PersistentTidIsKnownResult PersistentStore_TidIsKnown(
PersistentStoreSharedData *storeSharedData,
ItemPointer persistentTid,
ItemPointer maxTid)
{
*maxTid = storeSharedData->maxTid;
Assert(!PersistentStore_IsZeroTid(persistentTid));
// UNDONE: I think the InRecovery test only applies to physical Master Mirroring on Standby.
/* Only test this outside of recovery scenarios */
if (Persistent_BeforePersistenceWork())
return PersistentTidIsKnownResult_BeforePersistenceWork;
if (storeSharedData->needToScanIntoSharedMemory)
return PersistentTidIsKnownResult_ScanNotPerformedYet;
if (PersistentStore_IsZeroTid(&storeSharedData->maxTid))
return PersistentTidIsKnownResult_MaxTidIsZero;
if (ItemPointerCompare(
persistentTid,
&storeSharedData->maxTid) <= 0) // Less-than or equal.
return PersistentTidIsKnownResult_Known;
else
return PersistentTidIsKnownResult_NotKnown;
}
void
PersistentTablespace_ActivateStandby(int16 oldmaster, int16 newmaster)
{
TablespaceDirEntry tablespaceDirEntry;
HASH_SEQ_STATUS hstat;
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
hash_seq_init(&hstat, persistentTablespaceSharedHashTable);
if (Persistent_BeforePersistenceWork())
elog(ERROR, "persistent table changes forbidden");
PersistentTablespace_VerifyInitScan();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
LWLockAcquire(TablespaceHashLock, LW_SHARED);
while ((tablespaceDirEntry = hash_seq_search(&hstat)) != NULL)
{
PersistentFileSysObjName fsObjName;
Oid tblspc = tablespaceDirEntry->key.tablespaceOid;
ItemPointerData persistentTid;
uint64 persistentSerialNum;
tablespaceDirEntry = PersistentTablespace_FindEntryUnderLock(tblspc);
if (tablespaceDirEntry == NULL)
elog(ERROR, "cannot find persistent tablespace entry %u",
tblspc);
persistentSerialNum = tablespaceDirEntry->persistentSerialNum;
ItemPointerCopy(&tablespaceDirEntry->persistentTid, &persistentTid);
/*
* We release TablespaceHashLock in the middle of the loop and
* re-acquire it after doing persistent table change. This is needed
* to prevent holding the lock for any purpose other than to protect
* the tablespace shared hash table. Not releasing this lock could
* result in file I/O and potential deadlock due to other LW locks
* being acquired in the process. Releasing the lock this way is safe
* because we are still holding PersistentObjLock in exclusive mode.
* Any change to the filespace shared hash table is also protected by
* PersistentObjLock.
*/
LWLockRelease(TablespaceHashLock);
PersistentFileSysObjName_SetTablespaceDir(&fsObjName, tblspc);
PersistentFileSysObj_ActivateStandby(&fsObjName,
&persistentTid,
persistentSerialNum,
oldmaster,
newmaster,
/* flushToXlog */ false);
LWLockAcquire(TablespaceHashLock, LW_SHARED);
}
LWLockRelease(TablespaceHashLock);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
/*
* Add a mirror.
*/
void
PersistentFilespace_AddMirror(Oid filespace,
char *mirpath,
int16 pridbid, int16 mirdbid,
bool set_mirror_existence)
{
PersistentFileSysObjName fsObjName;
char *newpath;
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
FilespaceDirEntry fde;
if (Persistent_BeforePersistenceWork())
elog(ERROR, "persistent table changes forbidden");
PersistentFilespace_VerifyInitScan();
PersistentFileSysObjName_SetFilespaceDir(&fsObjName, filespace);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
fde = PersistentFilespace_FindDirUnderLock(filespace);
if (fde == NULL)
elog(ERROR, "did not find persistent filespace entry %u",
filespace);
if (fde->dbId1 == pridbid)
{
fde->dbId2 = mirdbid;
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded2,
mirpath);
newpath = fde->locationBlankPadded2;
}
else if (fde->dbId2 == pridbid)
{
fde->dbId1 = mirdbid;
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded1,
mirpath);
newpath = fde->locationBlankPadded1;
}
else
{
Insist(false);
}
PersistentFileSysObj_AddMirror(&fsObjName,
&fde->persistentTid,
fde->persistentSerialNum,
pridbid,
mirdbid,
(void *)newpath,
set_mirror_existence,
/* flushToXlog */ false);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
/*
* 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);
}
bool
PersistentDatabase_DirIsCreated(DbDirNode *dbDirNode)
{
READ_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
DatabaseDirEntry databaseDirEntry;
bool result;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent database '%s' because we are before "
"persistence work",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace));
/*
* The initdb process will load the persistent table once we out of
* bootstrap mode.
*/
return true;
}
PersistentDatabase_VerifyInitScan();
READ_PERSISTENT_STATE_ORDERED_LOCK;
databaseDirEntry =
(DatabaseDirEntry)
SharedOidSearch_Find(
&persistentDatabaseSharedData->databaseDirSearchTable,
dbDirNode->database,
dbDirNode->tablespace);
result = (databaseDirEntry != NULL);
if (result &&
databaseDirEntry->state != PersistentFileSysState_Created &&
databaseDirEntry->state != PersistentFileSysState_CreatePending &&
databaseDirEntry->state != PersistentFileSysState_JustInTimeCreatePending)
elog(ERROR, "Persistent database entry %s expected to be in 'Create Pending' or 'Created' (actual state '%s')",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
READ_PERSISTENT_STATE_ORDERED_UNLOCK;
return result;
}
void
PersistentTablespace_RemoveSegment(int16 dbid, bool ismirror)
{
TablespaceDirEntry tablespaceDirEntry;
HASH_SEQ_STATUS hstat;
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
hash_seq_init(&hstat, persistentTablespaceSharedHashTable);
if (Persistent_BeforePersistenceWork())
elog(ERROR, "persistent table changes forbidden");
PersistentTablespace_VerifyInitScan();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
LWLockAcquire(TablespaceHashLock, LW_SHARED);
while ((tablespaceDirEntry = hash_seq_search(&hstat)) != NULL)
{
PersistentFileSysObjName fsObjName;
Oid tblspc = tablespaceDirEntry->key.tablespaceOid;
ItemPointerData persistentTid;
uint64 persistentSerialNum;
tablespaceDirEntry = PersistentTablespace_FindEntryUnderLock(tblspc);
LWLockRelease(TablespaceHashLock);
if (tablespaceDirEntry == NULL)
elog(ERROR, "Did not find persistent tablespace entry %u",
tblspc);
persistentSerialNum = tablespaceDirEntry->persistentSerialNum;
ItemPointerCopy(&tablespaceDirEntry->persistentTid, &persistentTid);
PersistentFileSysObjName_SetTablespaceDir(&fsObjName, tblspc);
PersistentFileSysObj_RemoveSegment(&fsObjName,
&persistentTid,
persistentSerialNum,
dbid,
ismirror,
/* flushToXlog */ false);
LWLockAcquire(TablespaceHashLock, LW_SHARED);
}
LWLockRelease(TablespaceHashLock);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
/*
* Remove all reference to a segment from the gp_persistent_filespace_node table
* and share memory structure.
*/
void
PersistentFilespace_RemoveSegment(int16 dbid, bool ismirror)
{
HASH_SEQ_STATUS hstat;
FilespaceDirEntry fde;
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
if (Persistent_BeforePersistenceWork())
elog(ERROR, "persistent table changes forbidden");
hash_seq_init(&hstat, persistentFilespaceSharedHashTable);
PersistentFilespace_VerifyInitScan();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
while ((fde = hash_seq_search(&hstat)) != NULL)
{
Oid filespace = fde->key.filespaceOid;
PersistentFileSysObjName fsObjName;
PersistentFileSysObjName_SetFilespaceDir(&fsObjName, filespace);
if (fde->dbId1 == dbid)
{
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded1,
"");
}
else if (fde->dbId2 == dbid)
{
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded2,
"");
}
PersistentFileSysObj_RemoveSegment(&fsObjName,
&fde->persistentTid,
fde->persistentSerialNum,
dbid,
ismirror,
/* flushToXlog */ false);
}
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
void PersistentDatabase_MarkJustInTimeCreatePending(
DbDirNode *dbDirNode,
ItemPointer persistentTid,
int64 *persistentSerialNum)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
DatabaseDirEntry databaseDirEntry;
SharedOidSearchAddResult addResult;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent database '%s' because we are before persistence work",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace));
return; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentDatabase_VerifyInitScan();
PersistentFileSysObjName_SetDatabaseDir(&fsObjName,dbDirNode->tablespace,dbDirNode->database,is_tablespace_shared);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
databaseDirEntry =
(DatabaseDirEntry)
SharedOidSearch_Find(
&persistentDatabaseSharedData->databaseDirSearchTable,
dbDirNode->database,
dbDirNode->tablespace);
if (databaseDirEntry != NULL)
{
/*
* An existence check should have been done before calling this routine.
*/
elog(ERROR, "Persistent database entry '%s' already exists in state '%s'",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
}
addResult =
SharedOidSearch_Add(
&persistentDatabaseSharedData->databaseDirSearchTable,
dbDirNode->database,
dbDirNode->tablespace,
(SharedOidSearchObjHeader**)&databaseDirEntry);
if (addResult == SharedOidSearchAddResult_NoMemory)
{
/* If out of shared memory, no need to promote to PANIC. */
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("Out of shared-memory for persistent databases"),
errhint("You may need to increase the gp_max_databases and "
"gp_max_tablespaces value"),
errOmitLocation(true)));
}
else if (addResult == SharedOidSearchAddResult_Exists)
elog(PANIC, "Persistent database entry '%s' already exists in state '%s'",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
else
Assert(addResult == SharedOidSearchAddResult_Ok);
databaseDirEntry->state = PersistentFileSysState_JustInTimeCreatePending;
PersistentDatabase_AddTuple(
databaseDirEntry,
/* reserved */ 0,
/* parentXid */ InvalidTransactionId,
/* flushToXLog */ true);
*persistentTid = databaseDirEntry->persistentTid;
*persistentSerialNum = databaseDirEntry->persistentSerialNum;
/*
* This XLOG must be generated under the persistent write-lock.
*/
#ifdef MASTER_MIRROR_SYNC
mmxlog_log_create_database(dbDirNode->tablespace,
dbDirNode->database);
#endif
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
/*
* Indicate the non-transaction just-in-time database create was NOT successful.
*/
void PersistentDatabase_AbandonJustInTimeCreatePending(
DbDirNode *dbDirNode,
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
int64 persistentSerialNum)
/* Serial number for the relation. Distinquishes the uses of the tuple. */
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
DatabaseDirEntry databaseDirEntry;
PersistentFileSysObjStateChangeResult stateChangeResult;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent database '%s' because we are before persistence work",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace));
return; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentDatabase_VerifyInitScan();
PersistentFileSysObjName_SetDatabaseDir(&fsObjName,dbDirNode->tablespace,dbDirNode->database,is_tablespace_shared);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
PersistentDatabase_LookupExistingDbDir(
dbDirNode,
&databaseDirEntry);
if (databaseDirEntry->state != PersistentFileSysState_JustInTimeCreatePending)
elog(ERROR, "Persistent database entry %s expected to be in 'Just-In-Time Create Pending' state (actual state '%s')",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
stateChangeResult =
PersistentFileSysObj_StateChange(
&fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_Free,
/* retryPossible */ false,
/* flushToXlog */ false,
/* oldState */ NULL,
/* verifiedActionCallback */ NULL);
databaseDirEntry->state = PersistentFileSysState_Free;
if (databaseDirEntry->iteratorRefCount == 0)
SharedOidSearch_Delete(
&persistentDatabaseSharedData->databaseDirSearchTable,
&databaseDirEntry->header);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent database directory: Abandon '%s' in state 'Just-In-Time Create Pending', serial number " INT64_FORMAT " at TID %s (State-Change result '%s')",
PersistentFileSysObjName_ObjectName(&fsObjName),
persistentSerialNum,
ItemPointerToString(persistentTid),
PersistentFileSysObjStateChangeResult_Name(stateChangeResult));
}
/*
* Indicate we physically removed the relation file.
*/
void PersistentDatabase_Dropped(
PersistentFileSysObjName *fsObjName,
/* The tablespace and database OIDs for the dropped relation. */
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
int64 persistentSerialNum)
/* Serial number for the relation. Distinquishes the uses of the tuple. */
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
DbDirNode *dbDirNode = &fsObjName->variant.dbDirNode;
DatabaseDirEntry databaseDirEntry;
PersistentFileSysState oldState;
PersistentFileSysObjStateChangeResult stateChangeResult;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent database '%s' because we are before persistence work",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace));
return; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentDatabase_VerifyInitScan();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
PersistentDatabase_LookupExistingDbDir(
dbDirNode,
&databaseDirEntry);
if (databaseDirEntry->state != PersistentFileSysState_DropPending &&
databaseDirEntry->state != PersistentFileSysState_AbortingCreate)
elog(ERROR, "Persistent database entry %s expected to be in 'Drop Pending' or 'Aborting Create' (actual state '%s')",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
stateChangeResult =
PersistentFileSysObj_StateChange(
fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_Free,
/* retryPossible */ false,
/* flushToXlog */ false,
&oldState,
PersistentDatabase_DroppedVerifiedActionCallback);
databaseDirEntry->state = PersistentFileSysState_Free;
if (databaseDirEntry->iteratorRefCount == 0)
SharedOidSearch_Delete(
&persistentDatabaseSharedData->databaseDirSearchTable,
&databaseDirEntry->header);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent database directory: '%s' changed state from '%s' to (Free), serial number " INT64_FORMAT " at TID %s (State-Change result '%s')",
PersistentFileSysObjName_ObjectName(fsObjName),
PersistentFileSysObjState_Name(oldState),
persistentSerialNum,
ItemPointerToString(persistentTid),
PersistentFileSysObjStateChangeResult_Name(stateChangeResult));
}
/*
* Indicate we physically removed the filespace file.
*/
void PersistentFilespace_Dropped(
Oid filespaceOid,
/* The filespace OID for the dropped filespace. */
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
int64 persistentSerialNum)
/* Serial number for the filespace. Distinquishes the uses of the tuple. */
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
FilespaceDirEntry filespaceDirEntry;
PersistentFileSysState oldState;
PersistentFileSysObjStateChangeResult stateChangeResult;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent filespace %u because we are before persistence work",
filespaceOid);
return; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentFilespace_VerifyInitScan();
PersistentFileSysObjName_SetFilespaceDir(&fsObjName,filespaceOid);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
stateChangeResult =
PersistentFileSysObj_StateChange(
&fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_Free,
/* retryPossible */ false,
/* flushToXlog */ false,
&oldState,
PersistentFilespace_DroppedVerifiedActionCallback);
WRITE_FILESPACE_HASH_LOCK;
filespaceDirEntry = PersistentFilespace_FindDirUnderLock(filespaceOid);
if (filespaceDirEntry == NULL)
elog(ERROR, "Did not find persistent filespace entry %u",
filespaceOid);
if (filespaceDirEntry->state != PersistentFileSysState_DropPending &&
filespaceDirEntry->state != PersistentFileSysState_AbortingCreate)
elog(ERROR, "Persistent filespace entry %u expected to be in 'Drop Pending' or 'Aborting Create' (actual state '%s')",
filespaceOid,
PersistentFileSysObjState_Name(filespaceDirEntry->state));
filespaceDirEntry->state = PersistentFileSysState_Free;
PersistentFilespace_RemoveDirUnderLock(filespaceDirEntry);
WRITE_FILESPACE_HASH_UNLOCK;
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent filespace directory: '%s' changed state from '%s' to (Free), serial number " INT64_FORMAT " at TID %s (State-Change result '%s')",
PersistentFileSysObjName_ObjectName(&fsObjName),
PersistentFileSysObjState_Name(oldState),
persistentSerialNum,
ItemPointerToString(persistentTid),
PersistentFileSysObjStateChangeResult_Name(stateChangeResult));
}
/*
* 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));
}
PersistentFileSysObjStateChangeResult PersistentRelation_MarkAbortingCreate(
PersistentFileSysObjName *fsObjName,
ItemPointer persistentTid,
int64 persistentSerialNum,
bool retryPossible)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
RelFileNode *relFileNode = &fsObjName->variant.rel.relFileNode;
RelationDirEntry relationDirEntry;
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));
/*
* The initdb process will load the persistent table once we out of bootstrap mode.
*/
return PersistentFileSysObjStateChangeResult_None;
}
}
PersistentRelation_VerifyInitScan();
/*
* 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;
relationDirEntry = PersistentRelation_FindEntryUnderLock(relFileNode);
if (relationDirEntry == NULL)
{
elog(ERROR, "Did not find persistent relation entry %u/%u/%u",
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode);
}
if (relationDirEntry->state != PersistentFileSysState_CreatePending)
{
elog(ERROR, "Persistent relation entry %u/%u/%u expected to be in 'Create Pending' (actual state '%s')",
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode,
PersistentFileSysObjState_Name(relationDirEntry->state));
}
stateChangeResult = PersistentFileSysObj_StateChange(
fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_AbortingCreate,
retryPossible,
/* flushToXLog */ false,
/* oldState */ NULL,
/* verifiedActionCallback */ NULL);
relationDirEntry->state = PersistentFileSysState_AbortingCreate;
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;
}
/*
* Activate a standby master by removing reference to the dead master
* and changing our dbid to the old master's dbid
*/
void
PersistentFilespace_ActivateStandby(int16 oldmaster, int16 newmaster)
{
HASH_SEQ_STATUS hstat;
FilespaceDirEntry fde;
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
if (Persistent_BeforePersistenceWork())
elog(ERROR, "persistent table changes forbidden");
hash_seq_init(&hstat, persistentFilespaceSharedHashTable);
PersistentFilespace_VerifyInitScan();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
while ((fde = hash_seq_search(&hstat)) != NULL)
{
Oid filespace = fde->key.filespaceOid;
PersistentFileSysObjName fsObjName;
PersistentFileSysObjName_SetFilespaceDir(&fsObjName, filespace);
if (fde->dbId1 == oldmaster)
{
fde->dbId1 = InvalidDbid;
fde->dbId2 = newmaster;
/* Copy standby filespace location into new master location */
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded2,
fde->locationBlankPadded1);
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded1,
"");
}
else if (fde->dbId2 == oldmaster)
{
fde->dbId2 = InvalidDbid;
fde->dbId1 = newmaster;
/* Copy standby filespace location into new master location */
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded1,
fde->locationBlankPadded2);
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded2,
"");
}
PersistentFileSysObj_ActivateStandby(&fsObjName,
&fde->persistentTid,
fde->persistentSerialNum,
oldmaster,
newmaster,
/* flushToXlog */ false);
}
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
/*
* Indicate we are aborting the create of a relation file.
*
* This state will make sure the relation gets dropped after a system crash.
*/
PersistentFileSysObjStateChangeResult PersistentDatabase_MarkAbortingCreate(
PersistentFileSysObjName *fsObjName,
/* The tablespace and database OIDs for the aborting create. */
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
int64 persistentSerialNum,
/* Serial number for the relation. Distinquishes the uses of the tuple. */
bool retryPossible)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
DbDirNode *dbDirNode = &fsObjName->variant.dbDirNode;
DatabaseDirEntry databaseDirEntry;
PersistentFileSysObjStateChangeResult stateChangeResult;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent database '%s' because we are before persistence work",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace));
return false; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentDatabase_VerifyInitScan();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
PersistentDatabase_LookupExistingDbDir(
dbDirNode,
&databaseDirEntry);
if (databaseDirEntry->state != PersistentFileSysState_CreatePending)
elog(ERROR, "Persistent database entry %s expected to be in 'Create Pending' (actual state '%s')",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
stateChangeResult =
PersistentFileSysObj_StateChange(
fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_AbortingCreate,
retryPossible,
/* flushToXlog */ false,
/* oldState */ NULL,
/* verifiedActionCallback */ NULL);
databaseDirEntry->state = PersistentFileSysState_AbortingCreate;
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent database directory: '%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 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;
}
void
PersistentTablespace_AddCreated(
Oid filespaceOid,
/* The filespace where the tablespace lives. */
Oid tablespaceOid,
/* The tablespace OID to be added. */
MirroredObjectExistenceState mirrorExistenceState,
bool flushToXLog)
/* When true, the XLOG record for this change will be flushed to disk. */
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
ItemPointerData persistentTid;
int64 persistentSerialNum;
TablespaceDirEntry tablespaceDirEntry;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent tablespace %u because we are before persistence work",
tablespaceOid);
return;
/*
* The initdb process will load the persistent table once we out of
* bootstrap mode.
*/
}
PersistentTablespace_VerifyInitScan();
PersistentFileSysObjName_SetTablespaceDir(&fsObjName, tablespaceOid);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
PersistentTablespace_AddTuple(
filespaceOid,
tablespaceOid,
PersistentFileSysState_Created,
/* createMirrorDataLossTrackingSessionNum */ 0,
mirrorExistenceState,
/* reserved */ 0,
InvalidTransactionId,
flushToXLog,
&persistentTid,
&persistentSerialNum);
WRITE_TABLESPACE_HASH_LOCK;
tablespaceDirEntry =
PersistentTablespace_CreateEntryUnderLock(filespaceOid, tablespaceOid);
Assert(tablespaceDirEntry != NULL);
tablespaceDirEntry->state = PersistentFileSysState_Created;
ItemPointerCopy(&persistentTid, &tablespaceDirEntry->persistentTid);
tablespaceDirEntry->persistentSerialNum = persistentSerialNum;
WRITE_TABLESPACE_HASH_UNLOCK;
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent tablespace directory: Add '%s' in state 'Created', mirror existence state '%s', serial number " INT64_FORMAT " at TID '%s' ",
PersistentFileSysObjName_ObjectName(&fsObjName),
MirroredObjectExistenceState_Name(mirrorExistenceState),
persistentSerialNum,
ItemPointerToString(&persistentTid));
}
/*
* 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;
}
/*
* Indicate we intend to create a filespace 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 filespace files that are going
* to be created, call ~_DoPendingCreates to do the actual file-system creates. (See its
* note on XLOG flushing).
*/
void PersistentFilespace_MarkCreatePending(
Oid filespaceOid,
/* The filespace where the filespace lives. */
int16 primaryDbId,
char *primaryFilespaceLocation,
/*
* The primary filespace directory path. NOT Blank padded.
* Just a NULL terminated string.
*/
int16 mirrorDbId,
char *mirrorFilespaceLocation,
MirroredObjectExistenceState mirrorExistenceState,
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
int64 *persistentSerialNum,
bool flushToXLog)
/* When true, the XLOG record for this change will be flushed to disk. */
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
FilespaceDirEntry filespaceDirEntry;
TransactionId topXid;
Datum values[Natts_gp_persistent_filespace_node];
char mirrorFilespaceLocationBlankPadded[FilespaceLocationBlankPaddedWithNullTermLen];
char primaryFilespaceLocationBlankPadded[FilespaceLocationBlankPaddedWithNullTermLen];
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent filespace %u because we are before persistence work",
filespaceOid);
return; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentFilespace_VerifyInitScan();
PersistentFileSysObjName_SetFilespaceDir(&fsObjName,filespaceOid);
topXid = GetTopTransactionId();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
PersistentFilespace_BlankPadCopyLocation(
primaryFilespaceLocationBlankPadded,
primaryFilespaceLocation);
PersistentFilespace_BlankPadCopyLocation(
mirrorFilespaceLocationBlankPadded,
mirrorFilespaceLocation);
GpPersistentFilespaceNode_SetDatumValues(
values,
filespaceOid,
primaryDbId,
primaryFilespaceLocationBlankPadded,
mirrorDbId,
mirrorFilespaceLocationBlankPadded,
PersistentFileSysState_CreatePending,
/* createMirrorDataLossTrackingSessionNum */ 0,
mirrorExistenceState,
/* reserved */ 0,
/* parentXid */ topXid,
/* persistentSerialNum */ 0); // This will be set by PersistentFileSysObj_AddTuple.
PersistentFileSysObj_AddTuple(
PersistentFsObjType_FilespaceDir,
values,
flushToXLog,
persistentTid,
persistentSerialNum);
WRITE_FILESPACE_HASH_LOCK;
filespaceDirEntry = PersistentFilespace_CreateDirUnderLock(filespaceOid);
Assert(filespaceDirEntry != NULL);
filespaceDirEntry->dbId1 = primaryDbId;
memcpy(filespaceDirEntry->locationBlankPadded1, primaryFilespaceLocationBlankPadded,
FilespaceLocationBlankPaddedWithNullTermLen);
filespaceDirEntry->dbId2 = mirrorDbId;
memcpy(filespaceDirEntry->locationBlankPadded2, mirrorFilespaceLocationBlankPadded,
FilespaceLocationBlankPaddedWithNullTermLen);
filespaceDirEntry->state = PersistentFileSysState_CreatePending;
ItemPointerCopy(persistentTid, &filespaceDirEntry->persistentTid);
filespaceDirEntry->persistentSerialNum = *persistentSerialNum;
WRITE_FILESPACE_HASH_UNLOCK;
/*
* This XLOG must be generated under the persistent write-lock.
*/
#ifdef MASTER_MIRROR_SYNC
mmxlog_log_create_filespace(filespaceOid);
#endif
SIMPLE_FAULT_INJECTOR(FaultBeforePendingDeleteFilespaceEntry);
/*
* MPP-18228
* To make adding 'Create Pending' entry to persistent table and adding
* to the PendingDelete list atomic
*/
PendingDelete_AddCreatePendingEntryWrapper(
&fsObjName,
persistentTid,
*persistentSerialNum);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent filespace directory: Add '%s' in state 'Created', mirror existence state '%s', serial number " INT64_FORMAT " at TID %s",
PersistentFileSysObjName_ObjectName(&fsObjName),
MirroredObjectExistenceState_Name(mirrorExistenceState),
*persistentSerialNum,
ItemPointerToString(persistentTid));
}
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.
*
* 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 PersistentDatabase_MarkCreatePending(
DbDirNode *dbDirNode,
ItemPointer persistentTid,
int64 *persistentSerialNum,
bool flushToXLog)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
DatabaseDirEntry databaseDirEntry;
SharedOidSearchAddResult addResult;
PersistentFileSysObjName fsObjName;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent database '%s' because we are before persistence work",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace));
/*
* The initdb process will load the persistent table once we
* out of bootstrap mode.
*/
return;
}
PersistentDatabase_VerifyInitScan();
PersistentFileSysObjName_SetDatabaseDir(
&fsObjName,
dbDirNode->tablespace,
dbDirNode->database,
is_tablespace_shared);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
databaseDirEntry =
(DatabaseDirEntry)
SharedOidSearch_Find(
&persistentDatabaseSharedData->databaseDirSearchTable,
dbDirNode->database,
dbDirNode->tablespace);
if (databaseDirEntry != NULL)
elog(ERROR, "Persistent database entry '%s' already exists in state '%s'",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
addResult =
SharedOidSearch_Add(
&persistentDatabaseSharedData->databaseDirSearchTable,
dbDirNode->database,
dbDirNode->tablespace,
(SharedOidSearchObjHeader**)&databaseDirEntry);
if (addResult == SharedOidSearchAddResult_NoMemory)
{
/* If out of shared memory, no need to promote to PANIC. */
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("Out of shared-memory for persistent databases"),
errhint("You may need to increase the gp_max_databases and "
"gp_max_tablespaces value"),
errOmitLocation(true)));
}
else if (addResult == SharedOidSearchAddResult_Exists)
elog(PANIC, "Persistent database entry '%s' already exists in state '%s'",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
else
Assert(addResult == SharedOidSearchAddResult_Ok);
databaseDirEntry->state = PersistentFileSysState_CreatePending;
databaseDirEntry->iteratorRefCount = 0;
PersistentDatabase_AddTuple(
databaseDirEntry,
/* reserved */ 0,
/* parentXid */ GetTopTransactionId(),
flushToXLog);
*persistentTid = databaseDirEntry->persistentTid;
*persistentSerialNum = databaseDirEntry->persistentSerialNum;
/*
* This XLOG must be generated under the persistent write-lock.
*/
#ifdef MASTER_MIRROR_SYNC
mmxlog_log_create_database(dbDirNode->tablespace, dbDirNode->database);
#endif
#ifdef FAULT_INJECTOR
FaultInjector_InjectFaultIfSet(
FaultBeforePendingDeleteDatabaseEntry,
DDLNotSpecified,
"", // databaseName
""); // tableName
#endif
/*
* MPP-18228
* To make adding 'Create Pending' entry to persistent table and adding
* to the PendingDelete list atomic
*/
PendingDelete_AddCreatePendingEntryWrapper(
&fsObjName,
persistentTid,
*persistentSerialNum);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
/*
* Activate a standby master by removing reference to the dead master
* and changing our dbid to the old master's dbid
*/
void
PersistentFilespace_ActivateStandby(int16 oldmaster, int16 newmaster)
{
HASH_SEQ_STATUS hstat;
FilespaceDirEntry fde;
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
if (Persistent_BeforePersistenceWork())
elog(ERROR, "persistent table changes forbidden");
hash_seq_init(&hstat, persistentFilespaceSharedHashTable);
PersistentFilespace_VerifyInitScan();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
/*
* We release FilespaceHashLock in the middle of the loop and re-acquire
* it after doing persistent table change. This is needed to prevent
* holding the lock for any purpose other than to protect the filespace
* shared hash table. Not releasing this lock could result in file I/O
* and potential deadlock due to other LW locks being acquired in the
* process. Releasing the lock this way is safe because we are still
* holding PersistentObjLock in exclusive mode. Any change to the
* filespace shared hash table is also protected by PersistentObjLock.
*/
WRITE_FILESPACE_HASH_LOCK;
while ((fde = hash_seq_search(&hstat)) != NULL)
{
Oid filespace = fde->key.filespaceOid;
PersistentFileSysObjName fsObjName;
ItemPointerData persistentTid;
int64 persistentSerialNum = fde->persistentSerialNum;
ItemPointerCopy(&fde->persistentTid, &persistentTid);
PersistentFileSysObjName_SetFilespaceDir(&fsObjName, filespace);
if (fde->dbId1 == oldmaster)
{
fde->dbId1 = InvalidDbid;
fde->dbId2 = newmaster;
/* Copy standby filespace location into new master location */
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded2,
fde->locationBlankPadded1);
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded1,
"");
}
else if (fde->dbId2 == oldmaster)
{
fde->dbId2 = InvalidDbid;
fde->dbId1 = newmaster;
/* Copy standby filespace location into new master location */
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded1,
fde->locationBlankPadded2);
PersistentFilespace_BlankPadCopyLocation(
fde->locationBlankPadded2,
"");
}
WRITE_FILESPACE_HASH_UNLOCK;
PersistentFileSysObj_ActivateStandby(&fsObjName,
&persistentTid,
persistentSerialNum,
oldmaster,
newmaster,
/* flushToXlog */ false);
WRITE_FILESPACE_HASH_LOCK;
}
WRITE_FILESPACE_HASH_UNLOCK;
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
void PersistentDatabase_AddCreated(
DbDirNode *dbDirNode,
/* The tablespace and database OIDs for the create. */
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
bool flushToXLog)
/* When true, the XLOG record for this change will be flushed to disk. */
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
DatabaseDirEntry databaseDirEntry;
SharedOidSearchAddResult addResult;
int64 persistentSerialNum;
if (!Persistent_BeforePersistenceWork())
elog(ERROR, "We can only add to persistent meta-data when special states");
// Verify PersistentFileSysObj_BuildInitScan has been called.
PersistentDatabase_VerifyInitScan();
PersistentFileSysObjName_SetDatabaseDir(&fsObjName,dbDirNode->tablespace,dbDirNode->database,is_tablespace_shared);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
/*
* GpIdentity.segindex
* In the initdb, GpIdentity.segindex is set to -10000. It will update this
* value to the correct GpIdentity.segindex.
*/
databaseDirEntry =
(DatabaseDirEntry)
SharedOidSearch_Find(
&persistentDatabaseSharedData->databaseDirSearchTable,
dbDirNode->database,
dbDirNode->tablespace);
if (databaseDirEntry != NULL)
elog(ERROR, "Persistent database entry '%s' already exists in state '%s'",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
addResult =
SharedOidSearch_Add(
&persistentDatabaseSharedData->databaseDirSearchTable,
dbDirNode->database,
dbDirNode->tablespace,
(SharedOidSearchObjHeader**)&databaseDirEntry);
if (addResult == SharedOidSearchAddResult_NoMemory)
{
/* If out of shared memory, no need to promote to PANIC. */
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("Out of shared-memory for persistent databases"),
errhint("You may need to increase the gp_max_databases and "
"gp_max_tablespaces value"),
errOmitLocation(true)));
}
else if (addResult == SharedOidSearchAddResult_Exists)
elog(PANIC, "Persistent database entry '%s' already exists in state '%s'",
GetDatabasePath(
dbDirNode->database,
dbDirNode->tablespace),
PersistentFileSysObjState_Name(databaseDirEntry->state));
else
Assert(addResult == SharedOidSearchAddResult_Ok);
databaseDirEntry->state = PersistentFileSysState_Created;
databaseDirEntry->iteratorRefCount = 0;
PersistentDatabase_AddTuple(
databaseDirEntry,
/* reserved */ 0,
InvalidTransactionId,
flushToXLog);
*persistentTid = databaseDirEntry->persistentTid;
persistentSerialNum = databaseDirEntry->persistentSerialNum;
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent database directory: Add '%s' in state 'Created', serial number " INT64_FORMAT " at TID %s",
PersistentFileSysObjName_ObjectName(&fsObjName),
persistentSerialNum,
ItemPointerToString(persistentTid));
}
void PersistentRelation_AddCreated(
RelFileNode *relFileNode,
ItemPointer persistentTid,
bool flushToXLog)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
ItemPointerData previousFreeTid;
int64 persistentSerialNum;
RelationDirEntry relationDirEntry;
Datum values[Natts_gp_persistent_relation_node];
if (RelFileNode_IsEmpty(relFileNode))
{
elog(ERROR, "Invalid RelFileNode (0,0,0)");
}
MemSet(&previousFreeTid, 0, sizeof(ItemPointerData));
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_SetRelationDir(
&fsObjName,
relFileNode,
is_tablespace_shared);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
relationDirEntry = PersistentRelation_CreateEntryUnderLock(relFileNode);
if (relationDirEntry == NULL)
{
/* If out of shared memory, no need to promote to PANIC. */
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("Out of shared-memory for persistent relations"),
errhint("You may need to increase the gp_max_relations value"),
errOmitLocation(true)));
}
relationDirEntry->state = PersistentFileSysState_Created;
GpPersistentRelationNode_SetDatumValues(
values,
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode,
PersistentFileSysState_Created,
/* reserved */ 0,
/* parentXid */ InvalidTransactionId,
/* persistentSerialNum */ 0,
&previousFreeTid,
is_tablespace_shared(relFileNode->spcNode));
PersistentFileSysObj_AddTuple(
PersistentFsObjType_RelationDir,
values,
flushToXLog,
persistentTid,
&persistentSerialNum);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
{
elog(Persistent_DebugPrintLevel(),
"Persistent relation: Add '%s', in state 'Created', serial number " INT64_FORMAT " at TID %s",
PersistentFileSysObjName_ObjectName(&fsObjName),
persistentSerialNum,
ItemPointerToString(persistentTid));
}
}
/*
* Indicate we intend to create a tablespace 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 tablespace files that are going
* to be created, call ~_DoPendingCreates to do the actual file-system creates. (See its
* note on XLOG flushing).
*/
void
PersistentTablespace_MarkCreatePending(
Oid filespaceOid,
/* The filespace where the tablespace lives. */
Oid tablespaceOid,
/* The tablespace OID for the create. */
MirroredObjectExistenceState mirrorExistenceState,
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
int64 *persistentSerialNum,
bool flushToXLog)
/* When true, the XLOG record for this change will be flushed to disk. */
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
TablespaceDirEntry tablespaceDirEntry;
TransactionId topXid;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent tablespace %u because we are before persistence work",
tablespaceOid);
return;
/*
* The initdb process will load the persistent table once we out of
* bootstrap mode.
*/
}
PersistentTablespace_VerifyInitScan();
PersistentFileSysObjName_SetTablespaceDir(&fsObjName, tablespaceOid);
topXid = GetTopTransactionId();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
PersistentTablespace_AddTuple(
filespaceOid,
tablespaceOid,
PersistentFileSysState_CreatePending,
/* createMirrorDataLossTrackingSessionNum */ 0,
mirrorExistenceState,
/* reserved */ 0,
/* parentXid */ topXid,
flushToXLog,
persistentTid,
persistentSerialNum);
WRITE_TABLESPACE_HASH_LOCK;
tablespaceDirEntry =
PersistentTablespace_CreateEntryUnderLock(filespaceOid, tablespaceOid);
Assert(tablespaceDirEntry != NULL);
tablespaceDirEntry->state = PersistentFileSysState_CreatePending;
ItemPointerCopy(persistentTid, &tablespaceDirEntry->persistentTid);
tablespaceDirEntry->persistentSerialNum = *persistentSerialNum;
WRITE_TABLESPACE_HASH_UNLOCK;
/*
* This XLOG must be generated under the persistent write-lock.
*/
#ifdef MASTER_MIRROR_SYNC
mmxlog_log_create_tablespace(
filespaceOid,
tablespaceOid);
#endif
SIMPLE_FAULT_INJECTOR(FaultBeforePendingDeleteTablespaceEntry);
/*
* MPP-18228 To make adding 'Create Pending' entry to persistent table and
* adding to the PendingDelete list atomic
*/
PendingDelete_AddCreatePendingEntryWrapper(
&fsObjName,
persistentTid,
*persistentSerialNum);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent tablespace directory: Add '%s' in state 'Created', mirror existence state '%s', serial number " INT64_FORMAT " at TID %s",
PersistentFileSysObjName_ObjectName(&fsObjName),
MirroredObjectExistenceState_Name(mirrorExistenceState),
*persistentSerialNum,
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));
}
/*
* Indicate the transaction commited and the tablespace is officially created.
*/
void
PersistentTablespace_Created(
Oid tablespaceOid,
/* The tablespace OID for the create. */
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
int64 persistentSerialNum,
/* Serial number for the tablespace. Distinquishes the uses of the tuple. */
bool retryPossible)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
TablespaceDirEntry tablespaceDirEntry;
PersistentFileSysObjStateChangeResult stateChangeResult;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent tablespace %u because we are before persistence work",
tablespaceOid);
return;
/*
* The initdb process will load the persistent table once we out of
* bootstrap mode.
*/
}
PersistentTablespace_VerifyInitScan();
PersistentFileSysObjName_SetTablespaceDir(&fsObjName, tablespaceOid);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
WRITE_TABLESPACE_HASH_LOCK;
tablespaceDirEntry = PersistentTablespace_FindEntryUnderLock(tablespaceOid);
if (tablespaceDirEntry == NULL)
elog(ERROR, "Did not find persistent tablespace entry %u",
tablespaceOid);
if (tablespaceDirEntry->state != PersistentFileSysState_CreatePending)
elog(ERROR, "Persistent tablespace entry %u expected to be in 'Create Pending' state (actual state '%s')",
tablespaceOid,
PersistentFileSysObjState_Name(tablespaceDirEntry->state));
tablespaceDirEntry->state = PersistentFileSysState_Created;
WRITE_TABLESPACE_HASH_UNLOCK;
stateChangeResult =
PersistentFileSysObj_StateChange(
&fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_Created,
retryPossible,
/* flushToXlog */ false,
/* oldState */ NULL,
/* verifiedActionCallback */ NULL);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent tablespace directory: '%s' changed state from 'Create Pending' to 'Created', serial number " INT64_FORMAT " at TID %s (State-Change result '%s')",
PersistentFileSysObjName_ObjectName(&fsObjName),
persistentSerialNum,
ItemPointerToString(persistentTid),
PersistentFileSysObjStateChangeResult_Name(stateChangeResult));
}
static void
PersistentEndXactRec_VerifyFileSysActionInfos(
EndXactRecKind endXactRecKind,
PersistentEndXactFileSysActionInfo *fileSysActionInfos,
int count)
{
int i;
ItemPointerData maxTid;
if (InRecovery || Persistent_BeforePersistenceWork())
return;
for (i = 0; i < count; i++)
{
PersistentTidIsKnownResult persistentTidIsKnownResult;
if (!PersistentEndXactFileSysAction_IsValid(fileSysActionInfos[i].action))
elog(ERROR, "Persistent file-system action is invalid (%d) (index %d, transaction kind '%s')",
fileSysActionInfos[i].action,
i,
EndXactRecKind_Name(endXactRecKind));
if (!PersistentFsObjType_IsValid(fileSysActionInfos[i].fsObjName.type))
elog(ERROR, "Persistent file-system object type is invalid (%d) (index %d, transaction kind '%s')",
fileSysActionInfos[i].fsObjName.type,
i,
EndXactRecKind_Name(endXactRecKind));
if (PersistentStore_IsZeroTid(&fileSysActionInfos[i].persistentTid))
elog(ERROR, "TID for persistent '%s' tuple is invalid (0,0) (index %d, transaction kind '%s')",
PersistentFileSysObjName_TypeAndObjectName(&fileSysActionInfos[i].fsObjName),
i,
EndXactRecKind_Name(endXactRecKind));
persistentTidIsKnownResult = PersistentFileSysObj_TidIsKnown(
fileSysActionInfos[i].fsObjName.type,
&fileSysActionInfos[i].persistentTid,
&maxTid);
switch (persistentTidIsKnownResult)
{
case PersistentTidIsKnownResult_BeforePersistenceWork:
elog(ERROR, "Shouldn't being trying to verify persistent TID before persistence work");
break;
case PersistentTidIsKnownResult_ScanNotPerformedYet:
// UNDONE: For now, just debug log this.
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Can't verify persistent TID if we haven't done the persistent scan yet");
break;
case PersistentTidIsKnownResult_MaxTidIsZero:
// UNDONE: For now, just debug log this.
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"TID for persistent '%s' tuple TID %s and the last known TID zero (0,0) (index %d, transaction kind '%s')",
PersistentFileSysObjName_TypeAndObjectName(&fileSysActionInfos[i].fsObjName),
ItemPointerToString(&fileSysActionInfos[i].persistentTid),
i,
EndXactRecKind_Name(endXactRecKind));
break;
case PersistentTidIsKnownResult_NotKnown:
// UNDONE: For now, just debug log this.
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"TID for persistent '%s' tuple TID %s is beyond the last known TID %s (index %d, transaction kind '%s')",
PersistentFileSysObjName_TypeAndObjectName(&fileSysActionInfos[i].fsObjName),
ItemPointerToString(&fileSysActionInfos[i].persistentTid),
ItemPointerToString2(&maxTid),
i,
EndXactRecKind_Name(endXactRecKind));
break;
case PersistentTidIsKnownResult_Known:
/* OK */
break;
default:
elog(ERROR, "Unexpected persistent file-system TID is known result: %d",
persistentTidIsKnownResult);
}
if (fileSysActionInfos[i].persistentSerialNum == 0)
elog(ERROR, "Persistent '%s' serial number is invalid (0) (index %d, transaction kind '%s')",
PersistentFileSysObjName_TypeAndObjectName(&fileSysActionInfos[i].fsObjName),
i,
EndXactRecKind_Name(endXactRecKind));
if (fileSysActionInfos[i].fsObjName.type == PersistentFsObjType_RelationFile &&
!PersistentFileSysRelStorageMgr_IsValid(fileSysActionInfos[i].relStorageMgr))
elog(ERROR, "Persistent '%s' relation storage manager has invalid value (%d) (index %d, transaction kind '%s')",
PersistentFileSysObjName_TypeAndObjectName(&fileSysActionInfos[i].fsObjName),
fileSysActionInfos[i].relStorageMgr,
i,
EndXactRecKind_Name(endXactRecKind));
}
}
void PersistentRelation_MarkCreatePending(
RelFileNode *relFileNode,
ItemPointer persistentTid,
int64 *persistentSerialNum,
bool flushToXLog)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
PersistentFileSysObjName fsObjName;
RelationDirEntry relationDirEntry;
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));
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
{
elog(Persistent_DebugPrintLevel(),
"Skipping persistent relation '%s' because we are before persistence work",
relpath(*relFileNode));
}
*persistentSerialNum = 0;
/*
* The initdb process will load the persistent table once we out
* of bootstrap mode.
*/
return;
}
PersistentRelation_VerifyInitScan();
PersistentFileSysObjName_SetRelationDir(
&fsObjName,
relFileNode,
is_tablespace_shared);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
relationDirEntry = PersistentRelation_CreateEntryUnderLock(relFileNode);
if (relationDirEntry == NULL)
{
/* If out of shared memory, no need to promote to PANIC. */
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("Out of shared-memory for persistent relations"),
errhint("You may need to increase the gp_max_relations value"),
errOmitLocation(true)));
}
relationDirEntry->state = PersistentFileSysState_CreatePending;
GpPersistentRelationNode_SetDatumValues(
values,
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode,
PersistentFileSysState_CreatePending,
/* reserved */ 0,
/* parentXid */ GetTopTransactionId(),
/* persistentSerialNum */ 0, // This will be set by PersistentFileSysObj_AddTuple.
&previousFreeTid,
is_tablespace_shared(relFileNode->spcNode));
PersistentFileSysObj_AddTuple(
PersistentFsObjType_RelationDir,
values,
flushToXLog,
&relationDirEntry->persistentTid,
&relationDirEntry->persistentSerialNum);
*persistentTid = relationDirEntry->persistentTid;
*persistentSerialNum = relationDirEntry->persistentSerialNum;
/*
* This XLOG must be generated under the persistent write-lock.
*/
#ifdef MASTER_MIRROR_SYNC
mmxlog_log_create_relation(
relFileNode->spcNode,
relFileNode->dbNode,
relFileNode->relNode,
persistentTid, persistentSerialNum);
#endif
/*
* MPP-18228
* To make adding 'Create Pending' entry to persistent table and
* adding to the PendingDelete list atomic
*/
PendingDelete_AddCreatePendingEntryWrapper(
&fsObjName,
persistentTid,
*persistentSerialNum);
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
}
/*
* Indicate we are aborting the create of a filespace file.
*
* This state will make sure the filespace gets dropped after a system crash.
*/
PersistentFileSysObjStateChangeResult PersistentFilespace_MarkAbortingCreate(
PersistentFileSysObjName *fsObjName,
/* The filespace OID for the aborting create. */
ItemPointer persistentTid,
/* TID of the gp_persistent_rel_files tuple for the rel file */
int64 persistentSerialNum,
/* Serial number for the filespace. Distinquishes the uses of the tuple. */
bool retryPossible)
{
WRITE_PERSISTENT_STATE_ORDERED_LOCK_DECLARE;
Oid filespaceOid = fsObjName->variant.filespaceOid;
FilespaceDirEntry filespaceDirEntry;
PersistentFileSysObjStateChangeResult stateChangeResult;
if (Persistent_BeforePersistenceWork())
{
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Skipping persistent filespace %u because we are before persistence work",
filespaceOid);
return false; // The initdb process will load the persistent table once we out of bootstrap mode.
}
PersistentFilespace_VerifyInitScan();
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
filespaceDirEntry =
PersistentFilespace_FindDirUnderLock(
filespaceOid);
if (filespaceDirEntry == NULL)
elog(ERROR, "Did not find persistent filespace entry %u", filespaceOid);
if (filespaceDirEntry->state != PersistentFileSysState_CreatePending)
elog(ERROR, "Persistent filespace entry %u expected to be in 'Create Pending' (actual state '%s')",
filespaceOid,
PersistentFileSysObjState_Name(filespaceDirEntry->state));
stateChangeResult =
PersistentFileSysObj_StateChange(
fsObjName,
persistentTid,
persistentSerialNum,
PersistentFileSysState_AbortingCreate,
retryPossible,
/* flushToXlog */ false,
/* oldState */ NULL,
/* verifiedActionCallback */ NULL);
filespaceDirEntry->state = PersistentFileSysState_AbortingCreate;
WRITE_PERSISTENT_STATE_ORDERED_UNLOCK;
if (Debug_persistent_print)
elog(Persistent_DebugPrintLevel(),
"Persistent filespace directory: '%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;
}
