIdToPgIndex(newestXact) != 0 &&// 	!TransactionIdEquals(newestXact, FirstNormalTransactionId))// 	return;if (ClogPageMark[TransactionIdToPage(newestXact)])return;pageno = TransactionIdToPage(newestXact);LWLockAcquire(XactSLRULock, LW_EXCLUSIVE);/* Zero the page and make an XLOG entry about it */ZeroCLOGPage(pageno, true);LWLockRelease(XactSLRULock);ClogPageMark[TransactionIdToPage(newestXact)] = true;
}