static void
cdb_position_to_end(void)
{
XLogRecPtr redoCheckpointLoc;
CheckPoint redoCheckpoint;
XLogRecPtr endLocation;
// Throw in extra new line to make log more readable.
elog(LOG,"--------------------------");
XLogGetRecoveryStart("QDSYNC", "to get initial restart location", &redoCheckpointLoc, &redoCheckpoint);
syncRedoLoc = redoCheckpoint.redo;
// UNDONE: Minimum of redoCheckpointLoc and redoCheckpoint.redo?
XLogScanForStandbyEndLocation(&syncRedoLoc, &endLocation);
ereport(LOG,
(errmsg("QDSYNC: reporting recovery start location %s and scanned end location %s",
XLogLocationToString(&syncRedoLoc),
XLogLocationToString2(&endLocation))));
// Throw in extra new line to make log more readable.
elog(LOG,"--------------------------");
/*
* Open up end location segment and set offset to end.
*/
openXlogEnd(&endLocation);
/*
* Extra reply information that gives our standby master XLOG end location
* to the primary.
*/
putEndLocationReply(&endLocation);
}
static int
FileRepPrimary_ResyncBufferPoolIncrementalWrite(ChangeTrackingRequest *request)
{
int status = STATUS_OK;
Page page;
Buffer buf;
BlockNumber numBlocks = 0;
SMgrRelation smgr_relation = NULL;
char relidstr[OIDCHARS + 1 + OIDCHARS + 1 + OIDCHARS + 1];
int ii;
XLogRecPtr loc;
XLogRecPtr loc1;
int count = 0;
int thresholdCount = 0;
bool mirrorDataLossOccurred = FALSE;
int NumberOfRelations = request->count;
FileRepResyncHashEntry_s entry;
ChangeTrackingResult *result = NULL;
while (1)
{
/* allow flushing buffers from buffer pool during scan */
FileRepResync_SetReadBufferRequest();
if ((result = ChangeTracking_GetChanges(request)) != NULL)
{
FileRepResync_ResetReadBufferRequest();
for (ii = 0; ii < result->count; ii++)
{
if (smgr_relation == NULL)
{
NumberOfRelations--;
smgr_relation = smgropen(result->entries[ii].relFileNode);
snprintf(relidstr, sizeof(relidstr), "%u/%u/%u",
smgr_relation->smgr_rnode.spcNode,
smgr_relation->smgr_rnode.dbNode,
smgr_relation->smgr_rnode.relNode);
numBlocks = smgrnblocks(smgr_relation);
if (Debug_filerep_print)
elog(LOG, "resynchronize buffer pool relation '%u/%u/%u' "
"number of blocks:'%u' ",
smgr_relation->smgr_rnode.spcNode,
smgr_relation->smgr_rnode.dbNode,
smgr_relation->smgr_rnode.relNode,
numBlocks);
thresholdCount = Min(numBlocks, 1024);
}
loc1 = result->entries[ii].lsn_end;
/*
* if relation was truncated then block_num from change tracking can be beyond numBlocks
*/
if (result->entries[ii].block_num >= numBlocks)
{
ereport(LOG,
(errmsg("could not resynchonize buffer pool relation '%s' block '%d' (maybe due to truncate), "
"lsn change tracking '%s(%u/%u)' "
"number of blocks '%d' ",
relidstr,
result->entries[ii].block_num,
XLogLocationToString(&loc1),
loc1.xlogid,
loc1.xrecoff,
numBlocks),
FileRep_errcontext()));
goto flush_check;
}
/* allow flushing buffers from buffer pool during scan */
FileRepResync_SetReadBufferRequest();
buf = ReadBuffer_Resync(smgr_relation,
result->entries[ii].block_num,
relidstr);
FileRepResync_ResetReadBufferRequest();
Assert(result->entries[ii].block_num < numBlocks);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
page = BufferGetPage(buf);
loc = PageGetLSN(page);
if(Debug_filerep_print)
{
elog(LOG,
"incremental resync buffer pool identifier '%s' num blocks '%d' blkno '%d' lsn page '%s(%u/%u)' "
"lsn end change tracking '%s(%u/%u)' ",
relidstr,
numBlocks,
result->entries[ii].block_num,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff,
XLogLocationToString(&loc1),
result->entries[ii].lsn_end.xlogid,
result->entries[ii].lsn_end.xrecoff);
}
else
{
char tmpBuf[FILEREP_MAX_LOG_DESCRIPTION_LEN];
snprintf(tmpBuf, sizeof(tmpBuf),
"incremental resync buffer pool identifier '%s' num blocks '%d' blkno '%d' lsn page '%s(%u/%u)' ",
relidstr,
numBlocks,
result->entries[ii].block_num,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff);
FileRep_InsertConfigLogEntry(tmpBuf);
snprintf(tmpBuf, sizeof(tmpBuf),
"incremental resync buffer pool identifier '%s' lsn end change tracking '%s(%u/%u)' ",
relidstr,
XLogLocationToString(&loc1),
result->entries[ii].lsn_end.xlogid,
result->entries[ii].lsn_end.xrecoff);
FileRep_InsertConfigLogEntry(tmpBuf);
}
if (XLByteLE(result->entries[ii].lsn_end, PageGetLSN(page)))
{
if (! XLByteEQ(PageGetLSN(page), result->entries[ii].lsn_end))
{
ereport(LOG,
(errmsg("Resynchonize buffer pool relation '%s' block '%d' has page lsn less than CT lsn, "
"lsn end change tracking '%s(%u/%u)' lsn page '%s(%u/%u)' "
"number of blocks '%d'",
relidstr,
result->entries[ii].block_num,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff,
XLogLocationToString(&loc1),
loc1.xlogid,
loc1.xrecoff,
numBlocks),
FileRep_errcontext()));
}
/*
* It's safe and better to perform write of the page to mirror,
* for this case, as primary and mirror data pages should always
* be same. So, we might do some extra work but definitely won't
* loose out blocks, or error out and need to perform full recovery.
* Need to cover for this case as there are some known scenarios where
* CT file can have extra records which should have been discarded,
* but as we loose out information of xlog LSN cannot be discarded.
* One such case is when CT_TRANSIENT being compacted to CT_COMPACT
* with specific xlog LSN (to discard extra records) in CT mode gets
* interrupted by resync. Compaction during Resync collects all the
* CT records and doesn't have xlog LSN information to discard any
* extra records from CT_TRANSIENT.
*/
smgrwrite(smgr_relation,
result->entries[ii].block_num,
(char *)BufferGetBlock(buf),
FALSE);
}
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepResyncWorker,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
UnlockReleaseBuffer(buf);
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepResyncWorker,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
flush_check:
if (((ii + 1) == result->count) ||
! (result->entries[ii].relFileNode.spcNode == result->entries[ii+1].relFileNode.spcNode &&
result->entries[ii].relFileNode.dbNode == result->entries[ii+1].relFileNode.dbNode &&
result->entries[ii].relFileNode.relNode == result->entries[ii+1].relFileNode.relNode))
{
if (result->ask_for_more == false)
{
smgrimmedsync(smgr_relation);
smgrclose(smgr_relation);
smgr_relation = NULL;
FileRep_GetRelationPath(
entry.fileName,
result->entries[ii].relFileNode,
0 /* segment file number is always 0 for Buffer Pool */);
status = FileRepResync_UpdateEntry(&entry);
if (status != STATUS_OK)
{
break;
}
}
}
if (count > thresholdCount)
{
count = 0;
FileRepSubProcess_ProcessSignals();
if (! (FileRepSubProcess_GetState() == FileRepStateReady &&
dataState == DataStateInResync))
{
mirrorDataLossOccurred = TRUE;
break;
}
}
else
count++;
} // for (ii = 0; ii < result->count; ii++)
} // if ((result = ChangeTracking_GetChanges(request)) != NULL)
FileRepResync_ResetReadBufferRequest();
if (result != NULL && result->ask_for_more == true)
{
Assert(request->count == 1);
request->entries[0].lsn_start = result->next_start_lsn;
}
else
{
break;
}
} // while(1)
ChangeTracking_FreeRequest(request);
ChangeTracking_FreeResult(result);
Insist(NumberOfRelations == 0);
if (mirrorDataLossOccurred)
status = STATUS_ERROR;
return status;
}
static int
FileRepPrimary_ResyncWrite(FileRepResyncHashEntry_s *entry)
{
int status = STATUS_OK;
Page page;
Buffer buf;
BlockNumber numBlocks;
BlockNumber blkno;
SMgrRelation smgr_relation;
char relidstr[OIDCHARS + 1 + OIDCHARS + 1 + OIDCHARS + 1];
XLogRecPtr loc;
int count = 0;
int thresholdCount = 0;
bool mirrorDataLossOccurred = FALSE;
switch (entry->relStorageMgr)
{
case PersistentFileSysRelStorageMgr_BufferPool:
switch (entry->mirrorDataSynchronizationState)
{
case MirroredRelDataSynchronizationState_BufferPoolScanIncremental:
case MirroredRelDataSynchronizationState_FullCopy:
smgr_relation = smgropen(entry->relFileNode);
numBlocks = smgrnblocks(smgr_relation);
snprintf(relidstr, sizeof(relidstr), "%u/%u/%u",
smgr_relation->smgr_rnode.spcNode,
smgr_relation->smgr_rnode.dbNode,
smgr_relation->smgr_rnode.relNode);
if (Debug_filerep_print)
elog(LOG, "resync buffer pool relation '%s' number of blocks '%d' ",
relidstr, numBlocks);
thresholdCount = Min(numBlocks, 1024);
/*
* required in order to report how many blocks were synchronized
* if gp_persistent_relation_node does not return that information
*/
if (entry->mirrorBufpoolResyncChangedPageCount == 0)
{
entry->mirrorBufpoolResyncChangedPageCount = numBlocks - entry->mirrorBufpoolResyncCkptBlockNum;
}
for (blkno = entry->mirrorBufpoolResyncCkptBlockNum; blkno < numBlocks; blkno++)
{
XLogRecPtr endResyncLSN = (isFullResync() ?
FileRepResync_GetEndFullResyncLSN() :
FileRepResync_GetEndIncrResyncLSN());
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepResyncWorkerRead,
DDLNotSpecified,
"", //databaseName
""); // tableName
#endif
FileRepResync_SetReadBufferRequest();
buf = ReadBuffer_Resync(smgr_relation, blkno, relidstr);
FileRepResync_ResetReadBufferRequest();
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
page = BufferGetPage(buf);
loc = PageGetLSN(page);
if (Debug_filerep_print)
{
elog(LOG,
"full resync buffer pool identifier '%s' num blocks '%d' blkno '%d' lsn begin change tracking '%s(%u/%u)' "
"lsn page '%s(%u/%u)' lsn end change tracking '%s(%u/%u)' ",
relidstr,
numBlocks,
blkno,
XLogLocationToString(&entry->mirrorBufpoolResyncCkptLoc),
entry->mirrorBufpoolResyncCkptLoc.xlogid,
entry->mirrorBufpoolResyncCkptLoc.xrecoff,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff,
XLogLocationToString(&endResyncLSN),
endResyncLSN.xlogid,
endResyncLSN.xrecoff);
}
else
{
char tmpBuf[FILEREP_MAX_LOG_DESCRIPTION_LEN];
snprintf(tmpBuf, sizeof(tmpBuf),
"full resync buffer pool identifier '%s' num blocks '%d' blkno '%d' lsn begin change tracking '%s(%u/%u)' ",
relidstr,
numBlocks,
blkno,
XLogLocationToString(&entry->mirrorBufpoolResyncCkptLoc),
entry->mirrorBufpoolResyncCkptLoc.xlogid,
entry->mirrorBufpoolResyncCkptLoc.xrecoff);
FileRep_InsertConfigLogEntry(tmpBuf);
snprintf(tmpBuf, sizeof(tmpBuf),
"full resync buffer pool identifier '%s' lsn page '%s(%u/%u)' lsn end change tracking '%s(%u/%u)' ",
relidstr,
XLogLocationToString(&loc),
loc.xlogid,
loc.xrecoff,
XLogLocationToString(&endResyncLSN),
endResyncLSN.xlogid,
endResyncLSN.xrecoff);
FileRep_InsertConfigLogEntry(tmpBuf);
}
if (XLByteLE(PageGetLSN(page), endResyncLSN) &&
XLByteLE(entry->mirrorBufpoolResyncCkptLoc, PageGetLSN(page)))
{
smgrwrite(smgr_relation,
blkno,
(char *)BufferGetBlock(buf),
FALSE);
}
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepResyncWorker,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
UnlockReleaseBuffer(buf);
if (count > thresholdCount)
{
count = 0;
FileRepSubProcess_ProcessSignals();
if (! (FileRepSubProcess_GetState() == FileRepStateReady &&
dataState == DataStateInResync))
{
mirrorDataLossOccurred = TRUE;
break;
}
}
else
count++;
}
if (mirrorDataLossOccurred)
break;
if (entry->mirrorDataSynchronizationState != MirroredRelDataSynchronizationState_FullCopy)
{
LockRelationForResyncExtension(&smgr_relation->smgr_rnode, ExclusiveLock);
numBlocks = smgrnblocks(smgr_relation);
smgrtruncate(smgr_relation,
numBlocks,
TRUE /* isTemp, TRUE means to not record in XLOG */,
FALSE /* isLocalBuf */,
&entry->persistentTid,
entry->persistentSerialNum);
UnlockRelationForResyncExtension(&smgr_relation->smgr_rnode, ExclusiveLock);
}
smgrimmedsync(smgr_relation);
smgrclose(smgr_relation);
smgr_relation = NULL;
break;
case MirroredRelDataSynchronizationState_None:
case MirroredRelDataSynchronizationState_DataSynchronized:
break;
default:
ereport(LOG,
(errmsg("could not resynchronize relation '%u/%u/%u' "
"mirror synchronization state:'%s(%d)' ",
entry->relFileNode.relNode,
entry->relFileNode.spcNode,
entry->relFileNode.dbNode,
MirroredRelDataSynchronizationState_Name(entry->mirrorDataSynchronizationState),
entry->mirrorDataSynchronizationState)));
break;
}
break;
case PersistentFileSysRelStorageMgr_AppendOnly:
{
MirroredAppendOnlyOpen mirroredOpen;
int primaryError;
bool mirrorDataLossOccurred;
char *buffer = NULL;
int64 endOffset = entry->mirrorAppendOnlyNewEof;
int64 startOffset = entry->mirrorAppendOnlyLossEof;
int32 bufferLen = 0;
int retval = 0;
switch (entry->mirrorDataSynchronizationState)
{
case MirroredRelDataSynchronizationState_AppendOnlyCatchup:
case MirroredRelDataSynchronizationState_FullCopy:
/*
* required in order to report how many blocks were synchronized
* if gp_persistent_relation_node does not return that information
*/
if (entry->mirrorBufpoolResyncChangedPageCount == 0)
{
entry->mirrorBufpoolResyncChangedPageCount = (endOffset - startOffset) / BLCKSZ;
}
/*
* The MirroredAppendOnly_OpenResynchonize routine knows we are a resynch worker and
* will open BOTH, but write only the MIRROR!!!
*/
MirroredAppendOnly_OpenResynchonize(
&mirroredOpen,
&entry->relFileNode,
entry->segmentFileNum,
startOffset,
&primaryError,
&mirrorDataLossOccurred);
if (primaryError != 0)
{
ereport(ERROR,
(errcode_for_file_access(),
errmsg("could not open file %u/%u/%u.%u : %s",
entry->relFileNode.dbNode,
entry->relFileNode.spcNode,
entry->relFileNode.relNode,
entry->segmentFileNum,
strerror(primaryError))));
break;
}
if (mirrorDataLossOccurred)
break;
/* AO and CO Data Store writes 64k size by default */
bufferLen = (Size) Min(2*BLCKSZ, endOffset - startOffset);
buffer = (char*) palloc(bufferLen);
if (buffer == NULL)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
(errmsg("not enough memory for resynchronization"))));
MemSet(buffer, 0, bufferLen);
while (startOffset < endOffset)
{
retval = MirroredAppendOnly_Read(
&mirroredOpen,
buffer,
bufferLen);
if (retval != bufferLen)
{
ereport(ERROR,
(errcode_for_file_access(),
errmsg("could not read from position:" INT64_FORMAT " in file %u/%u/%u.%u : %m",
startOffset,
entry->relFileNode.dbNode,
entry->relFileNode.spcNode,
entry->relFileNode.relNode,
entry->segmentFileNum)));
break;
}
MirroredAppendOnly_Append(
&mirroredOpen,
buffer,
bufferLen,
&primaryError,
&mirrorDataLossOccurred);
if (mirrorDataLossOccurred)
break;
Assert(primaryError == 0); // No primary writes as resync worker.
startOffset += bufferLen;
/* AO and CO Data Store writes 64k size by default */
bufferLen = (Size) Min(2*BLCKSZ, endOffset - startOffset);
}
if (buffer)
{
pfree(buffer);
buffer = NULL;
}
if (mirrorDataLossOccurred)
break;
/* Flush written data on Mirror */
MirroredAppendOnly_Flush(
&mirroredOpen,
&primaryError,
&mirrorDataLossOccurred);
if (mirrorDataLossOccurred)
break;
Assert(primaryError == 0); // Not flushed on primary as resync worker.
/* Close Primary and Mirror */
MirroredAppendOnly_Close(
&mirroredOpen,
&mirrorDataLossOccurred);
break;
case MirroredRelDataSynchronizationState_None:
case MirroredRelDataSynchronizationState_DataSynchronized:
break;
default:
ereport(LOG,
(errmsg("could not resynchronize relation '%u/%u/%u' "
"mirror synchronization state:'%s(%d)' ",
entry->relFileNode.relNode,
entry->relFileNode.spcNode,
entry->relFileNode.dbNode,
MirroredRelDataSynchronizationState_Name(entry->mirrorDataSynchronizationState),
entry->mirrorDataSynchronizationState)));
break;
}
break;
} //case
default:
Assert(0);
break;
} //switch
if (mirrorDataLossOccurred)
status = STATUS_ERROR;
return status;
}
static void
WalSendServerDoRequest(WalSendRequest *walSendRequest)
{
bool successful;
struct timeval standbyTimeout;
WalSendServerGetStandbyTimeout(&standbyTimeout);
switch (walSendRequest->command)
{
case PositionToEnd:
elog((Debug_print_qd_mirroring ? LOG : DEBUG5), "PositionToEnd");
successful = write_position_to_end(&originalEndLocation,
NULL, &walsend_shutdown_requested);
if (successful)
elog(LOG,"Standby master returned transaction log end location %s",
XLogLocationToString(&originalEndLocation));
else
{
disableQDMirroring_ConnectionError(
"Unable to connect to standby master and determine transaction log end location",
GetStandbyErrorString());
disconnectMirrorQD_SendClose();
}
break;
case Catchup:
elog((Debug_print_qd_mirroring ? LOG : DEBUG5), "Catchup");
if (isQDMirroringCatchingUp())
{
bool tooFarBehind = false;
elog(LOG,"Current master transaction log is flushed through location %s",
XLogLocationToString(&walSendRequest->flushedLocation));
if (XLByteLT(originalEndLocation, walSendRequest->flushedLocation))
{
/*
* Standby master is behind the primary. Send catchup WAL.
*/
/*
* Use a TRY block to catch errors from our attempt to read
* the primary's WAL. Errors from sending to the standby
* come up as a boolean return (successful).
*/
PG_TRY();
{
successful = XLogCatchupQDMirror(
&originalEndLocation,
&walSendRequest->flushedLocation,
&standbyTimeout,
&walsend_shutdown_requested);
}
PG_CATCH();
{
/*
* Report the error related to reading the primary's WAL
* to the server log
*/
/*
* But first demote the error to something much less
* scary.
*/
if (!elog_demote(WARNING))
{
elog(LOG,"unable to demote error");
PG_RE_THROW();
}
EmitErrorReport();
FlushErrorState();
successful = false;
tooFarBehind = true;
}
PG_END_TRY();
if (successful)
{
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),
"catchup send from standby end %s through primary flushed location %s",
XLogLocationToString(&originalEndLocation),
XLogLocationToString2(&walSendRequest->flushedLocation));
}
}
else if (XLByteEQ(originalEndLocation, walSendRequest->flushedLocation))
{
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),"Mirror was already caught up");
successful = true;
}
else
{
elog(WARNING,"Standby master transaction log location %s is beyond the current master end location %s",
XLogLocationToString(&originalEndLocation),
XLogLocationToString2(&walSendRequest->flushedLocation));
successful = false;
}
if (successful)
{
char detail[200];
int count;
count = snprintf(
detail, sizeof(detail),
"Transaction log copied from locations %s through %s to the standby master",
XLogLocationToString(&originalEndLocation),
XLogLocationToString2(&walSendRequest->flushedLocation));
if (count >= sizeof(detail))
{
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("format command string failure")));
}
enableQDMirroring("Master mirroring is now synchronized", detail);
currentEndLocation = walSendRequest->flushedLocation;
periodicLen = 0;
periodicLocation = currentEndLocation;
}
else
{
if (tooFarBehind)
{
disableQDMirroring_TooFarBehind(
"The current master was unable to synchronize the standby master "
"because the transaction logs on the current master were recycled. "
"A gpinitstandby (at an appropriate time) will be necessary to copy "
"over the whole master database to the standby master so it may be synchronized");
}
else
{
disableQDMirroring_ConnectionError(
"Connection to the standby master was lost during transaction log catchup",
GetStandbyErrorString());
}
disconnectMirrorQD_SendClose();
}
}
else if (isQDMirroringDisabled())
{
elog((Debug_print_qd_mirroring ? LOG : DEBUG5), "Master Mirror Send: Master mirroring not catching-up (state is disabled)");
}
else
{
elog(ERROR,"unexpected master mirroring state %s",
QDMirroringStateString());
}
break;
case WriteWalPages:
if (Debug_print_qd_mirroring)
elog(LOG, "WriteWalPages");
if (isQDMirroringEnabled())
{
char *from;
Size nbytes;
bool more= false;
/*
* For now, save copy of data until flush. This could be
* optimized.
*/
if (saveBuffer == NULL)
{
uint32 totalBufferLen = XLOGbuffers * XLOG_BLCKSZ;
saveBuffer = malloc(totalBufferLen);
if (saveBuffer == NULL)
elog(ERROR,"Could not allocate buffer for xlog data (%d bytes)",
totalBufferLen);
saveBufferLen = 0;
}
XLogGetBuffer(walSendRequest->startidx, walSendRequest->npages,
&from, &nbytes);
if (saveBufferLen == 0)
{
more = false;
writeLogId = walSendRequest->logId;
writeLogSeg = walSendRequest->logSeg;
writeLogOff = walSendRequest->logOff;
memcpy(saveBuffer, from, nbytes);
saveBufferLen = nbytes;
}
else
{
more = true;
memcpy(&saveBuffer[saveBufferLen], from, nbytes);
saveBufferLen += nbytes;
}
if (Debug_print_qd_mirroring)
elog(LOG,
"Master Mirror Send: WriteWalPages (%s) startidx %d, npages %d, timeLineID %d, logId %u, logSeg %u, logOff 0x%X, nbytes 0x%X",
(more ? "more" : "new"),
walSendRequest->startidx,
walSendRequest->npages,
walSendRequest->timeLineID,
walSendRequest->logId,
walSendRequest->logSeg,
walSendRequest->logOff,
(int)nbytes);
}
case FlushWalPages:
if (Debug_print_qd_mirroring)
elog(LOG, "FlushWalPages");
if (isQDMirroringEnabled())
{
char cmd[MAXFNAMELEN + 50];
if (saveBufferLen == 0)
successful = true;
else
{
if (snprintf(cmd, sizeof(cmd),"xlog %d %d %d %d",
writeLogId, writeLogSeg, writeLogOff,
(int)saveBufferLen) >= sizeof(cmd))
elog(ERROR,"could not create cmd for qd mirror logid %d seg %d",
writeLogId, writeLogSeg);
successful = write_qd_sync(cmd, saveBuffer, saveBufferLen,
&standbyTimeout,
&walsend_shutdown_requested);
if (successful)
{
XLogRecPtr oldEndLocation;
oldEndLocation = currentEndLocation;
currentEndLocation.xlogid = writeLogId;
currentEndLocation.xrecoff = writeLogSeg * XLogSegSize + writeLogOff;
if (currentEndLocation.xrecoff >= XLogFileSize)
{
(currentEndLocation.xlogid)++;
currentEndLocation.xrecoff = 0;
}
if (XLByteLT(oldEndLocation,currentEndLocation))
{
periodicLen += saveBufferLen;
if (periodicLen > periodicReportLen)
{
elog(LOG,
"Master mirroring periodic report: %d bytes successfully send to standby master for locations %s through %s",
periodicLen,
XLogLocationToString(&periodicLocation),
XLogLocationToString2(¤tEndLocation));
periodicLen = 0;
periodicLocation = currentEndLocation;
}
}
else
{
if (Debug_print_qd_mirroring)
elog(LOG,
"Send to Master mirror successful. New end location %s (old %s)",
XLogLocationToString(¤tEndLocation),
XLogLocationToString2(&oldEndLocation));
}
}
else
{
disableQDMirroring_ConnectionError(
"Connection to the standby master was lost attempting to send new transaction log",
GetStandbyErrorString());
disconnectMirrorQD_SendClose();
}
/*
* Reset so WriteWalPages can fill the buffer again.
*/
saveBufferLen = 0;
writeLogId = 0;
writeLogSeg = 0;
writeLogOff = 0;
}
if (successful && walSendRequest->haveNewCheckpointLocation)
{
uint32 logid;
uint32 seg;
uint32 offset;
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),"New previous checkpoint location %s",
XLogLocationToString(&walSendRequest->newCheckpointLocation));
XLByteToSeg(walSendRequest->newCheckpointLocation, logid, seg);
offset = walSendRequest->newCheckpointLocation.xrecoff % XLogSegSize;
if (snprintf(cmd, sizeof(cmd),"new_checkpoint_location %d %d %d",
logid, seg, offset) >= sizeof(cmd))
elog(ERROR,"could not create cmd for qd mirror logid %d seg %d offset %d",
logid, seg, offset);
successful = write_qd_sync(cmd, NULL, 0,
NULL, &walsend_shutdown_requested);
if (successful)
{
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),"Send of new checkpoint location to master mirror successful");
}
else
{
disableQDMirroring_ConnectionError(
"Connection to the standby master was lost attempting to send new checkpoint location",
GetStandbyErrorString());
disconnectMirrorQD_SendClose();
}
}
}
else if (isQDMirroringDisabled())
{
elog((Debug_print_qd_mirroring ? LOG : DEBUG5), "Master Mirror Send: Master mirroring not enabled");
}
else
{
elog(ERROR,"unexpected master mirroring state %s",
QDMirroringStateString());
}
break;
case CloseForShutdown:
if (Debug_print_qd_mirroring)
elog(LOG, "CloseForShutdown");
/*
* Do the work we would normally do when signaled to stop.
*/
WalSendServer_ServiceShutdown();
break;
default:
elog(ERROR, "Unknown WalSendRequestCommand %d", walSendRequest->command);
}
}
bool ChangeTracking_PrintRelationChangeInfo(
RmgrId xl_rmid,
uint8 xl_info,
void *data,
XLogRecPtr *loc,
bool weAreGeneratingXLogNow,
bool printSkipIssuesOnly)
{
bool atLeastOneSkipIssue = false;
int relationChangeInfoArrayCount;
int i;
int arrlen = ChangeTracking_GetInfoArrayDesiredMaxLength(xl_rmid, xl_info);
RelationChangeInfo relationChangeInfoArray[arrlen];
ChangeTracking_GetRelationChangeInfoFromXlog(
xl_rmid,
xl_info,
data,
relationChangeInfoArray,
&relationChangeInfoArrayCount,
arrlen);
for (i = 0; i < relationChangeInfoArrayCount; i++)
{
RelationChangeInfo *relationChangeInfo;
int64 maxPersistentSerialNum;
bool skip;
bool zeroTid = false;
bool invalidTid = false;
bool zeroSerialNum = false;
bool invalidSerialNum = false;
bool skipIssue = false;
relationChangeInfo = &relationChangeInfoArray[i];
if (weAreGeneratingXLogNow)
maxPersistentSerialNum = PersistentRelfile_MyHighestSerialNum();
else
maxPersistentSerialNum = PersistentRelfile_CurrentMaxSerialNum();
skip = GpPersistent_SkipXLogInfo(relationChangeInfo->relFileNode.relNode);
if (!skip)
{
zeroTid = PersistentStore_IsZeroTid(&relationChangeInfo->persistentTid);
if (!zeroTid)
invalidTid = !ItemPointerIsValid(&relationChangeInfo->persistentTid);
zeroSerialNum = (relationChangeInfo->persistentSerialNum == 0);
if (!zeroSerialNum)
{
invalidSerialNum = (relationChangeInfo->persistentSerialNum < 0);
/*
* If we have'nt done the scan yet... do not do upper range check.
*/
if (maxPersistentSerialNum != 0 &&
relationChangeInfo->persistentSerialNum > maxPersistentSerialNum)
invalidSerialNum = true;
}
skipIssue = (zeroTid || invalidTid || zeroSerialNum || invalidSerialNum);
}
if (!printSkipIssuesOnly || skipIssue)
elog(LOG,
"ChangeTracking_PrintRelationChangeInfo: [%d] xl_rmid %d, xl_info 0x%X, %u/%u/%u, block number %u, LSN %s, persistent serial num " INT64_FORMAT ", TID %s, maxPersistentSerialNum " INT64_FORMAT ", skip %s, zeroTid %s, invalidTid %s, zeroSerialNum %s, invalidSerialNum %s, skipIssue %s",
i,
xl_rmid,
xl_info,
relationChangeInfo->relFileNode.spcNode,
relationChangeInfo->relFileNode.dbNode,
relationChangeInfo->relFileNode.relNode,
relationChangeInfo->blockNumber,
XLogLocationToString(loc),
relationChangeInfo->persistentSerialNum,
ItemPointerToString(&relationChangeInfo->persistentTid),
maxPersistentSerialNum,
(skip ? "true" : "false"),
(zeroTid ? "true" : "false"),
(invalidTid ? "true" : "false"),
(zeroSerialNum ? "true" : "false"),
(invalidSerialNum ? "true" : "false"),
(skipIssue ? "true" : "false"));
if (skipIssue)
atLeastOneSkipIssue = true;
}
return atLeastOneSkipIssue;
}
/*
* cdb_sync_xlog - process xlog sync
*/
static void
cdb_sync_xlog(void)
{
uint32 currentBlockOffset;
XLogRecPtr writeLoc;
elog((Debug_print_qd_mirroring ? LOG : DEBUG5), "QDSYNC: write logid %d seg %d woffset 0x%X, wlen 0x%X",
wlogid, wseg, woffset, wlen);
if (woffset % XLOG_BLCKSZ != 0)
{
elog(ERROR,"QDSYNC: not on block boundaries 0x%X",
woffset);
}
if (wlogid != xlogid || wseg != xseg)
{
elog((Debug_print_qd_mirroring ? LOG : DEBUG5), "QDSYNC: closing previous file %s",
xlogfilename);
if (xlogfilefd >= 0)
{
close(xlogfilefd);
xlogfilefd = -1;
xlogfileoffset = -1;
}
xlogid = wlogid;
xseg = wseg;
openXlogNextFile();
/*
* Assume caller knows where to write.
*/
xlogfileoffset = woffset;
}
/*
* Validate we are appending or overwritting previous block
*/
currentBlockOffset = (xlogfileoffset / XLOG_BLCKSZ) * XLOG_BLCKSZ;
if (woffset != currentBlockOffset &&
woffset + XLOG_BLCKSZ != currentBlockOffset)
{
elog(ERROR,"QDSYNC: not appending to end (primary: 0x%X, standby: 0x%X)",
woffset, xlogfileoffset);
}
/*
* no validation checking on xlog. xlog sync is by block and may repeat
* the same block, so we do not have any way to check it now. we will rely
* on tmlog checking for now.
*/
ensureBufferSize();
readLogMessage(buf, wlen);
if ((wlen / XLOG_BLCKSZ) * XLOG_BLCKSZ != wlen)
{
int roundedUp;
int padLen;
/*
* Pad buffer out with zeros.
*/
roundedUp = ((wlen + XLOG_BLCKSZ - 1) / XLOG_BLCKSZ) * XLOG_BLCKSZ;
Assert(buflen >= roundedUp);
padLen = roundedUp - wlen;
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),
"QDSYNC: padding buffer with %d zeros (wlen %d, roundedUp %d)",
padLen, wlen, roundedUp);
memset(&((char*) buf)[wlen], 0, padLen);
wlen = roundedUp;
}
writeLoc.xlogid = wlogid;
writeLoc.xrecoff = wseg * XLogSegSize +
woffset;
syncWriteLog(xlogfilefd, buf, woffset, wlen);
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),
"QDSYNC: wrote location %s len 0x%X",
XLogLocationToString(&writeLoc),
wlen);
xlogfileoffset = woffset + wlen;
}
void
cdb_perform_redo(XLogRecPtr *redoCheckPointLoc, CheckPoint *redoCheckPoint, XLogRecPtr *newCheckpointLoc)
{
CheckPoint oldRedoCheckpoint;
uint32 logid;
uint32 seg;
int nsegsremoved;
if (redoCheckPointLoc->xlogid == 0 && redoCheckPointLoc->xrecoff == 0)
{
XLogGetRecoveryStart("QDSYNC", "for redo apply", redoCheckPointLoc, redoCheckPoint);
}
XLogStandbyRecoverRange(redoCheckPointLoc, redoCheckPoint, newCheckpointLoc);
/*
* Sample the recovery start location now to see if appling redo
* processed checkpoint records and moved the restart location forward.
*/
oldRedoCheckpoint = *redoCheckPoint;
XLogGetRecoveryStart("QDSYNC", "for redo progress check", redoCheckPointLoc, redoCheckPoint);
if (XLByteLT(oldRedoCheckpoint.redo,redoCheckPoint->redo))
{
ereport(LOG,
(errmsg("QDSYNC: transaction redo moved the restart location from %s to %s",
XLogLocationToString(&oldRedoCheckpoint.redo),
XLogLocationToString2(&redoCheckPoint->redo))));
}
else
{
Assert(XLByteEQ(oldRedoCheckpoint.redo,redoCheckPoint->redo));
ereport(LOG,
(errmsg("QDSYNC: transaction redo did not move the restart location %s forward this pass",
XLogLocationToString(&oldRedoCheckpoint.redo))));
return;
}
XLByteToSeg(redoCheckPoint->redo, logid, seg);
/*
* Delete offline log files (those no longer needed even for previous
* checkpoint).
*/
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),
"QDSYNC: keep log files as far back as (logid %d, seg %d)",
logid, seg);
if (logid || seg)
{
PrevLogSeg(logid, seg);
elog((Debug_print_qd_mirroring ? LOG : DEBUG5),
"QDSYNC: delete offline log files up to (logid %d, seg %d)",
logid, seg);
XLogRemoveStandbyLogs(logid, seg, &nsegsremoved);
if (nsegsremoved > 0)
{
// Throw in extra new line to make log more readable.
ereport(LOG,
(errmsg("QDSYNC: %d logs removed through logid %d, seg %d\n",
nsegsremoved,
logid, seg)));
}
}
// Throw in extra new line to make log more readable.
elog(LOG,"--------------------------");
}
static void PersistentStore_DiagnoseDumpTable(
PersistentStoreData *storeData,
PersistentStoreSharedData *storeSharedData)
{
if (disable_persistent_diagnostic_dump)
{
return;
}
MIRROREDLOCK_BUFMGR_DECLARE;
PersistentStoreScan storeScan;
ItemPointerData persistentTid;
int64 persistentSerialNum;
Datum *values;
BlockNumber lastDisplayedBlockNum;
bool displayedOne;
BlockNumber currentBlockNum;
elog(LOG,
"Diagnostic dump of persistent table ('%s'): maximum in-use serial number " INT64_FORMAT ", maximum free order number " INT64_FORMAT ", free TID %s, maximum known TID %s",
storeData->tableName,
storeSharedData->maxInUseSerialNum,
storeSharedData->maxFreeOrderNum,
ItemPointerToString(&storeSharedData->freeTid),
ItemPointerToString2(&storeSharedData->maxTid));
values = (Datum*)palloc(storeData->numAttributes * sizeof(Datum));
PersistentStore_BeginScan(
storeData,
storeSharedData,
&storeScan);
lastDisplayedBlockNum = 0;
displayedOne = false;
while (PersistentStore_GetNext(
&storeScan,
values,
&persistentTid,
&persistentSerialNum))
{
/*
* Use the BlockIdGetBlockNumber routine because ItemPointerGetBlockNumber
* asserts for valid TID.
*/
currentBlockNum = BlockIdGetBlockNumber(&persistentTid.ip_blkid);
if (!displayedOne || currentBlockNum != lastDisplayedBlockNum)
{
Buffer buffer;
PageHeader page;
XLogRecPtr lsn;
/*
* Fetch the block and display the LSN.
*/
// -------- MirroredLock ----------
MIRROREDLOCK_BUFMGR_LOCK;
buffer = ReadBuffer(
storeScan.persistentRel,
currentBlockNum);
page = (PageHeader) BufferGetPage(buffer);
lsn = PageGetLSN(page);
ReleaseBuffer(buffer);
MIRROREDLOCK_BUFMGR_UNLOCK;
// -------- MirroredLock ----------
elog(LOG, "Diagnostic LSN %s of page %u",
XLogLocationToString(&lsn),
currentBlockNum);
lastDisplayedBlockNum = currentBlockNum;
displayedOne = true;
}
/*
* Display the persistent tuple.
*/
(*storeData->printTupleCallback)(
LOG,
"DIAGNOSE",
&persistentTid,
values);
}
PersistentStore_EndScan(&storeScan);
pfree(values);
}
