static void
PersistentBuild_NonTransactionTruncate(RelFileNode *relFileNode)
{
SMgrRelation smgrRelation;
PersistentFileSysObjName fsObjName;
PersistentFileSysObjName_SetRelationFile(
&fsObjName,
relFileNode,
/* segmentFileNum */ 0,
is_tablespace_shared);
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Non-transaction truncate of '%s'",
PersistentFileSysObjName_ObjectName(&fsObjName));
smgrRelation = smgropen(*relFileNode);
smgrtruncate(
smgrRelation,
0,
/* isTemp */ true,
/* isLocalBuf */ false,
/* persistentTid */ NULL,
/* persistentSerialNum */ 0);
smgrclose(smgrRelation);
}
/*
* Set mirror for all relation files.
*/
void
PersistentRelation_AddMirrorAll(int16 pridbid, int16 mirdbid)
{
Relation rel;
HeapScanDesc scandesc;
HeapTuple tuple;
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
if (Persistent_BeforePersistenceWork())
elog(ERROR, "persistent table changes forbidden");
rel = heap_open(GpPersistentRelationNodeRelationId,
AccessExclusiveLock);
scandesc = heap_beginscan(rel, SnapshotNow, 0, NULL);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
while ((tuple = heap_getnext(scandesc, ForwardScanDirection)) != NULL)
{
Form_gp_persistent_relation_node form =
(Form_gp_persistent_relation_node)GETSTRUCT(tuple);
Oid tblspcoid = form->tablespace_oid;
Oid dboid = form->database_oid;
Oid relfilenode_oid = form->relfilenode_oid;
int32 segment_file_num = form->segment_file_num;
int64 serial = form->persistent_serial_num;
PersistentFileSysObjName fsObjName;
RelFileNode node;
node.spcNode = tblspcoid;
node.dbNode = dboid;
node.relNode = relfilenode_oid;
PersistentFileSysObjName_SetRelationFile(&fsObjName,
&node,
segment_file_num);
PersistentFileSysObj_AddMirror(&fsObjName,
&tuple->t_self,
serial,
pridbid,
mirdbid,
NULL,
true,
/* flushToXlog */ false);
}
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
heap_endscan(scandesc);
heap_close(rel, NoLock);
}
void GpPersistent_GetCommonValues(
PersistentFsObjType fsObjType,
Datum *values,
PersistentFileSysObjName *fsObjName,
PersistentFileSysState *persistentState,
MirroredObjectExistenceState *mirrorExistenceState,
TransactionId *parentXid,
int64 *persistentSerialNum)
{
int64 createMirrorDataLossTrackingSessionNum;
int32 reserved;
switch (fsObjType)
{
case PersistentFsObjType_RelationFile:
{
RelFileNode relFileNode;
int32 segmentFileNum;
PersistentFileSysRelStorageMgr relationStorageManager;
MirroredRelDataSynchronizationState mirrorDataSynchronizationState;
bool mirrorBufpoolMarkedForScanIncrementalResync;
int64 mirrorBufpoolResyncChangedPageCount;
XLogRecPtr mirrorBufpoolResyncCkptLoc;
BlockNumber mirrorBufpoolResyncCkptBlockNum;
int64 mirrorAppendOnlyLossEof;
int64 mirrorAppendOnlyNewEof;
PersistentFileSysRelBufpoolKind relBufpoolKind;
GpPersistentRelationNode_GetValues(
values,
&relFileNode.spcNode,
&relFileNode.dbNode,
&relFileNode.relNode,
&segmentFileNum,
&relationStorageManager,
persistentState,
&createMirrorDataLossTrackingSessionNum,
mirrorExistenceState,
&mirrorDataSynchronizationState,
&mirrorBufpoolMarkedForScanIncrementalResync,
&mirrorBufpoolResyncChangedPageCount,
&mirrorBufpoolResyncCkptLoc,
&mirrorBufpoolResyncCkptBlockNum,
&mirrorAppendOnlyLossEof,
&mirrorAppendOnlyNewEof,
&relBufpoolKind,
parentXid,
persistentSerialNum);
PersistentFileSysObjName_SetRelationFile(
fsObjName,
&relFileNode,
segmentFileNum);
}
break;
case PersistentFsObjType_DatabaseDir:
{
DbDirNode dbDirNode;
GpPersistentDatabaseNode_GetValues(
values,
&dbDirNode.tablespace,
&dbDirNode.database,
persistentState,
&createMirrorDataLossTrackingSessionNum,
mirrorExistenceState,
&reserved,
parentXid,
persistentSerialNum);
PersistentFileSysObjName_SetDatabaseDir(
fsObjName,
dbDirNode.tablespace,
dbDirNode.database);
}
break;
case PersistentFsObjType_TablespaceDir:
{
Oid tablespaceOid;
Oid filespaceOid;
GpPersistentTablespaceNode_GetValues(
values,
&filespaceOid,
&tablespaceOid,
persistentState,
&createMirrorDataLossTrackingSessionNum,
mirrorExistenceState,
&reserved,
parentXid,
persistentSerialNum);
PersistentFileSysObjName_SetTablespaceDir(
fsObjName,
tablespaceOid);
}
break;
case PersistentFsObjType_FilespaceDir:
{
Oid filespaceOid;
int16 dbId1;
char locationBlankPadded1[FilespaceLocationBlankPaddedWithNullTermLen];
int16 dbId2;
char locationBlankPadded2[FilespaceLocationBlankPaddedWithNullTermLen];
GpPersistentFilespaceNode_GetValues(
values,
&filespaceOid,
&dbId1,
locationBlankPadded1,
&dbId2,
locationBlankPadded2,
persistentState,
&createMirrorDataLossTrackingSessionNum,
mirrorExistenceState,
&reserved,
parentXid,
persistentSerialNum);
PersistentFileSysObjName_SetFilespaceDir(
fsObjName,
filespaceOid);
}
break;
default:
elog(ERROR, "Unexpected persistent file-system object type: %d",
fsObjType);
}
}
/*
* Indicate we are aborting the create of a relation file.
*
* This state will make sure the relation gets dropped after a system crash.
*/
PersistentFileSysObjStateChangeResult PersistentRelation_MarkAbortingCreate(
RelFileNode *relFileNode,
/* The tablespace, database, and relation OIDs for the aborting create. */
int32 segmentFileNum,
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the relation. */
int64 persistentSerialNum,
/* Serial number for the relation. Distinquishes the uses of the tuple. */
bool retryPossible)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
PersistentFileSysObjStateChangeResult stateChangeResult;
if(RelFileNode_IsEmpty(relFileNode))
elog(ERROR, "Invalid RelFileNode (0,0,0)");
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent relation '%s' because we are before persistence work",
relpath(*relFileNode));
return false; // The initdb process will load the persistent table once we out of bootstrap mode.
}
/* MPP-16543: When inserting tuples into AO table, row numbers will be
* generated from gp_fastsequence catalog table, as part of the design,
* these sequence numbers are not reusable, even if the AO insert
* transaction is aborted. The entry in gp_fastsequence was inserted
* using frozen_heap_insert, which means it's always visible.
* Aborted AO insert transaction will cause inconsistency between
* gp_fastsequence and pg_class, the solution is to introduce "frozen
* delete" - inplace update tuple's MVCC header to make it invisible.
*/
Relation gp_fastsequence_rel = heap_open(FastSequenceRelationId, RowExclusiveLock);
HeapTuple tup;
SysScanDesc scan;
ScanKeyData skey;
ScanKeyInit(&skey,
Anum_gp_fastsequence_objid,
BTEqualStrategyNumber,
F_OIDEQ,
relFileNode->relNode);
scan = systable_beginscan(gp_fastsequence_rel,
InvalidOid,
false,
SnapshotNow,
1,
&skey);
while (HeapTupleIsValid(tup = systable_getnext(scan)))
{
Form_gp_fastsequence found = (Form_gp_fastsequence) GETSTRUCT(tup);
if (found->objid == relFileNode->relNode)
{
if (Debug_persistent_print)
{
elog(LOG, "frozen deleting gp_fastsequence entry for aborted AO insert transaction on relation %s", relpath(*relFileNode));
}
frozen_heap_inplace_delete(gp_fastsequence_rel, tup);
}
}
systable_endscan(scan);
heap_close(gp_fastsequence_rel, RowExclusiveLock);
PersistentRelation_VerifyInitScan();
PersistentFileSysObjName_SetRelationFile(
&fsObjName,
relFileNode,
segmentFileNum);
// Do this check after skipping out if in bootstrap mode.
if (PersistentStore_IsZeroTid(persistentTid))
elog(ERROR, "TID for persistent '%s' tuple for mark DROP pending is invalid (0,0)",
PersistentFileSysObjName_TypeAndObjectName(&fsObjName));
if (persistentSerialNum == 0)
elog(ERROR, "Persistent '%s' serial number for mark DROP pending is invalid (0)",
PersistentFileSysObjName_TypeAndObjectName(&fsObjName));
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
stateChangeResult =
PersistentFileSysObj_StateChange(
&fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_AbortingCreate,
retryPossible,
/* flushToXlog */ false,
/* oldState */ NULL,
/* verifiedActionCallback */ NULL);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent relation: '%s' changed state from 'Create Pending' to 'Aborting Create', serial number " INT64_FORMAT " at TID %s (State-Change result '%s')",
PersistentFileSysObjName_ObjectName(&fsObjName),
persistentSerialNum,
ItemPointerToString(persistentTid),
PersistentFileSysObjStateChangeResult_Name(stateChangeResult));
return stateChangeResult;
}
/*
* Indicate we intend to drop a relation file as part of the current transaction.
*
* This relation file to drop will be listed inside a commit, distributed commit, a distributed
* prepared, and distributed commit prepared XOG records.
*
* For any of the commit type records, once that XLOG record is flushed then the actual
* file-system delete will occur. The flush guarantees the action will be retried after system
* crash.
*/
PersistentFileSysObjStateChangeResult PersistentRelation_MarkDropPending(
RelFileNode *relFileNode,
/* The tablespace, database, and relation OIDs for the drop. */
int32 segmentFileNum,
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the relation. */
int64 persistentSerialNum,
/* Serial number for the relation. Distinquishes the uses of the tuple. */
bool retryPossible)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
PersistentFileSysState oldState;
PersistentFileSysObjStateChangeResult stateChangeResult;
if(RelFileNode_IsEmpty(relFileNode))
elog(ERROR, "Invalid RelFileNode (0,0,0)");
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent relation '%s' because we are before persistence work",
relpath(*relFileNode));
return false; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentRelation_VerifyInitScan();
PersistentFileSysObjName_SetRelationFile(
&fsObjName,
relFileNode,
segmentFileNum);
// Do this check after skipping out if in bootstrap mode.
if (PersistentStore_IsZeroTid(persistentTid))
elog(ERROR, "TID for persistent '%s' tuple for mark DROP pending is invalid (0,0)",
PersistentFileSysObjName_TypeAndObjectName(&fsObjName));
if (persistentSerialNum == 0)
elog(ERROR, "Persistent '%s' serial number for mark DROP pending is invalid (0)",
PersistentFileSysObjName_TypeAndObjectName(&fsObjName));
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
stateChangeResult =
PersistentFileSysObj_StateChange(
&fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_DropPending,
retryPossible,
/* flushToXlog */ false,
&oldState,
/* verifiedActionCallback */ NULL);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent relation: '%s' changed state from '%s' to 'Drop Pending', serial number " INT64_FORMAT " TID %s (State-Change result '%s')",
PersistentFileSysObjName_ObjectName(&fsObjName),
PersistentFileSysObjState_Name(oldState),
persistentSerialNum,
ItemPointerToString(persistentTid),
PersistentFileSysObjStateChangeResult_Name(stateChangeResult));
return stateChangeResult;
}
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));
}
/*
* 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));
}
void PersistentRelation_MarkBufPoolRelationForScanIncrementalResync(
RelFileNode *relFileNode,
/* The tablespace, database, and relation OIDs for the created relation. */
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the relation. */
int64 persistentSerialNum)
/* Serial number for the relation. Distinquishes the uses of the tuple. */
{
PersistentFileSysObjName fsObjName;
if(RelFileNode_IsEmpty(relFileNode))
elog(ERROR, "Invalid RelFileNode (0,0,0)");
if (GpPersistent_SkipXLogInfo(relFileNode->relNode))
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent relation '%s' because it is special",
relpath(*relFileNode));
return; // Resynchronize will always handle these relations as 'Scan Incremental'..
}
if (IsBootstrapProcessingMode())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent relation '%s' because we are in bootstrap mode",
relpath(*relFileNode));
return; // The initdb process will load the persistent table once we out of bootstrap mode.
}
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent relation '%s' because we are before persistence work",
relpath(*relFileNode));
return; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentRelation_VerifyInitScan();
PersistentFileSysObjName_SetRelationFile(
&fsObjName,
relFileNode,
/* segmentFileNum */ 0);
// Do this check after skipping out if in bootstrap mode.
if (PersistentStore_IsZeroTid(persistentTid))
elog(ERROR, "TID for persistent '%s' tuple for mark physically truncated is invalid (0,0)",
PersistentFileSysObjName_TypeAndObjectName(&fsObjName));
if (persistentSerialNum == 0)
elog(ERROR, "Persistent '%s' serial number for mark physcially truncated is invalid (0)",
PersistentFileSysObjName_TypeAndObjectName(&fsObjName));
PersistentFileSysObj_MarkBufPoolRelationForScanIncrementalResync(
&fsObjName,
persistentTid,
persistentSerialNum,
/* flushToXlog */ true);
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent relation: '%s' marked physically truncated, serial number " INT64_FORMAT " at TID %s",
PersistentFileSysObjName_ObjectName(&fsObjName),
persistentSerialNum,
ItemPointerToString(persistentTid));
}
void GpPersistent_GetCommonValues(
PersistentFsObjType fsObjType,
Datum *values,
PersistentFileSysObjName *fsObjName,
PersistentFileSysState *persistentState,
TransactionId *parentXid,
int64 *persistentSerialNum)
{
int32 reserved;
ItemPointerData previousFreeTid;
bool sharedStorage;
switch (fsObjType)
{
case PersistentFsObjType_RelationFile:
{
RelFileNode relFileNode;
int32 segmentFileNum;
PersistentFileSysRelStorageMgr relationStorageManager;
PersistentFileSysRelBufpoolKind relBufpoolKind;
GpPersistentRelfileNode_GetValues(
values,
&relFileNode.spcNode,
&relFileNode.dbNode,
&relFileNode.relNode,
&segmentFileNum,
&relationStorageManager,
persistentState,
&relBufpoolKind,
parentXid,
persistentSerialNum,
&previousFreeTid,
&sharedStorage);
PersistentFileSysObjName_SetRelationFile(
fsObjName,
&relFileNode,
segmentFileNum,
NULL);
fsObjName->hasInited = true;
fsObjName->sharedStorage = sharedStorage;
}
break;
case PersistentFsObjType_RelationDir:
{
RelFileNode relFileNode;
GpPersistentRelationNode_GetValues(
values,
&relFileNode.spcNode,
&relFileNode.dbNode,
&relFileNode.relNode,
persistentState,
&reserved,
parentXid,
persistentSerialNum,
&previousFreeTid,
&sharedStorage);
PersistentFileSysObjName_SetRelationDir(
fsObjName,
&relFileNode,
NULL);
fsObjName->hasInited = true;
fsObjName->sharedStorage = sharedStorage;
}
break;
case PersistentFsObjType_DatabaseDir:
{
DbDirNode dbDirNode;
GpPersistentDatabaseNode_GetValues(
values,
&dbDirNode.tablespace,
&dbDirNode.database,
persistentState,
&reserved,
parentXid,
persistentSerialNum,
&previousFreeTid,
&sharedStorage);
PersistentFileSysObjName_SetDatabaseDir(
fsObjName,
dbDirNode.tablespace,
dbDirNode.database,
NULL);
fsObjName->hasInited = true;
fsObjName->sharedStorage = sharedStorage;
}
break;
case PersistentFsObjType_TablespaceDir:
{
Oid tablespaceOid;
Oid filespaceOid;
GpPersistentTablespaceNode_GetValues(
values,
&filespaceOid,
&tablespaceOid,
persistentState,
&reserved,
parentXid,
persistentSerialNum,
&previousFreeTid,
&sharedStorage);
PersistentFileSysObjName_SetTablespaceDir(
fsObjName,
tablespaceOid,
NULL);
fsObjName->hasInited = true;
fsObjName->sharedStorage = sharedStorage;
}
break;
case PersistentFsObjType_FilespaceDir:
{
Oid filespaceOid;
int16 dbId1;
char locationBlankPadded1[FilespaceLocationBlankPaddedWithNullTermLen];
GpPersistentFilespaceNode_GetValues(
values,
&filespaceOid,
&dbId1,
locationBlankPadded1,
persistentState,
&reserved,
parentXid,
persistentSerialNum,
&previousFreeTid,
&sharedStorage);
PersistentFileSysObjName_SetFilespaceDir(
fsObjName,
filespaceOid,
NULL);
fsObjName->hasInited = true;
fsObjName->sharedStorage = sharedStorage;
}
break;
default:
elog(ERROR, "Unexpected persistent file-system object type: %d",
fsObjType);
}
}
static void PersistentBuild_ScanGpPersistentRelationNodeForGlobal(
Relation gp_relation_node,
int64 *count)
{
PersistentFileSysObjData *fileSysObjData;
PersistentFileSysObjSharedData *fileSysObjSharedData;
PersistentStoreScan storeScan;
Datum values[Natts_gp_persistent_relfile_node];
ItemPointerData persistentTid;
int64 persistentSerialNum;
PersistentFileSysObj_GetDataPtrs(
PersistentFsObjType_RelationFile,
&fileSysObjData,
&fileSysObjSharedData);
PersistentStore_BeginScan(
&fileSysObjData->storeData,
&fileSysObjSharedData->storeSharedData,
&storeScan);
while (PersistentStore_GetNext(
&storeScan,
values,
&persistentTid,
&persistentSerialNum))
{
RelFileNode relFileNode;
int32 segmentFileNum;
PersistentFileSysRelStorageMgr relationStorageManager;
PersistentFileSysState persistentState;
PersistentFileSysRelBufpoolKind relBufpoolKind;
TransactionId parentXid;
int64 serialNum;
ItemPointerData previousFreeTid;
PersistentFileSysObjName fsObjName;
bool sharedStorage;
GpPersistentRelfileNode_GetValues(
values,
&relFileNode.spcNode,
&relFileNode.dbNode,
&relFileNode.relNode,
&segmentFileNum,
&relationStorageManager,
&persistentState,
&relBufpoolKind,
&parentXid,
&serialNum,
&previousFreeTid,
&sharedStorage);
if (persistentState == PersistentFileSysState_Free)
continue;
PersistentFileSysObjName_SetRelationFile(
&fsObjName,
&relFileNode,
segmentFileNum,
NULL);
fsObjName.hasInited = true;
fsObjName.sharedStorage = sharedStorage;
if (relFileNode.spcNode != GLOBALTABLESPACE_OID)
continue;
if (relationStorageManager != PersistentFileSysRelStorageMgr_BufferPool)
elog(ERROR, "Only expecting global tables to be Buffer Pool managed");
InsertGpRelfileNodeTuple(
gp_relation_node,
relFileNode.relNode, // pg_class OID
/* relationName */ NULL, // Optional.
relFileNode.relNode, // pg_class relfilenode
/* segmentFileNum */ 0,
/* updateIndex */ false,
&persistentTid,
persistentSerialNum);
(*count)++;
}
PersistentStore_EndScan(&storeScan);
}
/*
* Indicate we intend to create a relation file as part of the current transaction.
*
* An XLOG IntentToCreate record is generated that will guard the subsequent file-system
* create in case the transaction aborts.
*
* After 1 or more calls to this routine to mark intention about relation files that are going
* to be created, call ~_DoPendingCreates to do the actual file-system creates. (See its
* note on XLOG flushing).
*/
void PersistentRelfile_AddCreatePending(
RelFileNode *relFileNode,
/* The tablespace, database, and relation OIDs for the create. */
int32 segmentFileNum,
PersistentFileSysRelStorageMgr relStorageMgr,
PersistentFileSysRelBufpoolKind relBufpoolKind,
bool bufferPoolBulkLoad,
char *relationName,
ItemPointer persistentTid,
/* Resulting TID of the gp_persistent_relation_files tuple for the relation. */
int64 *serialNum,
/* 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. */
bool isLocalBuf)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
XLogRecPtr mirrorBufpoolResyncCkptLoc;
ItemPointerData previousFreeTid;
Datum values[Natts_gp_persistent_relfile_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.
}
PersistentRelfile_VerifyInitScan();
PersistentFileSysObjName_SetRelationFile(
&fsObjName,
relFileNode,
segmentFileNum,
is_tablespace_shared);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
GpPersistentRelfileNode_SetDatumValues(
values,
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode,
segmentFileNum,
relStorageMgr,
(bufferPoolBulkLoad ?
PersistentFileSysState_BulkLoadCreatePending :
PersistentFileSysState_CreatePending),
relBufpoolKind,
GetTopTransactionId(),
/* persistentSerialNum */ 0, // This will be set by PersistentFileSysObj_AddTuple.
&previousFreeTid,
is_tablespace_shared(relFileNode->spcNode));
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,
persistentTid, serialNum);
#endif
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FaultBeforePendingDeleteRelationEntry,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
/*
* 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', serial number " INT64_FORMAT " at TID %s",
PersistentFileSysObjName_ObjectName(&fsObjName),
relationName,
PersistentFileSysRelStorageMgr_Name(relStorageMgr),
*serialNum,
ItemPointerToString(persistentTid));
}
