void
cdb_sync_oid_to_segments(void)
{
if (Gp_role == GP_ROLE_DISPATCH && IsNormalProcessingMode())
{
int i;
/* Burn a few extra just for safety */
for (i=0;i<10;i++)
GetNewObjectId();
}
}
/*
* hashbuild() -- build a new hash index.
*
* We use a global variable to record the fact that we're creating
* a new index. This is used to avoid high-concurrency locking,
* since the index won't be visible until this transaction commits
* and since building is guaranteed to be single-threaded.
*/
Datum
hashbuild(PG_FUNCTION_ARGS)
{
Relation heap = (Relation) PG_GETARG_POINTER(0);
Relation index = (Relation) PG_GETARG_POINTER(1);
IndexInfo *indexInfo = (IndexInfo *) PG_GETARG_POINTER(2);
double reltuples;
HashBuildState buildstate;
/*
* We expect to be called exactly once for any index relation. If
* that's not the case, big trouble's what we have.
*/
if (RelationGetNumberOfBlocks(index) != 0)
elog(ERROR, "index \"%s\" already contains data",
RelationGetRelationName(index));
/* initialize the hash index metadata page */
_hash_metapinit(index);
/* build the index */
buildstate.indtuples = 0;
/* do the heap scan */
reltuples = IndexBuildHeapScan(heap, index, indexInfo,
hashbuildCallback, (void *) &buildstate);
/*
* Since we just counted the tuples in the heap, we update its stats
* in pg_class to guarantee that the planner takes advantage of the
* index we just created. But, only update statistics during normal
* index definitions, not for indices on system catalogs created
* during bootstrap processing. We must close the relations before
* updating statistics to guarantee that the relcache entries are
* flushed when we increment the command counter in UpdateStats(). But
* we do not release any locks on the relations; those will be held
* until end of transaction.
*/
if (IsNormalProcessingMode())
{
Oid hrelid = RelationGetRelid(heap);
Oid irelid = RelationGetRelid(index);
heap_close(heap, NoLock);
index_close(index);
UpdateStats(hrelid, reltuples);
UpdateStats(irelid, buildstate.indtuples);
}
PG_RETURN_VOID();
}
void
cdb_sync_oid_to_segments(void)
{
if (Gp_role == GP_ROLE_DISPATCH && IsNormalProcessingMode())
{
Oid max_oid = get_max_oid_from_segDBs();
/* Move our oid counter ahead of QEs */
while(GetNewObjectId() <= max_oid);
/* Burn a few extra just for safety */
for (int i = 0; i < 10; i++)
GetNewObjectId();
}
}
/*
* AtCommit_Notify
*
* This is called at transaction commit.
*
* If there are pending LISTEN/UNLISTEN actions, insert or delete
* tuples in pg_listener accordingly.
*
* If there are outbound notify requests in the pendingNotifies list,
* scan pg_listener for matching tuples, and either signal the other
* backend or send a message to our own frontend.
*
* NOTE: we are still inside the current transaction, therefore can
* piggyback on its committing of changes.
*/
void
AtCommit_Notify(void)
{
Relation lRel;
ListCell *p;
if (pendingActions == NIL && pendingNotifies == NIL)
return; /* no relevant statements in this xact */
/*
* NOTIFY is disabled if not normal processing mode. This test used to be
* in xact.c, but it seems cleaner to do it here.
*/
if (!IsNormalProcessingMode())
{
ClearPendingActionsAndNotifies();
return;
}
if (Trace_notify)
elog(DEBUG1, "AtCommit_Notify");
/* Acquire ExclusiveLock on pg_listener */
lRel = heap_open(ListenerRelationId, ExclusiveLock);
/* Perform any pending listen/unlisten actions */
foreach(p, pendingActions)
{
ListenAction *actrec = (ListenAction *) lfirst(p);
switch (actrec->action)
{
case LISTEN_LISTEN:
Exec_Listen(lRel, actrec->condname);
break;
case LISTEN_UNLISTEN:
Exec_Unlisten(lRel, actrec->condname);
break;
case LISTEN_UNLISTEN_ALL:
Exec_UnlistenAll(lRel);
break;
}
/* We must CCI after each action in case of conflicting actions */
CommandCounterIncrement();
}
/*
*--------------------------------------------------------------
* AtCommit_Notify
*
* This is called at transaction commit.
*
* If there are outbound notify requests in the pendingNotifies list,
* scan pg_listener for matching tuples, and either signal the other
* backend or send a message to our own frontend.
*
* NOTE: we are still inside the current transaction, therefore can
* piggyback on its committing of changes.
*
* Results:
* XXX
*
* Side effects:
* Tuples in pg_listener that have matching relnames and other peoples'
* listenerPIDs are updated with a nonzero notification field.
*
*--------------------------------------------------------------
*/
void
AtCommit_Notify(void)
{
Relation lRel;
TupleDesc tdesc;
HeapScanDesc scan;
HeapTuple lTuple,
rTuple;
Datum value[Natts_pg_listener];
char repl[Natts_pg_listener],
nulls[Natts_pg_listener];
if (pendingNotifies == NIL)
return; /* no NOTIFY statements in this transaction */
/*
* NOTIFY is disabled if not normal processing mode. This test used to be
* in xact.c, but it seems cleaner to do it here.
*/
if (!IsNormalProcessingMode())
{
ClearPendingNotifies();
return;
}
if (Trace_notify)
elog(DEBUG1, "AtCommit_Notify");
/* preset data to update notify column to MyProcPid */
nulls[0] = nulls[1] = nulls[2] = ' ';
repl[0] = repl[1] = repl[2] = ' ';
repl[Anum_pg_listener_notify - 1] = 'r';
value[0] = value[1] = value[2] = (Datum) 0;
value[Anum_pg_listener_notify - 1] = Int32GetDatum(MyProcPid);
lRel = heap_open(ListenerRelationId, ExclusiveLock);
tdesc = RelationGetDescr(lRel);
scan = heap_beginscan(lRel, SnapshotNow, 0, NULL);
while ((lTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
Form_pg_listener listener = (Form_pg_listener) GETSTRUCT(lTuple);
char *relname = NameStr(listener->relname);
int32 listenerPID = listener->listenerpid;
if (!AsyncExistsPendingNotify(relname))
continue;
if (listenerPID == MyProcPid)
{
/*
* Self-notify: no need to bother with table update. Indeed, we
* *must not* clear the notification field in this path, or we
* could lose an outside notify, which'd be bad for applications
* that ignore self-notify messages.
*/
if (Trace_notify)
elog(DEBUG1, "AtCommit_Notify: notifying self");
NotifyMyFrontEnd(relname, listenerPID);
}
else
{
if (Trace_notify)
elog(DEBUG1, "AtCommit_Notify: notifying pid %d",
listenerPID);
/*
* If someone has already notified this listener, we don't bother
* modifying the table, but we do still send a SIGUSR2 signal,
* just in case that backend missed the earlier signal for some
* reason. It's OK to send the signal first, because the other
* guy can't read pg_listener until we unlock it.
*/
if (kill(listenerPID, SIGUSR2) < 0)
{
/*
* Get rid of pg_listener entry if it refers to a PID that no
* longer exists. Presumably, that backend crashed without
* deleting its pg_listener entries. This code used to only
* delete the entry if errno==ESRCH, but as far as I can see
* we should just do it for any failure (certainly at least
* for EPERM too...)
*/
simple_heap_delete(lRel, &lTuple->t_self);
}
else if (listener->notification == 0)
{
HTSU_Result result;
ItemPointerData update_ctid;
TransactionId update_xmax;
rTuple = heap_modifytuple(lTuple, tdesc,
value, nulls, repl);
/*
* We cannot use simple_heap_update here because the tuple
* could have been modified by an uncommitted transaction;
* specifically, since UNLISTEN releases exclusive lock on the
* table before commit, the other guy could already have tried
* to unlisten. There are no other cases where we should be
* able to see an uncommitted update or delete. Therefore, our
* response to a HeapTupleBeingUpdated result is just to
* ignore it. We do *not* wait for the other guy to commit
* --- that would risk deadlock, and we don't want to block
* while holding the table lock anyway for performance
* reasons. We also ignore HeapTupleUpdated, which could occur
* if the other guy commits between our heap_getnext and
* heap_update calls.
*/
result = heap_update(lRel, &lTuple->t_self, rTuple,
&update_ctid, &update_xmax,
GetCurrentCommandId(), InvalidSnapshot,
false /* no wait for commit */ );
switch (result)
{
case HeapTupleSelfUpdated:
/* Tuple was already updated in current command? */
elog(ERROR, "tuple already updated by self");
break;
case HeapTupleMayBeUpdated:
/* done successfully */
#ifdef NOT_USED /* currently there are no indexes */
CatalogUpdateIndexes(lRel, rTuple);
#endif
break;
case HeapTupleBeingUpdated:
/* ignore uncommitted tuples */
break;
case HeapTupleUpdated:
/* ignore just-committed tuples */
break;
default:
elog(ERROR, "unrecognized heap_update status: %u",
result);
break;
}
}
}
}
heap_endscan(scan);
/*
* We do NOT release the lock on pg_listener here; we need to hold it
* until end of transaction (which is about to happen, anyway) to ensure
* that notified backends see our tuple updates when they look. Else they
* might disregard the signal, which would make the application programmer
* very unhappy.
*/
heap_close(lRel, NoLock);
ClearPendingNotifies();
if (Trace_notify)
elog(DEBUG1, "AtCommit_Notify: done");
}
/*
* routine to build an index. Basically calls insert over and over
*/
Datum
gistbuild(PG_FUNCTION_ARGS)
{
Relation heap = (Relation) PG_GETARG_POINTER(0);
Relation index = (Relation) PG_GETARG_POINTER(1);
IndexInfo *indexInfo = (IndexInfo *) PG_GETARG_POINTER(2);
double reltuples;
GISTBuildState buildstate;
Buffer buffer;
/* no locking is needed */
initGISTstate(&buildstate.giststate, index);
/*
* We expect to be called exactly once for any index relation. If
* that's not the case, big trouble's what we have.
*/
if (RelationGetNumberOfBlocks(index) != 0)
elog(ERROR, "index \"%s\" already contains data",
RelationGetRelationName(index));
/* initialize the root page */
buffer = ReadBuffer(index, P_NEW);
GISTInitBuffer(buffer, F_LEAF);
WriteBuffer(buffer);
/* build the index */
buildstate.numindexattrs = indexInfo->ii_NumIndexAttrs;
buildstate.indtuples = 0;
/* do the heap scan */
reltuples = IndexBuildHeapScan(heap, index, indexInfo,
gistbuildCallback, (void *) &buildstate);
/* okay, all heap tuples are indexed */
/*
* Since we just counted the tuples in the heap, we update its stats
* in pg_class to guarantee that the planner takes advantage of the
* index we just created. But, only update statistics during normal
* index definitions, not for indices on system catalogs created
* during bootstrap processing. We must close the relations before
* updating statistics to guarantee that the relcache entries are
* flushed when we increment the command counter in UpdateStats(). But
* we do not release any locks on the relations; those will be held
* until end of transaction.
*/
if (IsNormalProcessingMode())
{
Oid hrelid = RelationGetRelid(heap);
Oid irelid = RelationGetRelid(index);
heap_close(heap, NoLock);
index_close(index);
UpdateStats(hrelid, reltuples);
UpdateStats(irelid, buildstate.indtuples);
}
freeGISTstate(&buildstate.giststate);
#ifdef GISTDEBUG
gist_dumptree(index, 0, GISTP_ROOT, 0);
#endif
PG_RETURN_VOID();
}
static void
DtmXactCallback(XactEvent event, void *arg)
{
//XTM_INFO("%d: DtmXactCallbackevent=%d nextxid=%d\n", getpid(), event, DtmNextXid);
switch (event)
{
case XACT_EVENT_START:
//XTM_INFO("%d: normal=%d, initialized=%d, replication=%d, bgw=%d, vacuum=%d\n",
// getpid(), IsNormalProcessingMode(), dtm->initialized, MMDoReplication, IsBackgroundWorker, IsAutoVacuumWorkerProcess());
if (IsNormalProcessingMode() && dtm->initialized && MMDoReplication && !am_walsender && !IsBackgroundWorker && !IsAutoVacuumWorkerProcess()) {
MMBeginTransaction();
}
break;
#if 0
case XACT_EVENT_PRE_COMMIT:
case XACT_EVENT_PARALLEL_PRE_COMMIT:
{
TransactionId xid = GetCurrentTransactionIdIfAny();
if (!MMIsDistributedTrans && TransactionIdIsValid(xid)) {
XTM_INFO("%d: Will ignore transaction %u\n", getpid(), xid);
MMMarkTransAsLocal(xid);
}
break;
}
#endif
case XACT_EVENT_COMMIT:
case XACT_EVENT_ABORT:
if (TransactionIdIsValid(DtmNextXid))
{
if (!DtmVoted) {
ArbiterSetTransStatus(DtmNextXid, TRANSACTION_STATUS_ABORTED, false);
}
if (event == XACT_EVENT_COMMIT)
{
/*
* Now transaction status is already written in CLOG,
* so we can remove information about it from hash table
*/
LWLockAcquire(dtm->hashLock, LW_EXCLUSIVE);
hash_search(xid_in_doubt, &DtmNextXid, HASH_REMOVE, NULL);
LWLockRelease(dtm->hashLock);
}
#if 0 /* should be handled now using DtmVoted flag */
else
{
/*
* Transaction at the node can be aborted because of transaction failure at some other node
* before it starts doing anything and assigned Xid, in this case Postgres is not calling SetTransactionStatus,
* so we have to send report to DTMD here
*/
if (!TransactionIdIsValid(GetCurrentTransactionIdIfAny())) {
XTM_INFO("%d: abort transation on DTMD\n", getpid());
ArbiterSetTransStatus(DtmNextXid, TRANSACTION_STATUS_ABORTED, false);
}
}
#endif
DtmNextXid = InvalidTransactionId;
DtmLastSnapshot = NULL;
}
MMIsDistributedTrans = false;
break;
default:
break;
}
}
/* ----------------------------------------------------------------
* index_create
*
* Returns OID of the created index.
* ----------------------------------------------------------------
*/
Oid
index_create(Oid heapRelationId,
const char *indexRelationName,
IndexInfo *indexInfo,
Oid accessMethodObjectId,
Oid *classObjectId,
bool primary,
bool isconstraint,
bool allow_system_table_mods)
{
Relation heapRelation;
Relation indexRelation;
TupleDesc indexTupDesc;
bool shared_relation;
Oid namespaceId;
Oid indexoid;
int i;
/*
* Only SELECT ... FOR UPDATE are allowed while doing this
*/
heapRelation = heap_open(heapRelationId, ShareLock);
/*
* The index will be in the same namespace as its parent table, and is
* shared across databases if and only if the parent is.
*/
namespaceId = RelationGetNamespace(heapRelation);
shared_relation = heapRelation->rd_rel->relisshared;
/*
* check parameters
*/
if (indexInfo->ii_NumIndexAttrs < 1)
elog(ERROR, "must index at least one column");
if (!allow_system_table_mods &&
IsSystemRelation(heapRelation) &&
IsNormalProcessingMode())
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("user-defined indexes on system catalog tables are not supported")));
/*
* We cannot allow indexing a shared relation after initdb (because
* there's no way to make the entry in other databases' pg_class).
* Unfortunately we can't distinguish initdb from a manually started
* standalone backend. However, we can at least prevent this mistake
* under normal multi-user operation.
*/
if (shared_relation && IsUnderPostmaster)
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("shared indexes cannot be created after initdb")));
if (get_relname_relid(indexRelationName, namespaceId))
ereport(ERROR,
(errcode(ERRCODE_DUPLICATE_TABLE),
errmsg("relation \"%s\" already exists",
indexRelationName)));
/*
* construct tuple descriptor for index tuples
*/
indexTupDesc = ConstructTupleDescriptor(heapRelation,
indexInfo,
classObjectId);
/*
* create the index relation's relcache entry and physical disk file.
* (If we fail further down, it's the smgr's responsibility to remove
* the disk file again.)
*/
indexRelation = heap_create(indexRelationName,
namespaceId,
indexTupDesc,
shared_relation,
true,
allow_system_table_mods);
/* Fetch the relation OID assigned by heap_create */
indexoid = RelationGetRelid(indexRelation);
/*
* Obtain exclusive lock on it. Although no other backends can see it
* until we commit, this prevents deadlock-risk complaints from lock
* manager in cases such as CLUSTER.
*/
LockRelation(indexRelation, AccessExclusiveLock);
/*
* Fill in fields of the index's pg_class entry that are not set
* correctly by heap_create.
*
* XXX should have a cleaner way to create cataloged indexes
*/
indexRelation->rd_rel->relowner = GetUserId();
indexRelation->rd_rel->relam = accessMethodObjectId;
indexRelation->rd_rel->relkind = RELKIND_INDEX;
indexRelation->rd_rel->relhasoids = false;
/*
* store index's pg_class entry
*/
UpdateRelationRelation(indexRelation);
/*
* now update the object id's of all the attribute tuple forms in the
* index relation's tuple descriptor
*/
InitializeAttributeOids(indexRelation,
indexInfo->ii_NumIndexAttrs,
indexoid);
/*
* append ATTRIBUTE tuples for the index
*/
AppendAttributeTuples(indexRelation, indexInfo->ii_NumIndexAttrs);
/* ----------------
* update pg_index
* (append INDEX tuple)
*
* Note that this stows away a representation of "predicate".
* (Or, could define a rule to maintain the predicate) --Nels, Feb '92
* ----------------
*/
UpdateIndexRelation(indexoid, heapRelationId, indexInfo,
classObjectId, primary);
/*
* Register constraint and dependencies for the index.
*
* If the index is from a CONSTRAINT clause, construct a pg_constraint
* entry. The index is then linked to the constraint, which in turn
* is linked to the table. If it's not a CONSTRAINT, make the
* dependency directly on the table.
*
* We don't need a dependency on the namespace, because there'll be an
* indirect dependency via our parent table.
*
* During bootstrap we can't register any dependencies, and we don't try
* to make a constraint either.
*/
if (!IsBootstrapProcessingMode())
{
ObjectAddress myself,
referenced;
myself.classId = RelOid_pg_class;
myself.objectId = indexoid;
myself.objectSubId = 0;
if (isconstraint)
{
char constraintType;
Oid conOid;
if (primary)
constraintType = CONSTRAINT_PRIMARY;
else if (indexInfo->ii_Unique)
constraintType = CONSTRAINT_UNIQUE;
else
{
elog(ERROR, "constraint must be PRIMARY or UNIQUE");
constraintType = 0; /* keep compiler quiet */
}
/* Shouldn't have any expressions */
if (indexInfo->ii_Expressions)
elog(ERROR, "constraints can't have index expressions");
conOid = CreateConstraintEntry(indexRelationName,
namespaceId,
constraintType,
false, /* isDeferrable */
false, /* isDeferred */
heapRelationId,
indexInfo->ii_KeyAttrNumbers,
indexInfo->ii_NumIndexAttrs,
InvalidOid, /* no domain */
InvalidOid, /* no foreign key */
NULL,
0,
' ',
' ',
' ',
InvalidOid, /* no associated index */
NULL, /* no check constraint */
NULL,
NULL);
referenced.classId = get_system_catalog_relid(ConstraintRelationName);
referenced.objectId = conOid;
referenced.objectSubId = 0;
recordDependencyOn(&myself, &referenced, DEPENDENCY_INTERNAL);
}
else
{
/* Create auto dependencies on simply-referenced columns */
for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
{
if (indexInfo->ii_KeyAttrNumbers[i] != 0)
{
referenced.classId = RelOid_pg_class;
referenced.objectId = heapRelationId;
referenced.objectSubId = indexInfo->ii_KeyAttrNumbers[i];
recordDependencyOn(&myself, &referenced, DEPENDENCY_AUTO);
}
}
}
/* Store dependency on operator classes */
referenced.classId = get_system_catalog_relid(OperatorClassRelationName);
for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
{
referenced.objectId = classObjectId[i];
referenced.objectSubId = 0;
recordDependencyOn(&myself, &referenced, DEPENDENCY_NORMAL);
}
/* Store dependencies on anything mentioned in index expressions */
if (indexInfo->ii_Expressions)
{
recordDependencyOnSingleRelExpr(&myself,
(Node *) indexInfo->ii_Expressions,
heapRelationId,
DEPENDENCY_NORMAL,
DEPENDENCY_AUTO);
}
/* Store dependencies on anything mentioned in predicate */
if (indexInfo->ii_Predicate)
{
recordDependencyOnSingleRelExpr(&myself,
(Node *) indexInfo->ii_Predicate,
heapRelationId,
DEPENDENCY_NORMAL,
DEPENDENCY_AUTO);
}
}
/*
* Advance the command counter so that we can see the newly-entered
* catalog tuples for the index.
*/
CommandCounterIncrement();
/*
* In bootstrap mode, we have to fill in the index strategy structure
* with information from the catalogs. If we aren't bootstrapping,
* then the relcache entry has already been rebuilt thanks to sinval
* update during CommandCounterIncrement.
*/
if (IsBootstrapProcessingMode())
RelationInitIndexAccessInfo(indexRelation);
else
Assert(indexRelation->rd_indexcxt != NULL);
/*
* If this is bootstrap (initdb) time, then we don't actually fill in
* the index yet. We'll be creating more indexes and classes later,
* so we delay filling them in until just before we're done with
* bootstrapping. Otherwise, we call the routine that constructs the
* index.
*
* In normal processing mode, the heap and index relations are closed by
* index_build() --- but we continue to hold the ShareLock on the heap
* and the exclusive lock on the index that we acquired above, until
* end of transaction.
*/
if (IsBootstrapProcessingMode())
{
index_register(heapRelationId, indexoid, indexInfo);
/* XXX shouldn't we close the heap and index rels here? */
}
else
index_build(heapRelation, indexRelation, indexInfo);
return indexoid;
}