PostgreSQL备机checkpoint

  数据库异常关闭时，数据库关闭时来不及或者没机会做checkpoint，则需要从上一个一致性检查的开始恢复。

    PostgreSQL备机checkpoint是不能产生checkpoint WAL的，因为如果写这样类型的checkpoint的话，就会将接收的WAL打乱，那么日志将混乱，回放会出问题。

    那么问题来了，备机支持checkpoint吗？他的checkpoint怎么做的？

    PostgreSQL为了缩短恢复时间，备机上也支持checkpoint，即CreateRestartPoint。但是其pg_control文件的checkpoint记录的位点是从主机传过来WAL里面的checkpoint记录位置。

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);

    
    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)
{
    
    WalSndInitStopping();

    
    WalSndWaitStopping();

    if (RecoveryInProgress())//备机写checkpoint
        CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
    else
    {
        
        if (XLogArchivingActive() && XLogArchiveCommandSet())
            RequestXLogSwitch(false);

        CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
    }
    ShutdownCLOG();
    ShutdownCommitTs();
    ShutdownSUBTRANS();
    ShutdownMultiXact();
}

7、总结

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。