Datum
currtid_byreloid(PG_FUNCTION_ARGS)
{
Oid reloid = PG_GETARG_OID(0);
ItemPointer tid = PG_GETARG_ITEMPOINTER(1);
ItemPointer result;
Relation rel;
result = (ItemPointer) palloc(sizeof(ItemPointerData));
if (!reloid)
{
*result = Current_last_tid;
PG_RETURN_ITEMPOINTER(result);
}
rel = heap_open(reloid, AccessShareLock);
if (rel->rd_rel->relkind == RELKIND_VIEW)
return currtid_for_view(rel, tid);
ItemPointerCopy(tid, result);
heap_get_latest_tid(rel, SnapshotNow, result);
heap_close(rel, AccessShareLock);
PG_RETURN_ITEMPOINTER(result);
}
Datum
currtid_byreloid(PG_FUNCTION_ARGS)
{
Oid reloid = PG_GETARG_OID(0);
ItemPointer tid = PG_GETARG_ITEMPOINTER(1);
ItemPointer result;
Relation rel;
AclResult aclresult;
result = (ItemPointer) palloc(sizeof(ItemPointerData));
if (!reloid)
{
*result = Current_last_tid;
PG_RETURN_ITEMPOINTER(result);
}
rel = heap_open(reloid, AccessShareLock);
aclresult = pg_class_aclcheck(RelationGetRelid(rel), GetUserId(),
ACL_SELECT);
if (aclresult != ACLCHECK_OK)
aclcheck_error(aclresult, ACL_KIND_CLASS,
RelationGetRelationName(rel));
if (rel->rd_rel->relkind == RELKIND_VIEW)
return currtid_for_view(rel, tid);
ItemPointerCopy(tid, result);
heap_get_latest_tid(rel, SnapshotNow, result);
heap_close(rel, AccessShareLock);
PG_RETURN_ITEMPOINTER(result);
}
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;
}
Datum
currtid_byrelname(PG_FUNCTION_ARGS)
{
text *relname = PG_GETARG_TEXT_P(0);
ItemPointer tid = PG_GETARG_ITEMPOINTER(1);
ItemPointer result;
RangeVar *relrv;
Relation rel;
AclResult aclresult;
Snapshot snapshot;
relrv = makeRangeVarFromNameList(textToQualifiedNameList(relname));
rel = heap_openrv(relrv, AccessShareLock);
aclresult = pg_class_aclcheck(RelationGetRelid(rel), GetUserId(),
ACL_SELECT);
if (aclresult != ACLCHECK_OK)
aclcheck_error(aclresult, ACL_KIND_CLASS,
RelationGetRelationName(rel));
if (rel->rd_rel->relkind == RELKIND_VIEW || rel->rd_rel->relkind == RELKIND_CONTVIEW)
return currtid_for_view(rel, tid);
result = (ItemPointer) palloc(sizeof(ItemPointerData));
ItemPointerCopy(tid, result);
snapshot = RegisterSnapshot(GetLatestSnapshot());
heap_get_latest_tid(rel, snapshot, result);
UnregisterSnapshot(snapshot);
heap_close(rel, AccessShareLock);
PG_RETURN_ITEMPOINTER(result);
}
/*
* Fetches the next entry from a visimap store index scan.
*
* Parameter visiMapEntry may be NULL. If it is not NULL and
* the scan returns an entry, the entry data is copied to the
* visimapEntry.
* Parameter tupleTid may be NULL. If it is not NULL and the scan
* returns an entry, the (heap) tuple id is copied to the parameter.
*/
bool
AppendOnlyVisimapStore_GetNext(
AppendOnlyVisimapStore *visiMapStore,
IndexScanDesc indexScan,
ScanDirection scanDirection,
AppendOnlyVisimapEntry* visiMapEntry,
ItemPointerData *tupleTid)
{
HeapTuple tuple;
TupleDesc heapTupleDesc;
Assert(visiMapStore);
Assert(RelationIsValid(visiMapStore->visimapRelation));
Assert(RelationIsValid(visiMapStore->visimapIndex));
Assert(indexScan);
tuple = AppendOnlyVisimapStore_GetNextTuple(visiMapStore, indexScan, scanDirection);
if (tuple == NULL)
{
return false;
}
heapTupleDesc = RelationGetDescr(visiMapStore->visimapRelation);
if (visiMapEntry)
{
AppendOnlyVisimapEntry_Copyout(visiMapEntry, tuple,
heapTupleDesc);
}
if (tupleTid)
{
ItemPointerCopy(&tuple->t_self, tupleTid);
}
return true;
}
/*
* add a task, it will be performed on all segments.
*/
void SharedStorageOpAddTask(const char * relname, RelFileNode *node,
int32 segno, ItemPointer persistentTid,
int64 persistentSerialNum,
SharedStorageOpTasks *tasks) {
Assert(NULL != node && NULL != tasks);
Assert(tasks->sizeTasks >= tasks->numTasks);
RelFileNode *n;
if (tasks->sizeTasks == tasks->numTasks) {
tasks->tasks = repalloc(tasks->tasks,
tasks->sizeTasks * sizeof(SharedStorageOpTask) * 2);
tasks->sizeTasks *= 2;
}
n = &tasks->tasks[tasks->numTasks].node;
n->dbNode = node->dbNode;
n->relNode = node->relNode;
n->spcNode = node->spcNode;
tasks->tasks[tasks->numTasks].segno = segno;
tasks->tasks[tasks->numTasks].relname = palloc(strlen(relname) + 1);
strcpy(tasks->tasks[tasks->numTasks].relname, relname);
ItemPointerCopy(persistentTid, &tasks->tasks[tasks->numTasks].persistentTid);
tasks->tasks[tasks->numTasks].persistentSerialNum = persistentSerialNum;
tasks->numTasks++;
}
static bool PersistentStore_GetFreeTuple(
PersistentStoreData *storeData,
PersistentStoreSharedData *storeSharedData,
ItemPointer freeTid)
{
ItemPointerData previousFreeTid;
MemSet(freeTid, 0, sizeof(ItemPointerData));
if (Debug_persistent_store_print)
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_GetFreeTuple: Enter: maximum free order number " INT64_FORMAT ", free TID %s ('%s')",
storeSharedData->maxFreeOrderNum,
ItemPointerToString(&storeSharedData->freeTid),
storeData->tableName);
if (storeSharedData->maxFreeOrderNum == 0)
{
return false; /* No free tuples. */
}
if (gp_persistent_skip_free_list)
{
if (Debug_persistent_store_print)
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_GetFreeTuple: Skipping because gp_persistent_skip_free_list GUC is ON ('%s')",
storeData->tableName);
return false; /* Pretend no free tuples. */
}
Assert(storeSharedData->freeTid.ip_posid != 0);
if (!PersistentStore_ValidateFreeTID(
storeData,
storeSharedData,
&previousFreeTid))
return false;
*freeTid = storeSharedData->freeTid;
storeSharedData->maxFreeOrderNum--;
ItemPointerCopy(&previousFreeTid /* previousFreeTid set inside the ValidateFreeTID function */,
&storeSharedData->freeTid);
if (Debug_persistent_store_print)
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_GetFreeTuple: Exit: maximum free order number " INT64_FORMAT ", free TID %s ('%s')",
storeSharedData->maxFreeOrderNum,
ItemPointerToString(&storeSharedData->freeTid),
storeData->tableName);
if (validate_previous_free_tid &&
!PersistentStore_ValidateFreeTID(
storeData,
storeSharedData,
&previousFreeTid))
return false;
return true;
}
/*
* update a tuple in in-memory heap table.
*
* if the target tuple already in the memory,
* update it in-place with flag INMEM_HEAP_TUPLE_UPDATED.
* else report an error.
*
* update should not change the otid of the old tuple,
* since updated tuple should write back to the master and update there.
*/
void
InMemHeap_Update(InMemHeapRelation relation, ItemPointer otid,
HeapTuple tup)
{
int pos;
HeapTuple target;
MemoryContext oldmem = CurrentMemoryContext;
Assert(ItemPointerIsValid(otid));
pos = InMemHeap_Find(relation, otid);
CurrentMemoryContext = relation->memcxt;
/*
* not found, report error
*/
if (pos >= relation->tupsize)
{
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("update a tuple which does not exist,"
" relname = %s, relid = %u", relation->rel->rd_rel->relname.data,
relation->relid)));
}
Insist(relation->hashIndex == NULL && "cannot handle index in in-memory heap when update");
/*
* already in table
*/
Assert(relation->tuples[pos].flags == INMEM_HEAP_TUPLE_DISPATCHED
|| relation->tuples[pos].flags == INMEM_HEAP_TUPLE_UPDATED);
relation->tuples[pos].flags = INMEM_HEAP_TUPLE_UPDATED;
target = heaptuple_copy_to(tup, NULL, NULL );
/*
* do not modify original tuple header
*/
ItemPointerCopy(&target->t_self, &relation->tuples[pos].tuple->t_self);
Assert(ItemPointerEquals(&target->t_self, otid));
memcpy(target->t_data, relation->tuples[pos].tuple->t_data,
sizeof(HeapTupleHeaderData));
CurrentMemoryContext = oldmem;
pfree(relation->tuples[pos].tuple);
relation->tuples[pos].tuple = target;
}
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;
}
Datum
currtid_byrelname(PG_FUNCTION_ARGS)
{
text *relname = PG_GETARG_TEXT_P(0);
ItemPointer tid = PG_GETARG_ITEMPOINTER(1);
ItemPointer result;
RangeVar *relrv;
Relation rel;
AclResult aclresult;
/*
* Immediately inform client that the function is not supported
*/
elog(ERROR, "Function currtid2 is not supported by GPDB");
relrv = makeRangeVarFromNameList(textToQualifiedNameList(relname));
rel = heap_openrv(relrv, AccessShareLock);
aclresult = pg_class_aclcheck(RelationGetRelid(rel), GetUserId(),
ACL_SELECT);
if (aclresult != ACLCHECK_OK)
aclcheck_error(aclresult, ACL_KIND_CLASS,
RelationGetRelationName(rel));
if (rel->rd_rel->relkind == RELKIND_VIEW)
return currtid_for_view(rel, tid);
result = (ItemPointer) palloc(sizeof(ItemPointerData));
ItemPointerCopy(tid, result);
heap_get_latest_tid(rel, SnapshotNow, result);
heap_close(rel, AccessShareLock);
PG_RETURN_ITEMPOINTER(result);
}
Datum
currtid_byrelname(PG_FUNCTION_ARGS)
{
text *relname = PG_GETARG_TEXT_P(0);
ItemPointer tid = PG_GETARG_ITEMPOINTER(1);
ItemPointer result;
RangeVar *relrv;
Relation rel;
relrv = makeRangeVarFromNameList(textToQualifiedNameList(relname,
"currtid_byrelname"));
rel = heap_openrv(relrv, AccessShareLock);
if (rel->rd_rel->relkind == RELKIND_VIEW)
return currtid_for_view(rel, tid);
result = (ItemPointer) palloc(sizeof(ItemPointerData));
ItemPointerCopy(tid, result);
heap_get_latest_tid(rel, SnapshotNow, result);
heap_close(rel, AccessShareLock);
PG_RETURN_ITEMPOINTER(result);
}
/*
* Search the relation 'rel' for tuple using the index.
*
* If a matching tuple is found, lock it with lockmode, fill the slot with its
* contents, and return true. Return false otherwise.
*/
bool
RelationFindReplTupleByIndex(Relation rel, Oid idxoid,
LockTupleMode lockmode,
TupleTableSlot *searchslot,
TupleTableSlot *outslot)
{
HeapTuple scantuple;
ScanKeyData skey[INDEX_MAX_KEYS];
IndexScanDesc scan;
SnapshotData snap;
TransactionId xwait;
Relation idxrel;
bool found;
/* Open the index. */
idxrel = index_open(idxoid, RowExclusiveLock);
/* Start an index scan. */
InitDirtySnapshot(snap);
scan = index_beginscan(rel, idxrel, &snap,
RelationGetNumberOfAttributes(idxrel),
0);
/* Build scan key. */
build_replindex_scan_key(skey, rel, idxrel, searchslot);
retry:
found = false;
index_rescan(scan, skey, RelationGetNumberOfAttributes(idxrel), NULL, 0);
/* Try to find the tuple */
if ((scantuple = index_getnext(scan, ForwardScanDirection)) != NULL)
{
found = true;
ExecStoreTuple(scantuple, outslot, InvalidBuffer, false);
ExecMaterializeSlot(outslot);
xwait = TransactionIdIsValid(snap.xmin) ?
snap.xmin : snap.xmax;
/*
* If the tuple is locked, wait for locking transaction to finish and
* retry.
*/
if (TransactionIdIsValid(xwait))
{
XactLockTableWait(xwait, NULL, NULL, XLTW_None);
goto retry;
}
}
/* Found tuple, try to lock it in the lockmode. */
if (found)
{
Buffer buf;
HeapUpdateFailureData hufd;
HTSU_Result res;
HeapTupleData locktup;
ItemPointerCopy(&outslot->tts_tuple->t_self, &locktup.t_self);
PushActiveSnapshot(GetLatestSnapshot());
res = heap_lock_tuple(rel, &locktup, GetCurrentCommandId(false),
lockmode,
LockWaitBlock,
false /* don't follow updates */ ,
&buf, &hufd);
/* the tuple slot already has the buffer pinned */
ReleaseBuffer(buf);
PopActiveSnapshot();
switch (res)
{
case HeapTupleMayBeUpdated:
break;
case HeapTupleUpdated:
/* XXX: Improve handling here */
ereport(LOG,
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
errmsg("concurrent update, retrying")));
goto retry;
case HeapTupleInvisible:
elog(ERROR, "attempted to lock invisible tuple");
default:
elog(ERROR, "unexpected heap_lock_tuple status: %u", res);
break;
}
}
index_endscan(scan);
/* Don't release lock until commit. */
index_close(idxrel, NoLock);
return found;
}
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 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));
}
/*
* _bt_mergeload - Merge two streams of index tuples into new index files.
*/
static void
_bt_mergeload(Spooler *self, BTWriteState *wstate, BTSpool *btspool, BTReader *btspool2, Relation heapRel)
{
BTPageState *state = NULL;
IndexTuple itup,
itup2;
bool should_free = false;
TupleDesc tupdes = RelationGetDescr(wstate->index);
int keysz = RelationGetNumberOfAttributes(wstate->index);
ScanKey indexScanKey;
ON_DUPLICATE on_duplicate = self->on_duplicate;
Assert(btspool != NULL);
/* the preparation of merge */
itup = BTSpoolGetNextItem(btspool, NULL, &should_free);
itup2 = BTReaderGetNextItem(btspool2);
indexScanKey = _bt_mkscankey_nodata(wstate->index);
for (;;)
{
bool load1 = true; /* load BTSpool next ? */
bool hasnull;
int32 compare;
if (self->dup_old + self->dup_new > self->max_dup_errors)
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("Maximum duplicate error count exceeded")));
if (itup2 == NULL)
{
if (itup == NULL)
break;
}
else if (itup != NULL)
{
compare = compare_indextuple(itup, itup2, indexScanKey,
keysz, tupdes, &hasnull);
if (compare == 0 && !hasnull && btspool->isunique)
{
ItemPointerData t_tid2;
/*
* t_tid is update by heap_is_visible(), because use it for an
* index, t_tid backup
*/
ItemPointerCopy(&itup2->t_tid, &t_tid2);
/* The tuple pointed by the old index should not be visible. */
if (!heap_is_visible(heapRel, &itup->t_tid))
{
itup = BTSpoolGetNextItem(btspool, itup, &should_free);
}
else if (!heap_is_visible(heapRel, &itup2->t_tid))
{
itup2 = BTReaderGetNextItem(btspool2);
}
else
{
if (on_duplicate == ON_DUPLICATE_KEEP_NEW)
{
self->dup_old++;
remove_duplicate(self, heapRel, itup2,
RelationGetRelationName(wstate->index));
itup2 = BTReaderGetNextItem(btspool2);
}
else
{
ItemPointerCopy(&t_tid2, &itup2->t_tid);
self->dup_new++;
remove_duplicate(self, heapRel, itup,
RelationGetRelationName(wstate->index));
itup = BTSpoolGetNextItem(btspool, itup, &should_free);
}
}
continue;
}
else if (compare > 0)
load1 = false;
}
else
load1 = false;
BULKLOAD_PROFILE(&prof_merge_unique);
/* When we see first tuple, create first index page */
if (state == NULL)
state = _bt_pagestate(wstate, 0);
if (load1)
{
IndexTuple next_itup = NULL;
bool next_should_free = false;
for (;;)
{
/* get next item */
next_itup = BTSpoolGetNextItem(btspool, next_itup,
&next_should_free);
if (!btspool->isunique || next_itup == NULL)
break;
compare = compare_indextuple(itup, next_itup, indexScanKey,
keysz, tupdes, &hasnull);
if (compare < 0 || hasnull)
break;
if (compare > 0)
{
/* shouldn't happen */
elog(ERROR, "faild in tuplesort_performsort");
}
/*
* If tupple is deleted by other unique indexes, not visible
*/
if (!heap_is_visible(heapRel, &next_itup->t_tid))
{
continue;
}
if (!heap_is_visible(heapRel, &itup->t_tid))
{
if (should_free)
pfree(itup);
itup = next_itup;
should_free = next_should_free;
next_should_free = false;
continue;
}
/* not unique between input files */
self->dup_new++;
remove_duplicate(self, heapRel, next_itup,
RelationGetRelationName(wstate->index));
if (self->dup_old + self->dup_new > self->max_dup_errors)
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("Maximum duplicate error count exceeded")));
}
_bt_buildadd(wstate, state, itup);
if (should_free)
pfree(itup);
itup = next_itup;
should_free = next_should_free;
}
else
{
_bt_buildadd(wstate, state, itup2);
itup2 = BTReaderGetNextItem(btspool2);
}
BULKLOAD_PROFILE(&prof_merge_insert);
}
_bt_freeskey(indexScanKey);
/* Close down final pages and write the metapage */
_bt_uppershutdown(wstate, state);
/*
* If the index isn't temp, we must fsync it down to disk before it's safe
* to commit the transaction. (For a temp index we don't care since the
* index will be uninteresting after a crash anyway.)
*
* It's obvious that we must do this when not WAL-logging the build. It's
* less obvious that we have to do it even if we did WAL-log the index
* pages. The reason is that since we're building outside shared buffers,
* a CHECKPOINT occurring during the build has no way to flush the
* previously written data to disk (indeed it won't know the index even
* exists). A crash later on would replay WAL from the checkpoint,
* therefore it wouldn't replay our earlier WAL entries. If we do not
* fsync those pages here, they might still not be on disk when the crash
* occurs.
*/
if (!RELATION_IS_LOCAL(wstate->index))
{
RelationOpenSmgr(wstate->index);
smgrimmedsync(wstate->index->rd_smgr, MAIN_FORKNUM);
}
BULKLOAD_PROFILE(&prof_merge_term);
}
void PersistentStore_FreeTuple(
PersistentStoreData *storeData,
PersistentStoreSharedData *storeSharedData,
ItemPointer persistentTid,
/* TID of the stored tuple. */
Datum *freeValues,
bool flushToXLog)
/* When true, the XLOG record for this change will be flushed to disk. */
{
Relation persistentRel;
HeapTuple persistentTuple = NULL;
ItemPointerData prevFreeTid;
XLogRecPtr xlogEndLoc;
/* The end location of the UPDATE XLOG record. */
Assert( LWLockHeldByMe(PersistentObjLock) );
#ifdef USE_ASSERT_CHECKING
if (storeSharedData == NULL ||
!PersistentStoreSharedData_EyecatcherIsValid(storeSharedData))
elog(ERROR, "Persistent store shared-memory not valid");
#endif
if (Debug_persistent_store_print)
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_FreeTuple: Going to free tuple at TID %s ('%s', shared data %p)",
ItemPointerToString(persistentTid),
storeData->tableName,
storeSharedData);
Assert(persistentTid->ip_posid != 0);
persistentRel = (*storeData->openRel)();
prevFreeTid = storeSharedData->freeTid;
if (validate_previous_free_tid)
{
/* Let us validate and have sanity check to make sure the prevFreeTid is really free. */
ItemPointerData tmpPrevFreeTid;
PersistentStore_ValidateFreeTID(
storeData,
storeSharedData,
&tmpPrevFreeTid);
}
storeSharedData->maxFreeOrderNum++;
if (storeSharedData->maxFreeOrderNum == 1)
ItemPointerCopy(persistentTid, &prevFreeTid); /* So non-zero PreviousFreeTid indicates free. */
storeSharedData->freeTid = *persistentTid;
PersistentStore_FormTupleSetOurs(
storeData,
persistentRel->rd_att,
freeValues,
storeSharedData->maxFreeOrderNum,
&prevFreeTid,
&persistentTuple);
persistentTuple->t_self = *persistentTid;
frozen_heap_inplace_update(persistentRel, persistentTuple);
/*
* XLOG location of the UPDATE tuple's XLOG record.
*/
xlogEndLoc = XLogLastInsertEndLoc();
heap_freetuple(persistentTuple);
(*storeData->closeRel)(persistentRel);
storeSharedData->inUseCount--;
if (Debug_persistent_store_print)
elog(PersistentStore_DebugPrintLevel(),
"PersistentStore_FreeTuple: Freed tuple at TID %s. Maximum free order number " INT64_FORMAT ", in use count " INT64_FORMAT " ('%s')",
ItemPointerToString(&storeSharedData->freeTid),
storeSharedData->maxFreeOrderNum,
storeSharedData->inUseCount,
storeData->tableName);
if (flushToXLog)
{
XLogFlush(xlogEndLoc);
XLogRecPtr_Zero(&nowaitXLogEndLoc);
}
else
nowaitXLogEndLoc = xlogEndLoc;
}
/*
* MultiRelease -- release a multi-level lock
*
* Returns: TRUE if successful, FALSE otherwise.
*/
bool
MultiRelease(LockTableId tableId,
LOCKTAG *tag,
LOCKT lockt,
LOCK_LEVEL level)
{
LOCKT locks[N_LEVELS];
int i,status;
LOCKTAG xxTag, *tmpTag = &xxTag;
/*
* same level scheme as MultiAcquire().
*/
switch (level) {
case RELN_LEVEL:
locks[0] = lockt;
locks[1] = NO_LOCK;
locks[2] = NO_LOCK;
break;
case PAGE_LEVEL:
locks[0] = lockt + INTENT;
locks[1] = lockt;
locks[2] = NO_LOCK;
break;
case TUPLE_LEVEL:
locks[0] = lockt + INTENT;
locks[1] = lockt + INTENT;
locks[2] = lockt;
break;
default:
elog(WARN,"MultiRelease: bad lockt");
}
/*
* again, construct the tag on the fly. This time, however,
* we release the locks in the REVERSE order -- from lowest
* level to highest level.
*
* Must zero out the tag to set padding byes to zero and ensure
* hashing consistency.
*/
memset(tmpTag, 0, sizeof(*tmpTag));
tmpTag->relId = tag->relId;
tmpTag->dbId = tag->dbId;
for (i=(N_LEVELS-1); i>=0; i--) {
if (locks[i] != NO_LOCK) {
switch (i) {
case RELN_LEVEL:
/* -------------
* Set the block # and offset to invalid
* -------------
*/
BlockIdSet(&(tmpTag->tupleId.ip_blkid), InvalidBlockNumber);
tmpTag->tupleId.ip_posid = InvalidOffsetNumber;
break;
case PAGE_LEVEL:
/* -------------
* Copy the block #, set the offset to invalid
* -------------
*/
BlockIdCopy(&(tmpTag->tupleId.ip_blkid),
&(tag->tupleId.ip_blkid));
tmpTag->tupleId.ip_posid = InvalidOffsetNumber;
break;
case TUPLE_LEVEL:
ItemPointerCopy(&tmpTag->tupleId, &tag->tupleId);
break;
}
status = LockRelease(tableId, tmpTag, locks[i]);
if (! status) {
elog(WARN,"MultiRelease: couldn't release after error");
}
}
}
/* shouldn't reach here */
return false;
}
/*
* MultiAcquire -- acquire multi level lock at requested level
*
* Returns: TRUE if lock is set, FALSE if not
* Side Effects:
*/
bool
MultiAcquire(LockTableId tableId,
LOCKTAG *tag,
LOCKT lockt,
LOCK_LEVEL level)
{
LOCKT locks[N_LEVELS];
int i,status;
LOCKTAG xxTag, *tmpTag = &xxTag;
int retStatus = TRUE;
/*
* Three levels implemented. If we set a low level (e.g. Tuple)
* lock, we must set INTENT locks on the higher levels. The
* intent lock detects conflicts between the low level lock
* and an existing high level lock. For example, setting a
* write lock on a tuple in a relation is disallowed if there
* is an existing read lock on the entire relation. The
* write lock would set a WRITE + INTENT lock on the relation
* and that lock would conflict with the read.
*/
switch (level) {
case RELN_LEVEL:
locks[0] = lockt;
locks[1] = NO_LOCK;
locks[2] = NO_LOCK;
break;
case PAGE_LEVEL:
locks[0] = lockt + INTENT;
locks[1] = lockt;
locks[2] = NO_LOCK;
break;
case TUPLE_LEVEL:
locks[0] = lockt + INTENT;
locks[1] = lockt + INTENT;
locks[2] = lockt;
break;
default:
elog(WARN,"MultiAcquire: bad lock level");
return(FALSE);
}
/*
* construct a new tag as we go. Always loop through all levels,
* but if we arent' seting a low level lock, locks[i] is set to
* NO_LOCK for the lower levels. Always start from the highest
* level and go to the lowest level.
*/
memset(tmpTag,0,sizeof(*tmpTag));
tmpTag->relId = tag->relId;
tmpTag->dbId = tag->dbId;
for (i=0;i<N_LEVELS;i++) {
if (locks[i] != NO_LOCK) {
switch (i) {
case RELN_LEVEL:
/* -------------
* Set the block # and offset to invalid
* -------------
*/
BlockIdSet(&(tmpTag->tupleId.ip_blkid), InvalidBlockNumber);
tmpTag->tupleId.ip_posid = InvalidOffsetNumber;
break;
case PAGE_LEVEL:
/* -------------
* Copy the block #, set the offset to invalid
* -------------
*/
BlockIdCopy(&(tmpTag->tupleId.ip_blkid),
&(tag->tupleId.ip_blkid));
tmpTag->tupleId.ip_posid = InvalidOffsetNumber;
break;
case TUPLE_LEVEL:
/* --------------
* Copy the entire tuple id.
* --------------
*/
ItemPointerCopy(&tmpTag->tupleId, &tag->tupleId);
break;
}
status = LockAcquire(tableId, tmpTag, locks[i]);
if (! status) {
/* failed for some reason. Before returning we have
* to release all of the locks we just acquired.
* MultiRelease(xx,xx,xx, i) means release starting from
* the last level lock we successfully acquired
*/
retStatus = FALSE;
(void) MultiRelease(tableId, tag, lockt, i);
/* now leave the loop. Don't try for any more locks */
break;
}
}
}
return(retStatus);
}
/*
* Rebuild free TID list based on freeEntryHashTable. Returns number
* of free tuples in the rebuilt free list.
*/
uint64
PersistentStore_RebuildFreeList(
PersistentStoreData *storeData,
PersistentStoreSharedData *storeSharedData)
{
Datum *values;
PersistentStoreScan storeScan;
ItemPointerData persistentTid;
ItemPointerData previousFreeTid;
ItemPointerData previousTid;
uint64 persistentSerialNum;
uint64 freeOrderNum;
values = (Datum*)palloc(storeData->numAttributes * sizeof(Datum));
/*
* PT shared data must be already initialized, even when we are
* called during recovery.
*/
Assert(!PersistentStore_IsZeroTid(&storeSharedData->maxTid));
if (storeSharedData->maxFreeOrderNum < 1)
{
elog(LOG, "no free tuples in %s, not building any free list",
storeData->tableName);
return 0;
}
elog(LOG, "rebuilding free list in %s with " INT64_FORMAT " free tuples",
storeData->tableName, storeSharedData->maxFreeOrderNum);
/*
* Scan PT for free entries (in TID order) and establish links
* with previous free entry as we go on.
*/
previousTid.ip_posid = 0;
freeOrderNum = 0;
PersistentStore_BeginScan(storeData, storeSharedData, &storeScan);
while (PersistentStore_GetNext(
&storeScan,
values,
&persistentTid,
(int64 *)&persistentSerialNum))
{
/*
* We are scanning from low to high TID. All TIDs we
* encounter should be smaller or equal to the known
* maxTid.
*/
Assert(ItemPointerCompare(
&storeSharedData->maxTid,
&persistentTid) >= 0);
PersistentStore_ExtractOurTupleData(
storeData,
values,
(int64 *)&persistentSerialNum,
&previousFreeTid);
if (!PersistentStore_IsZeroTid(&previousFreeTid))
{
values[storeData->attNumPersistentSerialNum - 1] =
Int64GetDatum(++freeOrderNum);
values[storeData->attNumPreviousFreeTid - 1] =
ItemPointerIsValid(&previousTid) ?
PointerGetDatum(&previousTid) :
PointerGetDatum(&persistentTid);
#ifdef FAULT_INJECTOR
/*
* Inject fault after free list is partially built - a few
* tuples are updated but at least one is yet to be
* updated.
*/
if (freeOrderNum > 3)
{
FaultInjector_InjectFaultIfSet(
RebuildPTDB,
DDLNotSpecified,
"", // databaseName
""); // tableName
}
#endif
PersistentStore_UpdateTuple(
storeData, storeSharedData, &persistentTid, values, true);
ItemPointerCopy(&persistentTid, &previousTid);
}
}
PersistentStore_EndScan(&storeScan);
pfree(values);
if (ItemPointerIsValid(&previousTid))
{
Assert(freeOrderNum > 0);
ItemPointerCopy(&previousTid, &storeSharedData->freeTid);
storeSharedData->maxFreeOrderNum = freeOrderNum;
elog(LOG, "rebuilt free list in %s: maxFreeOrderNum = " INT64_FORMAT
" freeTid = %s", storeData->tableName, freeOrderNum,
ItemPointerToString(&persistentTid));
}
return freeOrderNum;
}
static ArrayTuple
gistVacuumUpdate(GistVacuum *gv, BlockNumber blkno, bool needunion)
{
ArrayTuple res = {NULL, 0, false};
Buffer buffer;
Page page,
tempPage = NULL;
OffsetNumber i,
maxoff;
ItemId iid;
int lenaddon = 4,
curlenaddon = 0,
nOffToDelete = 0,
nBlkToDelete = 0;
IndexTuple idxtuple,
*addon = NULL;
bool needwrite = false;
OffsetNumber offToDelete[MaxOffsetNumber];
BlockNumber blkToDelete[MaxOffsetNumber];
ItemPointerData *completed = NULL;
int ncompleted = 0,
lencompleted = 16;
vacuum_delay_point();
buffer = ReadBufferWithStrategy(gv->index, blkno, gv->strategy);
LockBuffer(buffer, GIST_EXCLUSIVE);
gistcheckpage(gv->index, buffer);
page = (Page) BufferGetPage(buffer);
maxoff = PageGetMaxOffsetNumber(page);
if (GistPageIsLeaf(page))
{
if (GistTuplesDeleted(page))
needunion = needwrite = true;
}
else
{
completed = (ItemPointerData *) palloc(sizeof(ItemPointerData) * lencompleted);
addon = (IndexTuple *) palloc(sizeof(IndexTuple) * lenaddon);
/* get copy of page to work */
tempPage = GistPageGetCopyPage(page);
for (i = FirstOffsetNumber; i <= maxoff; i = OffsetNumberNext(i))
{
ArrayTuple chldtuple;
bool needchildunion;
iid = PageGetItemId(tempPage, i);
idxtuple = (IndexTuple) PageGetItem(tempPage, iid);
needchildunion = (GistTupleIsInvalid(idxtuple)) ? true : false;
if (needchildunion)
elog(DEBUG2, "gistVacuumUpdate: need union for block %u",
ItemPointerGetBlockNumber(&(idxtuple->t_tid)));
chldtuple = gistVacuumUpdate(gv, ItemPointerGetBlockNumber(&(idxtuple->t_tid)),
needchildunion);
if (chldtuple.ituplen || chldtuple.emptypage)
{
/* update tuple or/and inserts new */
if (chldtuple.emptypage)
blkToDelete[nBlkToDelete++] = ItemPointerGetBlockNumber(&(idxtuple->t_tid));
offToDelete[nOffToDelete++] = i;
PageIndexTupleDelete(tempPage, i);
i--;
maxoff--;
needwrite = needunion = true;
if (chldtuple.ituplen)
{
Assert(chldtuple.emptypage == false);
while (curlenaddon + chldtuple.ituplen >= lenaddon)
{
lenaddon *= 2;
addon = (IndexTuple *) repalloc(addon, sizeof(IndexTuple) * lenaddon);
}
memcpy(addon + curlenaddon, chldtuple.itup, chldtuple.ituplen * sizeof(IndexTuple));
curlenaddon += chldtuple.ituplen;
if (chldtuple.ituplen > 1)
{
/*
* child was split, so we need mark completion
* insert(split)
*/
int j;
while (ncompleted + chldtuple.ituplen > lencompleted)
{
lencompleted *= 2;
completed = (ItemPointerData *) repalloc(completed, sizeof(ItemPointerData) * lencompleted);
}
for (j = 0; j < chldtuple.ituplen; j++)
{
ItemPointerCopy(&(chldtuple.itup[j]->t_tid), completed + ncompleted);
ncompleted++;
}
}
pfree(chldtuple.itup);
}
}
}
Assert(maxoff == PageGetMaxOffsetNumber(tempPage));
if (curlenaddon)
{
/* insert updated tuples */
if (gistnospace(tempPage, addon, curlenaddon, InvalidOffsetNumber, 0))
{
/* there is no space on page to insert tuples */
res = vacuumSplitPage(gv, tempPage, buffer, addon, curlenaddon);
tempPage = NULL; /* vacuumSplitPage() free tempPage */
needwrite = needunion = false; /* gistSplit already forms
* unions and writes pages */
}
else
/* enough free space */
gistfillbuffer(gv->index, tempPage, addon, curlenaddon, InvalidOffsetNumber);
}
}
/*
* If page is empty, we should remove pointer to it before deleting page
* (except root)
*/
if (blkno != GIST_ROOT_BLKNO && (PageIsEmpty(page) || (tempPage && PageIsEmpty(tempPage))))
{
/*
* New version of page is empty, so leave it unchanged, upper call
* will mark our page as deleted. In case of page split we never will
* be here...
*
* If page was empty it can't become non-empty during processing
*/
res.emptypage = true;
UnlockReleaseBuffer(buffer);
}
else
{
/* write page and remove its childs if it need */
START_CRIT_SECTION();
if (tempPage && needwrite)
{
PageRestoreTempPage(tempPage, page);
tempPage = NULL;
}
/* Empty index */
if (PageIsEmpty(page) && blkno == GIST_ROOT_BLKNO)
{
needwrite = true;
GistPageSetLeaf(page);
}
if (needwrite)
{
MarkBufferDirty(buffer);
GistClearTuplesDeleted(page);
if (!gv->index->rd_istemp)
{
XLogRecData *rdata;
XLogRecPtr recptr;
char *xlinfo;
rdata = formUpdateRdata(gv->index->rd_node, buffer,
offToDelete, nOffToDelete,
addon, curlenaddon, NULL);
xlinfo = rdata->next->data;
recptr = XLogInsert(RM_GIST_ID, XLOG_GIST_PAGE_UPDATE, rdata);
PageSetLSN(page, recptr);
PageSetTLI(page, ThisTimeLineID);
pfree(xlinfo);
pfree(rdata);
}
else
PageSetLSN(page, XLogRecPtrForTemp);
}
END_CRIT_SECTION();
if (needunion && !PageIsEmpty(page))
{
res.itup = (IndexTuple *) palloc(sizeof(IndexTuple));
res.ituplen = 1;
res.itup[0] = PageMakeUnionKey(gv, buffer);
}
UnlockReleaseBuffer(buffer);
/* delete empty children, now we havn't any links to pointed subtrees */
for (i = 0; i < nBlkToDelete; i++)
gistDeleteSubtree(gv, blkToDelete[i]);
if (ncompleted && !gv->index->rd_istemp)
gistxlogInsertCompletion(gv->index->rd_node, completed, ncompleted);
}
for (i = 0; i < curlenaddon; i++)
pfree(addon[i]);
if (addon)
pfree(addon);
if (completed)
pfree(completed);
if (tempPage)
pfree(tempPage);
return res;
}
/*
* extract_minipage
*
* Extract the minipage info from the given tuple. The tupleTid
* is also set here.
*/
static void
extract_minipage(AppendOnlyBlockDirectory *blockDirectory,
HeapTuple tuple,
TupleDesc tupleDesc,
int columnGroupNo)
{
Datum *values = blockDirectory->values;
bool *nulls = blockDirectory->nulls;
MinipagePerColumnGroup *minipageInfo =
&blockDirectory->minipages[columnGroupNo];
FileSegInfo *fsInfo = blockDirectory->currentSegmentFileInfo;
int64 eof;
int start, end, mid=0;
bool found = false;
heap_deform_tuple(tuple, tupleDesc, values, nulls);
Assert(blockDirectory->currentSegmentFileNum ==
DatumGetInt32(values[Anum_pg_aoblkdir_segno - 1]));
/*
* Copy out the minipage
*/
copy_out_minipage(minipageInfo,
values[Anum_pg_aoblkdir_minipage - 1],
nulls[Anum_pg_aoblkdir_minipage - 1]);
ItemPointerCopy(&tuple->t_self, &minipageInfo->tupleTid);
/*
* When crashes during inserts, or cancellation during inserts,
* there are out-of-date minipage entries in the block directory.
* We reset those entries here.
*/
Assert(fsInfo != NULL);
if (!blockDirectory->isAOCol)
eof = fsInfo->eof;
else
eof = ((AOCSFileSegInfo *)fsInfo)->vpinfo.entry[columnGroupNo].eof;
start = 0;
end = minipageInfo->numMinipageEntries - 1;
while (start <= end)
{
mid = (end - start + 1) / 2 + start;
if (minipageInfo->minipage->entry[mid].fileOffset > eof)
end = mid - 1;
else if (minipageInfo->minipage->entry[mid].fileOffset < eof)
start = mid + 1;
else
{
found = true;
break;
}
}
minipageInfo->numMinipageEntries = 0;
if (found)
minipageInfo->numMinipageEntries = mid;
else if (start > 0)
{
minipageInfo->numMinipageEntries = start;
Assert(minipageInfo->minipage->entry[start - 1].fileOffset < eof);
}
}
/*
* 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;
ItemPointerData persistentTid;
int64 persistentSerialNum;
if (Persistent_BeforePersistenceWork())
elog(ERROR, "persistent table changes forbidden");
PersistentFilespace_VerifyInitScan();
PersistentFileSysObjName_SetFilespaceDir(&fsObjName, filespace);
WRITE_PERSISTENT_STATE_ORDERED_LOCK;
LWLockAcquire(FilespaceHashLock, LW_SHARED);
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);
}
ItemPointerCopy(&fde->persistentTid, &persistentTid);
persistentSerialNum = fde->persistentSerialNum;
LWLockRelease(FilespaceHashLock);
PersistentFileSysObj_AddMirror(&fsObjName,
&persistentTid,
persistentSerialNum,
pridbid,
mirdbid,
(void *)newpath,
set_mirror_existence,
/* flushToXlog */ false);
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;
}
/*
* 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));
}
/*
* Post vacuum, iterate over all entries in index, check if the h_tid
* of each entry exists and is not dead. For specific system tables,
* also ensure that the key in index entry matches the corresponding
* attribute in the heap tuple.
*/
void
_bt_validate_vacuum(Relation irel, Relation hrel, TransactionId oldest_xmin)
{
MIRROREDLOCK_BUFMGR_DECLARE;
BlockNumber blkno;
BlockNumber num_pages;
Buffer ibuf = InvalidBuffer;
Buffer hbuf = InvalidBuffer;
Page ipage;
BTPageOpaque opaque;
IndexTuple itup;
HeapTupleData htup;
OffsetNumber maxoff,
minoff,
offnum;
Oid ioid,
hoid;
bool isnull;
blkno = BTREE_METAPAGE + 1;
num_pages = RelationGetNumberOfBlocks(irel);
elog(LOG, "btvalidatevacuum: index %s, heap %s",
RelationGetRelationName(irel), RelationGetRelationName(hrel));
MIRROREDLOCK_BUFMGR_LOCK;
for (; blkno < num_pages; blkno++)
{
ibuf = ReadBuffer(irel, blkno);
ipage = BufferGetPage(ibuf);
opaque = (BTPageOpaque) PageGetSpecialPointer(ipage);
if (!PageIsNew(ipage))
_bt_checkpage(irel, ibuf);
if (P_ISLEAF(opaque))
{
minoff = P_FIRSTDATAKEY(opaque);
maxoff = PageGetMaxOffsetNumber(ipage);
for (offnum = minoff;
offnum <= maxoff;
offnum = OffsetNumberNext(offnum))
{
itup = (IndexTuple) PageGetItem(ipage,
PageGetItemId(ipage, offnum));
ItemPointerCopy(&itup->t_tid, &htup.t_self);
/*
* TODO: construct a tid bitmap based on index tids
* and fetch heap tids in order afterwards. That will
* also allow validating if a heap tid appears twice
* in a unique index.
*/
if (!heap_release_fetch(hrel, SnapshotAny, &htup,
&hbuf, true, NULL))
{
elog(ERROR, "btvalidatevacuum: tid (%d,%d) from index %s "
"not found in heap %s",
ItemPointerGetBlockNumber(&itup->t_tid),
ItemPointerGetOffsetNumber(&itup->t_tid),
RelationGetRelationName(irel),
RelationGetRelationName(hrel));
}
switch (HeapTupleSatisfiesVacuum(hrel, htup.t_data, oldest_xmin, hbuf))
{
case HEAPTUPLE_RECENTLY_DEAD:
case HEAPTUPLE_LIVE:
case HEAPTUPLE_INSERT_IN_PROGRESS:
case HEAPTUPLE_DELETE_IN_PROGRESS:
/* these tuples are considered alive by vacuum */
break;
case HEAPTUPLE_DEAD:
elog(ERROR, "btvalidatevacuum: vacuum did not remove "
"dead tuple (%d,%d) from heap %s and index %s",
ItemPointerGetBlockNumber(&itup->t_tid),
ItemPointerGetOffsetNumber(&itup->t_tid),
RelationGetRelationName(hrel),
RelationGetRelationName(irel));
break;
default:
elog(ERROR, "btvalidatevacuum: invalid visibility");
break;
}
switch(RelationGetRelid(irel))
{
case DatabaseOidIndexId:
case TypeOidIndexId:
case ClassOidIndexId:
case ConstraintOidIndexId:
hoid = HeapTupleGetOid(&htup);
ioid = index_getattr(itup, 1, RelationGetDescr(irel), &isnull);
if (hoid != ioid)
{
elog(ERROR,
"btvalidatevacuum: index oid(%d) != heap oid(%d)"
" tuple (%d,%d) index %s", ioid, hoid,
ItemPointerGetBlockNumber(&itup->t_tid),
ItemPointerGetOffsetNumber(&itup->t_tid),
RelationGetRelationName(irel));
}
break;
case GpRelationNodeOidIndexId:
hoid = heap_getattr(&htup, 1, RelationGetDescr(hrel), &isnull);
ioid = index_getattr(itup, 1, RelationGetDescr(irel), &isnull);
if (hoid != ioid)
{
elog(ERROR,
"btvalidatevacuum: index oid(%d) != heap oid(%d)"
" tuple (%d,%d) index %s", ioid, hoid,
ItemPointerGetBlockNumber(&itup->t_tid),
ItemPointerGetOffsetNumber(&itup->t_tid),
RelationGetRelationName(irel));
}
int4 hsegno = heap_getattr(&htup, 2, RelationGetDescr(hrel), &isnull);
int4 isegno = index_getattr(itup, 2, RelationGetDescr(irel), &isnull);
if (isegno != hsegno)
{
elog(ERROR,
"btvalidatevacuum: index segno(%d) != heap segno(%d)"
" tuple (%d,%d) index %s", isegno, hsegno,
ItemPointerGetBlockNumber(&itup->t_tid),
ItemPointerGetOffsetNumber(&itup->t_tid),
RelationGetRelationName(irel));
}
break;
default:
break;
}
if (RelationGetNamespace(irel) == PG_AOSEGMENT_NAMESPACE)
{
int4 isegno = index_getattr(itup, 1, RelationGetDescr(irel), &isnull);
int4 hsegno = heap_getattr(&htup, 1, RelationGetDescr(hrel), &isnull);
if (isegno != hsegno)
{
elog(ERROR,
"btvalidatevacuum: index segno(%d) != heap segno(%d)"
" tuple (%d,%d) index %s", isegno, hsegno,
ItemPointerGetBlockNumber(&itup->t_tid),
ItemPointerGetOffsetNumber(&itup->t_tid),
RelationGetRelationName(irel));
}
}
}
}
if (BufferIsValid(ibuf))
ReleaseBuffer(ibuf);
}
if (BufferIsValid(hbuf))
ReleaseBuffer(hbuf);
MIRROREDLOCK_BUFMGR_UNLOCK;
}
/*
* Search the relation 'rel' for tuple using the sequential scan.
*
* If a matching tuple is found, lock it with lockmode, fill the slot with its
* contents, and return true. Return false otherwise.
*
* Note that this stops on the first matching tuple.
*
* This can obviously be quite slow on tables that have more than few rows.
*/
bool
RelationFindReplTupleSeq(Relation rel, LockTupleMode lockmode,
TupleTableSlot *searchslot, TupleTableSlot *outslot)
{
HeapTuple scantuple;
HeapScanDesc scan;
SnapshotData snap;
TransactionId xwait;
bool found;
TupleDesc desc = RelationGetDescr(rel);
Assert(equalTupleDescs(desc, outslot->tts_tupleDescriptor));
/* Start an index scan. */
InitDirtySnapshot(snap);
scan = heap_beginscan(rel, &snap, 0, NULL);
retry:
found = false;
heap_rescan(scan, NULL);
/* Try to find the tuple */
while ((scantuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
if (!tuple_equals_slot(desc, scantuple, searchslot))
continue;
found = true;
ExecStoreTuple(scantuple, outslot, InvalidBuffer, false);
ExecMaterializeSlot(outslot);
xwait = TransactionIdIsValid(snap.xmin) ?
snap.xmin : snap.xmax;
/*
* If the tuple is locked, wait for locking transaction to finish and
* retry.
*/
if (TransactionIdIsValid(xwait))
{
XactLockTableWait(xwait, NULL, NULL, XLTW_None);
goto retry;
}
}
/* Found tuple, try to lock it in the lockmode. */
if (found)
{
Buffer buf;
HeapUpdateFailureData hufd;
HTSU_Result res;
HeapTupleData locktup;
ItemPointerCopy(&outslot->tts_tuple->t_self, &locktup.t_self);
PushActiveSnapshot(GetLatestSnapshot());
res = heap_lock_tuple(rel, &locktup, GetCurrentCommandId(false),
lockmode,
LockWaitBlock,
false /* don't follow updates */ ,
&buf, &hufd);
/* the tuple slot already has the buffer pinned */
ReleaseBuffer(buf);
PopActiveSnapshot();
switch (res)
{
case HeapTupleMayBeUpdated:
break;
case HeapTupleUpdated:
/* XXX: Improve handling here */
ereport(LOG,
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
errmsg("concurrent update, retrying")));
goto retry;
case HeapTupleInvisible:
elog(ERROR, "attempted to lock invisible tuple");
default:
elog(ERROR, "unexpected heap_lock_tuple status: %u", res);
break;
}
}
heap_endscan(scan);
return found;
}
