/*
*
* FileRepPrimary_RunHeartBeat()
*
*
*/
static void
FileRepPrimary_RunHeartBeat(void)
{
int retry = 0;
Insist(fileRepRole == FileRepPrimaryRole);
Insist(dataState == DataStateInSync ||
dataState == DataStateInResync);
while (1)
{
FileRepSubProcess_ProcessSignals();
while (FileRepSubProcess_GetState() == FileRepStateFault ||
(fileRepShmemArray[0]->state == FileRepStateNotInitialized &&
FileRepSubProcess_GetState() != FileRepStateShutdownBackends &&
FileRepSubProcess_GetState() != FileRepStateShutdown)) {
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
if (FileRepSubProcess_GetState() == FileRepStateShutdown ||
FileRepSubProcess_GetState() == FileRepStateShutdownBackends) {
break;
}
/* verify if flow from primary to mirror and back is alive once per minute */
pg_usleep(50000L); /* 50 ms */
if (FileRepSubProcess_ProcessSignals() == true ||
FileRepSubProcess_GetState() == FileRepStateFault)
{
continue;
}
retry++;
if (retry == 1200) /* 1200 * 50 ms = 60 sec */
{
FileRepPrimary_MirrorHeartBeat(FileRepMessageTypeXLog);
continue;
}
if (retry == 1201) /* 1200 * 50 ms = 60 sec */
{
FileRepPrimary_MirrorHeartBeat(FileRepMessageTypeWriter);
continue;
}
if (retry == 1202) /* 1200 * 50 ms = 60 sec */
{
FileRepPrimary_MirrorHeartBeat(FileRepMessageTypeAO01);
retry = 0;
}
} // while(1)
}
/*
* FileRepPrimary_RunResyncWorker()
*
*/
static int
FileRepPrimary_RunResyncWorker(void)
{
int status = STATUS_OK;
FileRepResyncHashEntry_s *entry = NULL;
ChangeTrackingRequest *request = NULL;
FileRep_InsertConfigLogEntry("run resync worker");
while (1) {
FileRepSubProcess_ProcessSignals();
if (! (FileRepSubProcess_GetState() == FileRepStateReady &&
dataState == DataStateInResync))
{
break;
}
entry = FileRepPrimary_GetResyncEntry(&request);
if (entry == NULL && request == NULL) {
pg_usleep(100000L); /* 100 ms */
continue;
}
Assert(! (entry != NULL && request != NULL));
if (entry != NULL)
{
status = FileRepPrimary_ResyncWrite(entry);
if (status == STATUS_OK)
{
if (entry->mirrorBufpoolResyncChangedPageCount == 0)
{
entry->mirrorBufpoolResyncChangedPageCount = (entry->mirrorAppendOnlyNewEof - entry->mirrorAppendOnlyLossEof) / BLCKSZ;
}
status = FileRepResync_UpdateEntry(entry);
}
}
if (request != NULL)
{
status = FileRepPrimary_ResyncBufferPoolIncrementalWrite(request);
request = NULL;
}
if (status != STATUS_OK)
{
break;
}
}
return status;
}
/*
* FileRepPrimary_StartResyncWorker()
*/
void
FileRepPrimary_StartResyncWorker(void)
{
int status = STATUS_OK;
FileRep_InsertConfigLogEntry("start resync worker");
Insist(fileRepRole == FileRepPrimaryRole);
while (1) {
if (status != STATUS_OK)
{
FileRep_SetSegmentState(SegmentStateFault, FaultTypeMirror);
FileRepSubProcess_SetState(FileRepStateFault);
}
/*
* We are waiting for following conditions to move forward:
*
* Database is running
* And
* if dataState is InResync, we wait for FileRepSubProcess to Ready state
* else don't wait
*/
while (!isDatabaseRunning() ||
!(dataState == DataStateInResync ? FileRepSubProcess_GetState() == FileRepStateReady : true))
{
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() == FileRepStateShutdown ||
FileRepSubProcess_GetState() == FileRepStateShutdownBackends)
{
break;
}
pg_usleep(50000L); /* 50 ms */
}
if (FileRepSubProcess_GetState() == FileRepStateShutdown ||
FileRepSubProcess_GetState() == FileRepStateShutdownBackends) {
break;
}
FileRepSubProcess_InitHeapAccess();
status = FileRepPrimary_RunResyncWorker();
if (status != STATUS_OK) {
continue;
}
break;
} // while(1)
}
/*
*
* FileRepAckPrimary_StartConsumer
*/
void
FileRepAckPrimary_StartConsumer(void)
{
int status = STATUS_OK;
FileRep_InsertConfigLogEntry("run consumer");
while (1) {
if (status != STATUS_OK)
{
FileRep_SetSegmentState(SegmentStateFault, FaultTypeMirror);
FileRepSubProcess_SetState(FileRepStateFault);
}
while (FileRepSubProcess_GetState() == FileRepStateFault ||
(fileRepShmemArray[0]->state == FileRepStateNotInitialized &&
FileRepSubProcess_GetState() != FileRepStateShutdown)) {
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
if (FileRepSubProcess_GetState() == FileRepStateShutdown) {
break;
}
status = FileRepAckPrimary_RunConsumer();
} // while(1)
if (FileRepSubProcess_GetState() == FileRepStateShutdown) {
/* perform graceful shutdown */
}
LWLockAcquire(FileRepAckHashShmemLock, LW_EXCLUSIVE);
FileRep_IpcSignal(fileRepIpcArray[fileRepAckHashShmem->ipcArrayIndex]->semP,
&fileRepIpcArray[fileRepAckHashShmem->ipcArrayIndex]->refCountSemP);
LWLockRelease(FileRepAckHashShmemLock);
/* NOTE free memory (if any) */
return;
}
/*
*
* FileRepPrimary_StartRecoveryInSync()
*
*
*/
static void
FileRepPrimary_StartRecoveryInSync(void)
{
int status = STATUS_OK;
FileRep_InsertConfigLogEntry("run recovery");
while (1) {
if (status != STATUS_OK)
{
FileRep_SetSegmentState(SegmentStateFault, FaultTypeMirror);
FileRepSubProcess_SetState(FileRepStateFault);
}
while (FileRepSubProcess_GetState() == FileRepStateFault ||
(fileRepShmemArray[0]->state == FileRepStateNotInitialized &&
FileRepSubProcess_GetState() != FileRepStateShutdownBackends &&
FileRepSubProcess_GetState() != FileRepStateShutdown)) {
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
if (FileRepSubProcess_GetState() == FileRepStateShutdown ||
FileRepSubProcess_GetState() == FileRepStateShutdownBackends) {
break;
}
if (FileRepSubProcess_GetState() == FileRepStateReady) {
break;
}
Insist(fileRepRole == FileRepPrimaryRole);
Insist(dataState == DataStateInSync);
status = FileRepPrimary_RunRecoveryInSync();
} // while(1)
}
/*
* FileRepPrimary_StartResyncWorker()
*/
void
FileRepPrimary_StartResyncWorker(void)
{
int status = STATUS_OK;
FileRep_InsertConfigLogEntry("start resync worker");
Insist(fileRepRole == FileRepPrimaryRole);
while (1) {
if (status != STATUS_OK)
{
FileRep_SetSegmentState(SegmentStateFault, FaultTypeMirror);
FileRepSubProcess_SetState(FileRepStateFault);
}
while (FileRepSubProcess_GetState() != FileRepStateShutdown &&
FileRepSubProcess_GetState() != FileRepStateShutdownBackends &&
! (FileRepSubProcess_GetState() == FileRepStateReady &&
dataState == DataStateInResync)) {
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
if (FileRepSubProcess_GetState() == FileRepStateShutdown ||
FileRepSubProcess_GetState() == FileRepStateShutdownBackends) {
break;
}
status = FileRepPrimary_RunResyncWorker();
if (status != STATUS_OK) {
continue;
}
break;
} // while(1)
}
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 int
FileRepAckPrimary_RunReceiver(void)
{
uint32_t msgLength = 0;
FileRepConsumerProcIndex_e msgType;
int status = STATUS_OK;
char *msgPositionInsert;
FileRepShmemMessageDescr_s *fileRepShmemMessageDescr;
uint32 spareField;
FileRep_InsertConfigLogEntry("run receiver");
while (1) {
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateReady &&
FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
if ( ! FileRepConnServer_AwaitMessageBegin()) {
/* call was interrupted ... go back to beginning to process signals */
continue;
}
status = FileRepConnServer_ReceiveMessageType(&msgType);
if (status != STATUS_OK) {
break;
}
/* DATA MESSAGE TYPE */
status = FileRepConnServer_ReceiveMessageLength(&msgLength);
if (status != STATUS_OK) {
break;
}
msgPositionInsert = FileRep_ReserveShmem(fileRepAckShmemArray[msgType],
msgLength,
/* not used */ &spareField,
FileRepOperationNotSpecified,
FileRepAckShmemLock);
if (msgPositionInsert == NULL) {
status = STATUS_ERROR;
ereport(WARNING,
(errmsg("mirror failure, "
"could not queue received ack message to be processed, "
"failover requested"),
errhint("run gprecoverseg to re-establish mirror connectivity"),
FileRep_errdetail_Shmem(),
FileRep_errdetail_ShmemAck(),
FileRep_errcontext()));
break;
}
status = FileRepConnServer_ReceiveMessageData(
msgPositionInsert + sizeof(FileRepShmemMessageDescr_s),
msgLength);
if (status != STATUS_OK) {
break;
}
SIMPLE_FAULT_INJECTOR(FileRepReceiver);
fileRepShmemMessageDescr =
(FileRepShmemMessageDescr_s*) msgPositionInsert;
/* it is not in use */
fileRepShmemMessageDescr->messageSync = FALSE;
fileRepShmemMessageDescr->messageState = FileRepShmemMessageStateReady;
LWLockAcquire(FileRepAckShmemLock, LW_EXCLUSIVE);
FileRep_IpcSignal(fileRepIpcArray[fileRepAckShmemArray[msgType]->ipcArrayIndex]->semC,
&fileRepIpcArray[fileRepAckShmemArray[msgType]->ipcArrayIndex]->refCountSemC);
LWLockRelease(FileRepAckShmemLock);
FileRep_InsertLogEntry(
"P_RunReceiver",
FileRep_GetFlatFileIdentifier("", ""),
FileRepRelationTypeNotSpecified,
FileRepOperationNotSpecified,
FILEREP_UNDEFINED,
FILEREP_UNDEFINED,
FileRepAckStateNotInitialized,
spareField,
FILEREP_UNDEFINED);
} // while(1)
FileRepConnServer_CloseConnection();
return status;
}
/*
*
* FileRepPrimary_RunChangeTrackingCompacting()
*
*/
static void
FileRepPrimary_RunChangeTrackingCompacting(void)
{
int retry = 0;
FileRep_InsertConfigLogEntry("run change tracking compacting if records has to be discarded");
/*
* We have to check if any records have to be discarded from Change Tracking log file.
* Due to crash it can happen that the highest change tracking log lsn > the highest xlog lsn.
*
* Records from change tracking log file can be discarded only after database is started.
* Full environhment has to be set up in order to run queries over SPI.
*/
while (FileRepSubProcess_GetState() != FileRepStateShutdown &&
FileRepSubProcess_GetState() != FileRepStateShutdownBackends &&
isDatabaseRunning() == FALSE)
{
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
ChangeTracking_DoFullCompactingRoundIfNeeded();
/*
* Periodically check if compacting is required.
* Periodic compacting is required in order to
* a) reduce space for change tracking log file
* b) reduce time for transition from Change Tracking to Resync
*/
FileRep_InsertConfigLogEntry("run change tracking compacting");
while (1) {
FileRepSubProcess_ProcessSignals();
while (FileRepSubProcess_GetState() == FileRepStateFault ||
segmentState == SegmentStateChangeTrackingDisabled)
{
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
if (! (FileRepSubProcess_GetState() == FileRepStateReady &&
dataState == DataStateInChangeTracking))
{
break;
}
Insist(fileRepRole == FileRepPrimaryRole);
Insist(dataState == DataStateInChangeTracking);
Insist(FileRepSubProcess_GetState() == FileRepStateReady);
/* retry compacting of change tracking log files once per minute */
pg_usleep(50000L); /* 50 ms */
if (++retry == 1200)
{
ChangeTracking_CompactLogsIfPossible();
retry=0;
}
}
}
/*
*
* FileRepPrimary_StartRecoveryInChangeTracking()
*
*/
static void
FileRepPrimary_StartRecoveryInChangeTracking(void)
{
FileRep_InsertConfigLogEntry("run recovery");
while (1) {
while (FileRepSubProcess_GetState() == FileRepStateFault) {
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
if (FileRepSubProcess_GetState() == FileRepStateShutdown ||
FileRepSubProcess_GetState() == FileRepStateShutdownBackends) {
break;
}
Insist(fileRepRole == FileRepPrimaryRole);
Insist(dataState == DataStateInChangeTracking);
Insist(FileRepSubProcess_GetState() != FileRepStateReady);
if (ChangeTracking_RetrieveIsTransitionToInsync())
{
ChangeTracking_DropAll();
}
else
{
if (ChangeTracking_RetrieveIsTransitionToResync() == FALSE &&
isFullResync())
{
ChangeTracking_MarkFullResync();
/* segmentState == SegmentStateChangeTrackingDisabled */
getFileRepRoleAndState(&fileRepRole, &segmentState, &dataState, NULL, NULL);
Assert(segmentState == SegmentStateChangeTrackingDisabled);
/* database is resumed */
primaryMirrorSetIOSuspended(FALSE);
FileRep_InsertConfigLogEntry("change tracking recovery completed");
break;
}
else
{
ChangeTracking_MarkIncrResync();
}
}
XLogInChangeTrackingTransition();
/* NOTE: Any error during change tracking will result in disabling Change Tracking */
FileRepSubProcess_SetState(FileRepStateReady);
/* database is resumed */
primaryMirrorSetIOSuspended(FALSE);
FileRep_InsertConfigLogEntry("change tracking recovery completed");
break;
} // while(1)
}
/*
*
* FileRepPrimary_RunRecoveryInSync()
*
* 1) Recover Flat Files
* a) pg_control file
* b) pg_database file
* c) pg_auth file
* d) pg_twophase directory
* e) Slru directories
* *) pg_clog
* *) pg_multixact
* *) pg_distributedlog
* *) pg_distributedxidmap
* *) pg_subtrans
*
* 2) Reconcile xlog EOF
*
*/
static int
FileRepPrimary_RunRecoveryInSync(void)
{
int status = STATUS_OK;
FileRep_InsertConfigLogEntry("run recovery of flat files");
while (1) {
status = XLogRecoverMirrorControlFile();
if (status != STATUS_OK) {
break;
}
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
status = XLogReconcileEofPrimary();
if (status != STATUS_OK) {
break;
}
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
MirroredFlatFile_DropTemporaryFiles();
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
MirroredFlatFile_MirrorDropTemporaryFiles();
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
status = FlatFilesRecoverMirror();
if (status != STATUS_OK) {
break;
}
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
status = TwoPhaseRecoverMirror();
if (status != STATUS_OK) {
break;
}
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
status = SlruRecoverMirror();
if (status != STATUS_OK) {
break;
}
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
FileRepSubProcess_SetState(FileRepStateReady);
break;
}
return status;
}
/*
* SenderLoop
*
*/
static int
FileRepAckMirror_RunSender(void)
{
FileRepShmemMessageDescr_s *fileRepShmemMessageDescr=NULL;
char *fileRepMessage;
int status = STATUS_OK;
bool movePositionConsume = FALSE;
FileRepConsumerProcIndex_e messageType;
FileRepMessageHeader_s *fileRepMessageHeader;
FileRepShmem_s *fileRepAckShmem = NULL;
FileRep_InsertConfigLogEntry("run sender ack");
fileRepAckShmem = fileRepAckShmemArray[FILEREP_OUTGOING_MESSAGE_QUEUE];
while (1) {
LWLockAcquire(FileRepAckShmemLock, LW_EXCLUSIVE);
if (movePositionConsume) {
fileRepAckShmem->positionConsume =
fileRepAckShmem->positionConsume +
fileRepShmemMessageDescr->messageLength +
sizeof(FileRepShmemMessageDescr_s);
if (fileRepAckShmem->positionConsume == fileRepAckShmem->positionWraparound &&
fileRepAckShmem->positionInsert != fileRepAckShmem->positionWraparound) {
fileRepAckShmem->positionConsume = fileRepAckShmem->positionBegin;
fileRepAckShmem->positionWraparound = fileRepAckShmem->positionEnd;
}
FileRep_IpcSignal(fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->semP,
&fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->refCountSemP);
}
fileRepShmemMessageDescr =
(FileRepShmemMessageDescr_s*) fileRepAckShmem->positionConsume;
while ((fileRepAckShmem->positionConsume == fileRepAckShmem->positionInsert) ||
((fileRepAckShmem->positionConsume != fileRepAckShmem->positionInsert) &&
(fileRepShmemMessageDescr->messageState != FileRepShmemMessageStateReady))) {
fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->refCountSemC++;
LWLockRelease(FileRepAckShmemLock);
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateReady) {
LWLockAcquire(FileRepAckShmemLock, LW_EXCLUSIVE);
break;
}
FileRep_IpcWait(fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->semC, &fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->refCountSemC, FileRepAckShmemLock);
LWLockAcquire(FileRepAckShmemLock, LW_EXCLUSIVE);
if (fileRepAckShmem->positionConsume == fileRepAckShmem->positionWraparound &&
fileRepAckShmem->positionInsert != fileRepAckShmem->positionWraparound) {
fileRepAckShmem->positionConsume = fileRepAckShmem->positionBegin;
fileRepAckShmem->positionWraparound = fileRepAckShmem->positionEnd;
}
/* Re-assign to find if messageState is changed */
fileRepShmemMessageDescr =
(FileRepShmemMessageDescr_s*) fileRepAckShmem->positionConsume;
} // while internal
fileRepAckShmem->consumeCount++;
LWLockRelease(FileRepAckShmemLock);
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateReady) {
break;
}
FileRep_InsertLogEntry(
"M_RunSenderAck",
FileRep_GetFlatFileIdentifier("", ""),
FileRepRelationTypeNotSpecified,
FileRepOperationNotSpecified,
FILEREP_UNDEFINED,
FILEREP_UNDEFINED,
FileRepAckStateNotInitialized,
FILEREP_UNDEFINED,
FILEREP_UNDEFINED);
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FileRepSender,
DDLNotSpecified,
"", //databaseName
""); // tableName
#endif
fileRepMessage = (char*) (fileRepAckShmem->positionConsume +
sizeof(FileRepShmemMessageDescr_s));
fileRepMessageHeader = (FileRepMessageHeader_s*) (fileRepAckShmem->positionConsume +
sizeof(FileRepShmemMessageDescr_s));
messageType = FileRepMessageTypeXLog;
if (! FileRepConnClient_SendMessage(
messageType,
fileRepShmemMessageDescr->messageSync,
fileRepMessage,
fileRepShmemMessageDescr->messageLength))
{
ereport(WARNING,
(errcode_for_socket_access(),
errmsg("mirror failure, "
"could not sent ack message to primary : %m, "
"failover requested"),
errhint("run gprecoverseg to re-establish mirror connectivity"),
FileRep_errdetail_ShmemAck(),
FileRep_errcontext()));
status = STATUS_ERROR;
break;
}
movePositionConsume = TRUE;
} // while(1)
FileRepConnClient_CloseConnection();
return status;
}
/*
*
* FileRepPrimary_StartSender
*/
void
FileRepAckMirror_StartSender(void)
{
int status = STATUS_OK;
int retry = 0;
struct timeval currentTime;
pg_time_t beginTime = 0;
pg_time_t endTime = 0;
FileRep_InsertConfigLogEntry("start sender ack");
while (1) {
if (status != STATUS_OK) {
FileRep_SetSegmentState(SegmentStateFault, FaultTypeMirror);
FileRepSubProcess_SetState(FileRepStateFault);
}
while (FileRepSubProcess_GetState() == FileRepStateInitialization ||
FileRepSubProcess_GetState() == FileRepStateFault ||
(fileRepShmemArray[0]->state == FileRepStateNotInitialized &&
FileRepSubProcess_GetState() != FileRepStateShutdown )) {
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
if (FileRepSubProcess_GetState() == FileRepStateShutdown) {
break;
}
{
char tmpBuf[FILEREP_MAX_LOG_DESCRIPTION_LEN];
snprintf(tmpBuf, sizeof(tmpBuf), "primary address(port) '%s(%d)' mirror address(port) '%s(%d)' ",
fileRepPrimaryHostAddress,
fileRepPrimaryPort,
fileRepMirrorHostAddress,
fileRepMirrorPort);
FileRep_InsertConfigLogEntry(tmpBuf);
}
Insist(fileRepRole == FileRepMirrorRole);
status = FileRepConnClient_EstablishConnection(
fileRepPrimaryHostAddress,
fileRepPrimaryPort,
FALSE /* reportError */);
if (status != STATUS_OK)
{
gettimeofday(¤tTime, NULL);
beginTime = (pg_time_t) currentTime.tv_sec;
}
while (status != STATUS_OK &&
FileRep_IsRetry(retry) &&
(endTime - beginTime) < gp_segment_connect_timeout)
{
FileRep_Sleep10ms(retry);
FileRep_IncrementRetry(retry);
gettimeofday(¤tTime, NULL);
endTime = (pg_time_t) currentTime.tv_sec;
status = FileRepConnClient_EstablishConnection(
fileRepPrimaryHostAddress,
fileRepPrimaryPort,
(retry == file_rep_retry && file_rep_retry != 0) ||
((endTime - beginTime) > gp_segment_connect_timeout) ? TRUE : FALSE);
if (FileRepSubProcess_IsStateTransitionRequested())
{
break;
}
}
if (status != STATUS_OK) {
continue;
}
FileRep_SetFileRepRetry();
status = FileRepAckMirror_RunSender();
} // while(1)
FileRepConnClient_CloseConnection();
return;
}
/*
* FileRepAckPrimary_RunConsumer()
*/
static int
FileRepAckPrimary_RunConsumer(void)
{
FileRepShmemMessageDescr_s *fileRepShmemMessageDescr = NULL;
FileRepMessageHeader_s *fileRepMessageHeader = NULL;
pg_crc32 *fileRepMessageHeaderCrc;
pg_crc32 messageHeaderCrcLocal = 0;
int status = STATUS_OK;
bool movePositionConsume = FALSE;
FileRepShmem_s *fileRepAckShmem = NULL;
FileRep_InsertConfigLogEntry("run consumer");
fileRepAckShmem = fileRepAckShmemArray[FILEREP_ACKSHMEM_MESSAGE_SLOT_PRIMARY_ACK];
while (1) {
LWLockAcquire(FileRepAckShmemLock, LW_EXCLUSIVE);
if (movePositionConsume) {
fileRepAckShmem->positionConsume =
fileRepAckShmem->positionConsume +
fileRepShmemMessageDescr->messageLength +
sizeof(FileRepShmemMessageDescr_s);
if (fileRepAckShmem->positionConsume == fileRepAckShmem->positionWraparound &&
fileRepAckShmem->positionInsert != fileRepAckShmem->positionWraparound) {
fileRepAckShmem->positionConsume = fileRepAckShmem->positionBegin;
fileRepAckShmem->positionWraparound = fileRepAckShmem->positionEnd;
}
FileRep_IpcSignal(fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->semP,
&fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->refCountSemP);
}
fileRepShmemMessageDescr =
(FileRepShmemMessageDescr_s*) fileRepAckShmem->positionConsume;
while ((fileRepAckShmem->positionConsume == fileRepAckShmem->positionInsert) ||
((fileRepAckShmem->positionConsume != fileRepAckShmem->positionInsert) &&
(fileRepShmemMessageDescr->messageState != FileRepShmemMessageStateReady))) {
fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->refCountSemC++;
LWLockRelease(FileRepAckShmemLock);
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateReady &&
FileRepSubProcess_GetState() != FileRepStateInitialization) {
LWLockAcquire(FileRepAckShmemLock, LW_EXCLUSIVE);
break;
}
FileRep_IpcWait(fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->semC, &fileRepIpcArray[fileRepAckShmem->ipcArrayIndex]->refCountSemC, FileRepAckShmemLock);
LWLockAcquire(FileRepAckShmemLock, LW_EXCLUSIVE);
if (fileRepAckShmem->positionConsume == fileRepAckShmem->positionWraparound &&
fileRepAckShmem->positionInsert != fileRepAckShmem->positionWraparound) {
fileRepAckShmem->positionConsume = fileRepAckShmem->positionBegin;
fileRepAckShmem->positionWraparound = fileRepAckShmem->positionEnd;
}
/* Re-assign to find if messageState is changed */
fileRepShmemMessageDescr =
(FileRepShmemMessageDescr_s*) fileRepAckShmem->positionConsume;
} // internal while
fileRepAckShmem->consumeCount++;
LWLockRelease(FileRepAckShmemLock);
FileRepSubProcess_ProcessSignals();
if (FileRepSubProcess_GetState() != FileRepStateReady &&
FileRepSubProcess_GetState() != FileRepStateInitialization) {
break;
}
SIMPLE_FAULT_INJECTOR(FileRepConsumer);
/* Calculate and compare FileRepMessageHeader_s Crc */
fileRepMessageHeader = (FileRepMessageHeader_s*) (fileRepAckShmem->positionConsume +
sizeof(FileRepShmemMessageDescr_s));
FileRep_CalculateCrc((char *) fileRepMessageHeader,
sizeof(FileRepMessageHeader_s),
&messageHeaderCrcLocal);
fileRepMessageHeaderCrc =
(pg_crc32 *) (fileRepAckShmem->positionConsume +
sizeof(FileRepMessageHeader_s) +
sizeof(FileRepShmemMessageDescr_s));
if (*fileRepMessageHeaderCrc != messageHeaderCrcLocal)
{
status = STATUS_ERROR;
ereport(WARNING,
(errmsg("mirror failure, "
"could not match ack message header checksum between primary '%u' and mirror '%u', "
"failover requested",
*fileRepMessageHeaderCrc,
messageHeaderCrcLocal),
errhint("run gprecoverseg to re-establish mirror connectivity"),
FileRep_errdetail(fileRepMessageHeader->fileRepIdentifier,
fileRepMessageHeader->fileRepRelationType,
fileRepMessageHeader->fileRepOperation,
fileRepMessageHeader->messageCount),
FileRep_errdetail_ShmemAck(),
FileRep_errcontext()));
break;
}
/* Write operation is never acknowledged.
* That means message should never have body.
* CRC of body should be always 0.
*/
Assert(fileRepMessageHeader->fileRepOperation != FileRepOperationWrite);
Assert(fileRepMessageHeader->fileRepMessageBodyCrc == 0);
switch (fileRepMessageHeader->fileRepOperation)
{
case FileRepOperationReconcileXLogEof:
xLogEof = fileRepMessageHeader->fileRepOperationDescription.reconcile.xLogEof;
if (Debug_filerep_print)
ereport(LOG,
(errmsg("ack reconcile xlogid '%d' xrecoff '%d' ",
xLogEof.xlogid,
xLogEof.xrecoff)));
break;
case FileRepOperationValidation:
mirrorStatus = fileRepMessageHeader->fileRepOperationDescription.validation.mirrorStatus;
if (Debug_filerep_print)
ereport(LOG,
(errmsg("ack validation status '%s' ",
FileRepStatusToString[mirrorStatus])));
break;
case FileRepOperationCreate:
mirrorStatus = fileRepMessageHeader->fileRepOperationDescription.create.mirrorStatus;
if (Debug_filerep_print)
ereport(LOG,
(errmsg("ack create status '%s' ",
FileRepStatusToString[mirrorStatus])));
break;
case FileRepOperationStartSlruChecksum:
mirrorStatus =
fileRepMessageHeader->fileRepOperationDescription.startChecksum.mirrorStatus;
if (Debug_filerep_print)
{
ereport(LOG,
(errmsg("ack start SLRU checksum: status = '%s', directory = '%s' ",
FileRepStatusToString[mirrorStatus],
fileRepMessageHeader->fileRepIdentifier.fileRepFlatFileIdentifier.directorySimpleName)));
}
break;
case FileRepOperationVerifySlruDirectoryChecksum:
mirrorStatus =
fileRepMessageHeader->fileRepOperationDescription.verifyDirectoryChecksum.mirrorStatus;
if (Debug_filerep_print)
{
ereport(LOG,
(errmsg("ack verify SLRU directory checksum: status = '%s', directory = '%s' ",
FileRepStatusToString[mirrorStatus],
fileRepMessageHeader->fileRepIdentifier.fileRepFlatFileIdentifier.directorySimpleName)));
}
break;
default:
break;
}
if (fileRepMessageHeader->fileRepAckState != FileRepAckStateCompleted) {
status = STATUS_ERROR;
ereport(WARNING,
(errmsg("mirror failure, "
"could not complete operation on mirror ack state '%s', "
"failover requested",
FileRepAckStateToString[fileRepMessageHeader->fileRepAckState]),
errhint("run gprecoverseg to re-establish mirror connectivity"),
errSendAlert(true),
FileRep_errdetail(fileRepMessageHeader->fileRepIdentifier,
fileRepMessageHeader->fileRepRelationType,
fileRepMessageHeader->fileRepOperation,
fileRepMessageHeader->messageCount),
FileRep_errdetail_Shmem(),
FileRep_errdetail_ShmemAck(),
FileRep_errcontext()));
/*
* FAULT has to be set before entry is updated in ack hash table
* in order to suspend backend process.
*/
FileRep_SetSegmentState(SegmentStateFault, FaultTypeMirror);
FileRepSubProcess_ProcessSignals();
}
if (FileRepAckPrimary_UpdateHashEntry(
fileRepMessageHeader->fileRepIdentifier,
fileRepMessageHeader->fileRepRelationType,
fileRepMessageHeader->fileRepAckState) != STATUS_OK) {
status = STATUS_ERROR;
ereport(WARNING,
(errmsg("mirror failure, "
"could not update ack state '%s' in ack hash table, "
"failover requested",
FileRepAckStateToString[fileRepMessageHeader->fileRepAckState]),
errhint("run gprecoverseg to re-establish mirror connectivity"),
errSendAlert(true),
FileRep_errdetail(fileRepMessageHeader->fileRepIdentifier,
fileRepMessageHeader->fileRepRelationType,
fileRepMessageHeader->fileRepOperation,
fileRepMessageHeader->messageCount),
FileRep_errdetail_Shmem(),
FileRep_errdetail_ShmemAck(),
FileRep_errcontext()));
}
FileRep_InsertLogEntry(
"P_RunConsumer",
fileRepMessageHeader->fileRepIdentifier,
fileRepMessageHeader->fileRepRelationType,
fileRepMessageHeader->fileRepOperation,
messageHeaderCrcLocal,
fileRepMessageHeader->fileRepMessageBodyCrc,
fileRepMessageHeader->fileRepAckState,
FILEREP_UNDEFINED,
fileRepMessageHeader->messageCount);
if (status != STATUS_OK) {
break;
}
movePositionConsume = TRUE;
} // while(1)
return status;
}
bool
FileRepSubProcess_IsStateTransitionRequested(void)
{
bool isStateTransitionRequested = FALSE;
getFileRepRoleAndState(&fileRepRole, &segmentState, &dataState, NULL, NULL);
switch (fileRepProcessType)
{
case FileRepProcessTypeMain:
/* Handle Shutdown request */
if (segmentState == SegmentStateImmediateShutdown)
{
isStateTransitionRequested = TRUE;
}
break;
case FileRepProcessTypeNotInitialized:
if (segmentState == SegmentStateShutdownFilerepBackends &&
fileRepShmemArray[0]->state == FileRepStateFault)
{
FileRep_InsertConfigLogEntry("failure is detected in segment mirroring during backend shutdown, abort requested");
}
/* no break */
default:
if (fileRepProcessType != FileRepProcessTypeNotInitialized)
{
FileRepSubProcess_ProcessSignals();
}
if (dataState == DataStateInChangeTracking)
{
isStateTransitionRequested = TRUE;
}
switch (segmentState)
{
case SegmentStateFault:
case SegmentStateImmediateShutdown:
case SegmentStateShutdown:
isStateTransitionRequested = TRUE;
break;
default:
break;
}
break;
}
if (isStateTransitionRequested)
{
FileRep_InsertConfigLogEntry("state transition requested ");
}
return isStateTransitionRequested;
}
void
FileRepSubProcess_Main()
{
const char *statmsg;
MemoryContext fileRepSubProcessMemoryContext;
sigjmp_buf local_sigjmp_buf;
MyProcPid = getpid();
MyStartTime = time(NULL);
/*
* Create a PGPROC so we can use LWLocks in FileRep sub-processes. The
* routine also register clean up at process exit
*/
InitAuxiliaryProcess();
InitBufferPoolBackend();
FileRepSubProcess_ConfigureSignals();
/*
* If an exception is encountered, processing resumes here.
*
* See notes in postgres.c about the design of this coding.
*/
if (sigsetjmp(local_sigjmp_buf, 1) != 0)
{
/* Prevents interrupts while cleaning up */
HOLD_INTERRUPTS();
/* Report the error to the server log */
EmitErrorReport();
LWLockReleaseAll();
if (FileRepPrimary_IsResyncManagerOrWorker())
{
LockReleaseAll(DEFAULT_LOCKMETHOD, false);
}
if (FileRepIsBackendSubProcess(fileRepProcessType))
{
AbortBufferIO();
UnlockBuffers();
/* buffer pins are released here: */
ResourceOwnerRelease(CurrentResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
false, true);
}
/*
* We can now go away. Note that because we'll call InitProcess, a
* callback will be registered to do ProcKill, which will clean up
* necessary state.
*/
proc_exit(0);
}
/* We can now handle ereport(ERROR) */
PG_exception_stack = &local_sigjmp_buf;
PG_SETMASK(&UnBlockSig);
/*
* Identify myself via ps
*/
statmsg = FileRepProcessTypeToString[fileRepProcessType];
init_ps_display(statmsg, "", "", "");
/* Create the memory context where cross-transaction state is stored */
fileRepSubProcessMemoryContext = AllocSetContextCreate(TopMemoryContext,
"filerep subprocess memory context",
ALLOCSET_DEFAULT_MINSIZE,
ALLOCSET_DEFAULT_INITSIZE,
ALLOCSET_DEFAULT_MAXSIZE);
MemoryContextSwitchTo(fileRepSubProcessMemoryContext);
stateChangeRequestCounter++;
FileRepSubProcess_ProcessSignals();
switch (fileRepProcessType)
{
case FileRepProcessTypePrimarySender:
FileRepPrimary_StartSender();
break;
case FileRepProcessTypeMirrorReceiver:
FileRepMirror_StartReceiver();
break;
case FileRepProcessTypeMirrorConsumer:
case FileRepProcessTypeMirrorConsumerWriter:
case FileRepProcessTypeMirrorConsumerAppendOnly1:
FileRepMirror_StartConsumer();
break;
case FileRepProcessTypeMirrorSenderAck:
FileRepAckMirror_StartSender();
break;
case FileRepProcessTypePrimaryReceiverAck:
FileRepAckPrimary_StartReceiver();
break;
case FileRepProcessTypePrimaryConsumerAck:
FileRepAckPrimary_StartConsumer();
break;
case FileRepProcessTypePrimaryRecovery:
FileRepSubProcess_InitProcess();
/*
* At this point, database is starting up and xlog is not yet
* replayed. Initializing relcache now is dangerous, a sequential
* scan of catalog tables may end up with incorrect hint bits.
* E.g. a committed transaction's dirty heap pages made it to disk
* but pg_clog update was still in memory and we crashed. If a
* tuple inserted by this transaction is read during relcache
* initialization, status of the tuple's xmin will be incorrectly
* determined as "not commited" from pg_clog. And
* HEAP_XMIN_INVALID hint bit will be set, rendering the tuple
* perpetually invisible. Relcache initialization must be
* deferred to only after all of xlog has been replayed.
*/
FileRepPrimary_StartRecovery();
ResourceOwnerRelease(CurrentResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
false, true);
break;
case FileRepProcessTypeResyncManager:
FileRepSubProcess_InitProcess();
FileRepPrimary_StartResyncManager();
ResourceOwnerRelease(CurrentResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
false, true);
break;
case FileRepProcessTypeResyncWorker1:
case FileRepProcessTypeResyncWorker2:
case FileRepProcessTypeResyncWorker3:
case FileRepProcessTypeResyncWorker4:
FileRepSubProcess_InitProcess();
FileRepPrimary_StartResyncWorker();
ResourceOwnerRelease(CurrentResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
false, true);
break;
default:
elog(PANIC, "unrecognized process type: %s(%d)",
statmsg, fileRepProcessType);
break;
}
switch (FileRepSubProcess_GetState())
{
case FileRepStateShutdown:
case FileRepStateReady:
proc_exit(0);
break;
default:
proc_exit(2);
break;
}
}
/*
*
* FileRepAckPrimary_StartReceiver
*/
void
FileRepAckPrimary_StartReceiver(void)
{
int status = STATUS_OK;
struct timeval currentTime;
pg_time_t beginTime = 0;
pg_time_t endTime = 0;
int retval = 0;
FileRep_InsertConfigLogEntry("start receiver ack");
{
char tmpBuf[FILEREP_MAX_LOG_DESCRIPTION_LEN];
snprintf(tmpBuf, sizeof(tmpBuf), "primary address(port) '%s(%d)' mirror address(port) '%s(%d)' ",
fileRepPrimaryHostAddress,
fileRepPrimaryPort,
fileRepMirrorHostAddress,
fileRepMirrorPort);
FileRep_InsertConfigLogEntry(tmpBuf);
}
FileRepAckPrimary_ShmemReInit();
Insist(fileRepRole == FileRepPrimaryRole);
if (filerep_inject_listener_fault)
{
status = STATUS_ERROR;
ereport(WARNING,
(errmsg("mirror failure, "
"injected fault by guc filerep_inject_listener_fault, "
"failover requested"),
FileRep_errcontext()));
FileRep_SetSegmentState(SegmentStateFault, FaultTypeMirror);
FileRepSubProcess_SetState(FileRepStateFault);
FileRepSubProcess_ProcessSignals();
return;
}
status = FileRepConnServer_StartListener(
fileRepPrimaryHostAddress,
fileRepPrimaryPort);
gettimeofday(¤tTime, NULL);
beginTime = (pg_time_t) currentTime.tv_sec;
while (1) {
if (status != STATUS_OK)
{
FileRep_SetSegmentState(SegmentStateFault, FaultTypeMirror);
FileRepSubProcess_SetState(FileRepStateFault);
}
while (FileRepSubProcess_GetState() == FileRepStateFault) {
FileRepSubProcess_ProcessSignals();
pg_usleep(50000L); /* 50 ms */
}
if (FileRepSubProcess_GetState() == FileRepStateShutdown) {
break;
}
PG_SETMASK(&BlockSig);
retval = FileRepConnServer_Select();
PG_SETMASK(&UnBlockSig);
gettimeofday(¤tTime, NULL);
endTime = (pg_time_t) currentTime.tv_sec;
if ((endTime - beginTime) > gp_segment_connect_timeout)
{
ereport(WARNING,
(errmsg("mirror failure, "
"no connection was established from client from mirror, "
"primary address(port) '%s(%d)' mirror address(port) '%s(%d)' timeout reached '%d' "
"failover requested",
fileRepPrimaryHostAddress,
fileRepPrimaryPort,
fileRepMirrorHostAddress,
fileRepMirrorPort,
gp_segment_connect_timeout),
errSendAlert(true),
FileRep_errcontext()));
status = STATUS_ERROR;
continue;
}
/*
* check and process any signals received
* The routine returns TRUE if the received signal requests
* process shutdown.
*/
if (FileRepSubProcess_ProcessSignals()) {
continue;
}
if (retval < 0) {
status = STATUS_ERROR;
continue;
}
if (retval == 0) {
continue;
}
Assert(retval > 0);
status = FileRepConnServer_CreateConnection();
if (status != STATUS_OK) {
continue;
}
status = FileRepConnServer_ReceiveStartupPacket();
if (status != STATUS_OK) {
continue;
}
fileRepShmemArray[0]->state = FileRepStateInitialization;
status = FileRepAckPrimary_RunReceiver();
} // while(1)
FileRepConnServer_CloseConnection();
return;
}
