void PersistentRelation_AddCreated(
RelFileNode *relFileNode,
/* The tablespace, database, and relation OIDs for the create. */
int32 segmentFileNum,
PersistentFileSysRelStorageMgr relStorageMgr,
PersistentFileSysRelBufpoolKind relBufpoolKind,
MirroredObjectExistenceState mirrorExistenceState,
MirroredRelDataSynchronizationState relDataSynchronizationState,
int64 mirrorAppendOnlyLossEof,
int64 mirrorAppendOnlyNewEof,
char *relationName,
ItemPointer persistentTid,
/* Resulting TID of the gp_persistent_rel_files tuple for the relation. */
int64 *persistentSerialNum,
/* Resulting serial number for the relation. Distinquishes the uses of the tuple. */
bool flushToXLog)
/* When true, the XLOG record for this change will be flushed to disk. */
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
XLogRecPtr mirrorBufpoolResyncCkptLoc;
ItemPointerData previousFreeTid;
Datum values[Natts_gp_persistent_relation_node];
if(RelFileNode_IsEmpty(relFileNode))
elog(ERROR, "Invalid RelFileNode (0,0,0)");
MemSet(&previousFreeTid, 0, sizeof(ItemPointerData));
MemSet(&mirrorBufpoolResyncCkptLoc, 0, sizeof(XLogRecPtr));
if (!Persistent_BeforePersistenceWork())
elog(ERROR, "We can only add to persistent meta-data when special states");
// Verify PersistentFileSysObj_BuildInitScan has been called.
PersistentRelation_VerifyInitScan();
PersistentFileSysObjName_SetRelationFile(
&fsObjName,
relFileNode,
segmentFileNum);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
GpPersistentRelationNode_SetDatumValues(
values,
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode,
segmentFileNum,
relStorageMgr,
PersistentFileSysState_Created,
/* createMirrorDataLossTrackingSessionNum */ 0,
mirrorExistenceState,
relDataSynchronizationState,
/* mirrorBufpoolMarkedForScanIncrementalResync */ false,
/* mirrorBufpoolResyncChangedPageCount */ 0,
&mirrorBufpoolResyncCkptLoc,
/* mirrorBufpoolResyncCkptBlockNum */ 0,
mirrorAppendOnlyLossEof,
mirrorAppendOnlyNewEof,
relBufpoolKind,
InvalidTransactionId,
/* persistentSerialNum */ 0, // This will be set by PersistentFileSysObj_AddTuple.
&previousFreeTid);
PersistentFileSysObj_AddTuple(
PersistentFsObjType_RelationFile,
values,
flushToXLog,
persistentTid,
persistentSerialNum);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent relation: Add '%s', relation name '%s', in state 'Created', relation storage manager '%s', mirror existence state '%s', relation data resynchronization state '%s', serial number " INT64_FORMAT " at TID %s",
PersistentFileSysObjName_ObjectName(&fsObjName),
relationName,
PersistentFileSysRelStorageMgr_Name(relStorageMgr),
MirroredObjectExistenceState_Name(mirrorExistenceState),
MirroredRelDataSynchronizationState_Name(relDataSynchronizationState),
*persistentSerialNum,
ItemPointerToString(persistentTid));
}
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;
}
/*
* Indicate we intend to create a relation file as part of the current transaction.
*
* This function adds an entry in 'gp_persistent_relation_node' for either a new table (segment file
* # 0) or a new segment file under AO table (segment file # > 0 for row/column-oriented AO) with a state
* 'Create Pending'. An XLOG IntentToCreate record is generated that will guard the subsequent file-system
* create in case the transaction aborts.
*
* Paramaters
* -----------
* relFileNode = The tablespace, database, and relation OIDs for the create
* segmentFileNum = As the name implies ( 0 for heap
* >= 0 for RO/CO AO as applicable)
* relStorageMgr = Persistent Relation storage Manager
* relBufpoolKind = Buffer pool type beneath corrosponding relation
* TODO bufferPollBulkLoad = ???
* TODO mirrorExistenceState = ???
* TODO relDataSynchronizationState = ???
* flushToXlog = If true, the XLOG record for this change will be flushed to disk.
* TODO isLocalBuf = ???
*
* Return
* ------
* relationName = Name of the relation used for either debugging or to store in PendingDelete LL.
* persistentTid = Resulting TID of the gp_persistent_rel_files tuple for the relation
* serialNum = Resulting serial number for the relation. Distinquishes the uses of the tuple
*/
void PersistentRelation_AddCreatePending(
RelFileNode *relFileNode,
int32 segmentFileNum,
PersistentFileSysRelStorageMgr relStorageMgr,
PersistentFileSysRelBufpoolKind relBufpoolKind,
bool bufferPoolBulkLoad,
MirroredObjectExistenceState mirrorExistenceState,
MirroredRelDataSynchronizationState relDataSynchronizationState,
char *relationName,
ItemPointer persistentTid,
int64 *serialNum,
bool flushToXLog,
bool isLocalBuf)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
XLogRecPtr mirrorBufpoolResyncCkptLoc;
ItemPointerData previousFreeTid;
Datum values[Natts_gp_persistent_relation_node];
if(RelFileNode_IsEmpty(relFileNode))
elog(ERROR, "Invalid RelFileNode (0,0,0)");
MemSet(&previousFreeTid, 0, sizeof(ItemPointerData));
MemSet(&mirrorBufpoolResyncCkptLoc, 0, sizeof(XLogRecPtr));
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent relation '%s' because we are before persistence work",
relpath(*relFileNode));
MemSet(persistentTid, 0, sizeof(ItemPointerData));
*serialNum = 0;
return; // The initdb process will load the persistent table once we out of bootstrap mode.
}
/* Verify if the needed shared mem data structures for persistent tables are setup and inited */
PersistentRelation_VerifyInitScan();
/* Setup the file system object name */
PersistentFileSysObjName_SetRelationFile(
&fsObjName,
relFileNode,
segmentFileNum);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
/* Create a values array which will be used to create a 'gp_persistent_relation_node' tuple */
GpPersistentRelationNode_SetDatumValues(
values,
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode,
segmentFileNum,
relStorageMgr,
(bufferPoolBulkLoad ?
PersistentFileSysState_BulkLoadCreatePending :
PersistentFileSysState_CreatePending),
/* createMirrorDataLossTrackingSessionNum */ 0,
mirrorExistenceState,
relDataSynchronizationState,
/* mirrorBufpoolMarkedForScanIncrementalResync */ false,
/* mirrorBufpoolResyncChangedPageCount */ 0,
&mirrorBufpoolResyncCkptLoc,
/* mirrorBufpoolResyncCkptBlockNum */ 0,
/* mirrorAppendOnlyLossEof */ 0,
/* mirrorAppendOnlyNewEof */ 0,
relBufpoolKind,
GetTopTransactionId(),
/* persistentSerialNum */ 0, // This will be set by PersistentFileSysObj_AddTuple.
&previousFreeTid);
/* Add a new tuple to 'gp_persistent_relation_node' table for the new relation/segment file
* we intend to create. This will also create and apply a new persistent serial number. */
PersistentFileSysObj_AddTuple(
PersistentFsObjType_RelationFile,
values,
flushToXLog,
persistentTid,
serialNum);
/*
* This XLOG must be generated under the persistent write-lock.
*/
#ifdef MASTER_MIRROR_SYNC
mmxlog_log_create_relfilenode(
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode,
segmentFileNum);
#endif
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FaultBeforePendingDeleteRelationEntry,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
/* We'll add an entry to the PendingDelete LinkedList (LL) to remeber what we
* created in this transaction (or sub-transaction). If the transaction
* aborts then we can search for all such entries in this LL and get rid of (delete)
* such relations or segment files on the disk.
*
* MPP-18228
* To make adding 'Create Pending' entry to persistent table and adding
* to the PendingDelete list atomic
*/
PendingDelete_AddCreatePendingRelationEntry(
&fsObjName,
persistentTid,
serialNum,
relStorageMgr,
relationName,
isLocalBuf,
bufferPoolBulkLoad);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent relation: Add '%s', relation name '%s' in state 'Create Pending', relation storage manager '%s', mirror existence state '%s', relation data resynchronization state '%s', serial number " INT64_FORMAT " at TID %s",
PersistentFileSysObjName_ObjectName(&fsObjName),
relationName,
PersistentFileSysRelStorageMgr_Name(relStorageMgr),
MirroredObjectExistenceState_Name(mirrorExistenceState),
MirroredRelDataSynchronizationState_Name(relDataSynchronizationState),
*serialNum,
ItemPointerToString(persistentTid));
}