PostgreSQL备机是否做checkpoint
背景
我们知道为了缩短崩溃恢复时间，PostgreSQL支持checkpoint功能，即异常重启时，PostgreSQL以最近的checkpoint为起点，不断回放之后的WAL日志。PostgreSQL主机做checkpoint时，产生一条CHECKPOINT WAL日志记录。那么备机会做checkpoint吗？如果可以做checkpoint，它是否也会产生一条CHECKPOINT WAL日志记录并将其写入到WAL文件中？
实验
1、备机的日志位置
[yzs@bogon pg_wal]$ pg_waldump 00000001000000000000000C
...
rmgr: Transaction len (rec/tot): 405/ 405, tx: 722, lsn: 0/0C019E18, prev 0/0C019DEC, desc: COMMIT 2019-05-02 00:49:43.519052 PDT; inval msgs: catcache 74 catcache 73 catcache 74 catcache 73 catcache 50 catcache 49 catcache 7 catcache 6 catcache 7 catcache 6 catcache 7 catcache 6 catcache 7 catcache 6 catcache 7 catcache 6 catcache 7 catcache 6 catcache 7 catcache 6 snapshot 2608 relcache 16447
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C019FB0, prev 0/0C019E18, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C019FE4, prev 0/0C019FB0, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
rmgr: XLOG len (rec/tot): 102/ 102, tx: 0, lsn: 0/0C01A02C, prev 0/0C019FE4, desc: CHECKPOINT_ONLINE redo 0/C019FE4; tli 1; prev tli 1; fpw true; xid 0:723; oid 24639; multi 1; offset 0; oldest xid 548 in DB 1; oldest multi 1 in DB 1; oldest/newest commit timestamp xid: 0/0; oldest running xid 723; online
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C01A094, prev 0/0C01A02C, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
pg_waldump: FATAL: error in WAL record at 0/C01A094: invalid record length at 0/C01A0C8: wanted 24, got 0
2、备机pg_control文件记录的checkpoint位置
[yzs@bogon pg_wal]$ pg_controldata -D /home/yzs/data
pg_control version number: 1002
Catalog version number: 201707211
Database system identifier: 6602905782398358082
Database cluster state: in archive recovery
pg_control last modified: Thu 02 May 2019 12:58:24 AM PDT
Latest checkpoint location: 0/C01A02C
Prior checkpoint location: 0/C005470
Latest checkpoint's REDO location: 0/C019FE4
Latest checkpoint's REDO WAL file: 00000001000000000000000C
3、备机手动执行checkpoint
[yzs@bogon pg_wal]$ psql
psql (10.5)
Type "help" for help.
yzs=# checkpoint;
CHECKPOINT
4、备机日志文件
[yzs@bogon pg_wal]$ pg_waldump 00000001000000000000000C
...
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C019FB0, prev 0/0C019E18, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C019FE4, prev 0/0C019FB0, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
rmgr: XLOG len (rec/tot): 102/ 102, tx: 0, lsn: 0/0C01A02C, prev 0/0C019FE4, desc: CHECKPOINT_ONLINE redo 0/C019FE4; tli 1; prev tli 1; fpw true; xid 0:723; oid 24639; multi 1; offset 0; oldest xid 548 in DB 1; oldest multi 1 in DB 1; oldest/newest commit timestamp xid: 0/0; oldest running xid 723; online
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C01A094, prev 0/0C01A02C, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
pg_waldump: FATAL: error in WAL record at 0/C01A094: invalid record length at 0/C01A0C8: wanted 24, got 0
5、此时备机pg_control文件
[yzs@bogon pg_wal]$ pg_controldata -D /home/yzs/data
pg_control version number: 1002
Catalog version number: 201707211
Database system identifier: 6602905782398358082
Database cluster state: in archive recovery
pg_control last modified: Thu 02 May 2019 12:58:24 AM PDT
Latest checkpoint location: 0/C01A02C
Prior checkpoint location: 0/C005470
Latest checkpoint's REDO location: 0/C019FE4
Latest checkpoint's REDO WAL file: 00000001000000000000000C
6、主机执行一次insert
yzs=# insert into t1 values(1,2);
INSERT 0 1
7、此时备机位置：
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C019FB0, prev 0/0C019E18, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C019FE4, prev 0/0C019FB0, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
rmgr: XLOG len (rec/tot): 102/ 102, tx: 0, lsn: 0/0C01A02C, prev 0/0C019FE4, desc: CHECKPOINT_ONLINE redo 0/C019FE4; tli 1; prev tli 1; fpw true; xid 0:723; oid 24639; multi 1; offset 0; oldest xid 548 in DB 1; oldest multi 1 in DB 1; oldest/newest commit timestamp xid: 0/0; oldest running xid 723; online
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C01A094, prev 0/0C01A02C, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
rmgr: Heap len (rec/tot): 54/ 6986, tx: 723, lsn: 0/0C01A0C8, prev 0/0C01A094, desc: INSERT off 192, blkref #0: rel 1663/16434/16443 blk 2899 FPW
rmgr: Transaction len (rec/tot): 34/ 34, tx: 723, lsn: 0/0C01BC14, prev 0/0C01A0C8, desc: COMMIT 2019-05-02 01:01:07.715235 PDT
pg_waldump: FATAL: error in WAL record at 0/C01BC14: invalid record length at 0/C01BC38: wanted 24, got 0
8、主机执行checkpoint
lti 1; offset 0; oldest xid 548 in DB 1; oldest multi 1 in DB 1; oldest/newest commit timestamp xid: 0/0; oldest running xid 723; online
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C01A094, prev 0/0C01A02C, desc: RUNNING_XACTS nextXid 723 latestCompletedXid 722 oldestRunningXid 723
rmgr: Heap len (rec/tot): 54/ 6986, tx: 723, lsn: 0/0C01A0C8, prev 0/0C01A094, desc: INSERT off 192, blkref #0: rel 1663/16434/16443 blk 2899 FPW
rmgr: Transaction len (rec/tot): 34/ 34, tx: 723, lsn: 0/0C01BC14, prev 0/0C01A0C8, desc: COMMIT 2019-05-02 01:01:07.715235 PDT
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C01BC38, prev 0/0C01BC14, desc: RUNNING_XACTS nextXid 724 latestCompletedXid 723 oldestRunningXid 724
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C01BC6C, prev 0/0C01BC38, desc: RUNNING_XACTS nextXid 724 latestCompletedXid 723 oldestRunningXid 724
rmgr: XLOG len (rec/tot): 102/ 102, tx: 0, lsn: 0/0C01BCA0, prev 0/0C01BC6C, desc: CHECKPOINT_ONLINE redo 0/C01BC6C; tli 1; prev tli 1; fpw true; xid 0:724; oid 24639; multi 1; offset 0; oldest xid 548 in DB 1; oldest multi 1 in DB 1; oldest/newest commit timestamp xid: 0/0; oldest running xid 724; online
rmgr: Standby len (rec/tot): 50/ 50, tx: 0, lsn: 0/0C01BD08, prev 0/0C01BCA0, desc: RUNNING_XACTS nextXid 724 latestCompletedXid 723 oldestRunningXid 724
pg_waldump: FATAL: error in WAL record at 0/C01BD08: invalid record length at 0/C01BD3C: wanted 24, got 0
9、备机执行完checkpoint后
g_control version number: 1002
Catalog version number: 201707211
Database system identifier: 6602905782398358082
Database cluster state: in archive recovery
pg_control last modified: Thu 02 May 2019 01:03:16 AM PDT
Latest checkpoint location: 0/C01BCA0
Prior checkpoint location: 0/C01A02C
Latest checkpoint's REDO location: 0/C01BC6C
Latest checkpoint's REDO WAL file: 00000001000000000000000C
可以看到备机可以执行checkpoint，但是并不会产生WAL日志。
具体过程
1、备机shutdown时，checkpoint调用流程
1）ShutdownXLOG函数调用CreateRestartPoint做checkpoint，checkpoint位点来自XLogCtl->lastCheckPointRecPtr：
ShutdownXLOG->CreateRestartPoint:
lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
ControlFile->checkPoint = lastCheckPointRecPtr;
2）恢复进程更新XLogCtl->lastCheckPointRecPtr值
StartupXLOG
do{
...
RmgrTable[record->xl_rmid].rm_redo(xlogreader);
|-- RecoveryRestartPoint(&checkPoint);
|-- lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
|-- ControlFile->checkPoint = lastCheckPointRecPtr;
...
record = ReadRecord(xlogreader, InvalidXLogRecPtr, LOG, false);
|-- record = XLogReadRecord(xlogreader, RecPtr, &errormsg);
|-- ReadRecPtr = xlogreader->ReadRecPtr;
} while (record != NULL);
ShutdownXLOG->CreateRestartPoint:
lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
ControlFile->checkPoint = lastCheckPointRecPtr;
1、备机回放
StartupXLOG
do{
...
RmgrTable[record->xl_rmid].rm_redo(xlogreader);//回放
...
record = ReadRecord(xlogreader, InvalidXLogRecPtr, LOG, false);//读取一个xlog
} while (record != NULL);
2、回放函数
void
xlog_redo(XLogReaderState *record)
{
...
else if (info == XLOG_CHECKPOINT_SHUTDOWN){
...
memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
...
RecoveryRestartPoint(&checkPoint);
}else if (info == XLOG_CHECKPOINT_ONLINE){
...
memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
...
RecoveryRestartPoint(&checkPoint);
}
...
}
3、RecoveryRestartPoint
static void
RecoveryRestartPoint(const CheckPoint *checkPoint)
{
...
SpinLockAcquire(&XLogCtl->info_lck);
XLogCtl->lastCheckPointRecPtr = ReadRecPtr;//ReadRecPtr为读取checkpoint记录后的位置
XLogCtl->lastCheckPointEndPtr = EndRecPtr;
XLogCtl->lastCheckPoint = *checkPoint;
SpinLockRelease(&XLogCtl->info_lck);
}
4、ReadRecPtr赋值
ReadRecord
for (;;)
{
char *errormsg;
record = XLogReadRecord(xlogreader, RecPtr, &errormsg);
ReadRecPtr = xlogreader->ReadRecPtr;
EndRecPtr = xlogreader->EndRecPtr;
...
}
5、备机createcheckpoint
bool
CreateRestartPoint(int flags)
{
LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
/* Get a local copy of the last safe checkpoint record. */
SpinLockAcquire(&XLogCtl->info_lck);
lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;//checkpoint的位置来自XLogCtl->lastCheckPointRecPtr
lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
lastCheckPoint = XLogCtl->lastCheckPoint;
SpinLockRelease(&XLogCtl->info_lck);
...
if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) || lastCheckPoint.redo <= ControlFile->checkPointCopy.redo){
//回放了最后一个checkpoint记录后，备机再次手动执行checkpoint命令
UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
if (flags & CHECKPOINT_IS_SHUTDOWN){
LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
ControlFile->time = (pg_time_t) time(NULL);
UpdateControlFile();
LWLockRelease(ControlFileLock);
}
LWLockRelease(CheckpointLock);
return false;
}
...
LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY && ControlFile->checkPointCopy.redo < lastCheckPoint.redo){
ControlFile->prevCheckPoint = ControlFile->checkPoint;
ControlFile->checkPoint = lastCheckPointRecPtr;//checkpoint的位置
ControlFile->checkPointCopy = lastCheckPoint;
ControlFile->time = (pg_time_t) time(NULL);
...
if (flags & CHECKPOINT_IS_SHUTDOWN)
ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
UpdateControlFile();
}
...
return true;
}
6、备机shutdown
void
ShutdownXLOG(int code, Datum arg)
{
/*
* Signal walsenders to move to stopping state.
*/
WalSndInitStopping();
/*
* Wait for WAL senders to be in stopping state. This prevents commands
* from writing new WAL.
*/
WalSndWaitStopping();
if (RecoveryInProgress())//备机写checkpoint
CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
else
{
/*
* If archiving is enabled, rotate the last XLOG file so that all the
* remaining records are archived (postmaster wakes up the archiver
* process one more time at the end of shutdown). The checkpoint
* record will go to the next XLOG file and won't be archived (yet).
*/
if (XLogArchivingActive() && XLogArchiveCommandSet())
RequestXLogSwitch(false);
CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
}
ShutdownCLOG();
ShutdownCommitTs();
ShutdownSUBTRANS();
ShutdownMultiXact();
}
小结
PostgreSQL备库也可以写检查点，目的是避免每次重启备库都需要从上一个检查点（由主库产生，在WAL中回放出来的）APPLY后面所有的WAL。但是他记录的checkpoint位点是从主库传过来的。这样的话就有问题了，如果主机很长时间都没有做checkpoint了，备机即使正常关闭，重启时，也会从上一个checkpoint开始恢复，这样也会恢复很长时间；并且多次重启也需要从上一次checkpoint开始重复恢复。