/*
* create_distributed_table gets a table name, distribution column,
* distribution method and colocate_with option, then it creates a
* distributed table.
*/
Datum
create_distributed_table(PG_FUNCTION_ARGS)
{
Oid relationId = InvalidOid;
text *distributionColumnText = NULL;
Oid distributionMethodOid = InvalidOid;
text *colocateWithTableNameText = NULL;
Relation relation = NULL;
char *distributionColumnName = NULL;
Var *distributionColumn = NULL;
char distributionMethod = 0;
char *colocateWithTableName = NULL;
bool viaDeprecatedAPI = false;
CheckCitusVersion(ERROR);
EnsureCoordinator();
relationId = PG_GETARG_OID(0);
distributionColumnText = PG_GETARG_TEXT_P(1);
distributionMethodOid = PG_GETARG_OID(2);
colocateWithTableNameText = PG_GETARG_TEXT_P(3);
/*
* Lock target relation with an exclusive lock - there's no way to make
* sense of this table until we've committed, and we don't want multiple
* backends manipulating this relation.
*/
relation = try_relation_open(relationId, ExclusiveLock);
if (relation == NULL)
{
ereport(ERROR, (errmsg("could not create distributed table: "
"relation does not exist")));
}
/*
* We should do this check here since the codes in the following lines rely
* on this relation to have a supported relation kind. More extensive checks
* will be performed in CreateDistributedTable.
*/
EnsureRelationKindSupported(relationId);
distributionColumnName = text_to_cstring(distributionColumnText);
distributionColumn = BuildDistributionKeyFromColumnName(relation,
distributionColumnName);
distributionMethod = LookupDistributionMethod(distributionMethodOid);
colocateWithTableName = text_to_cstring(colocateWithTableNameText);
CreateDistributedTable(relationId, distributionColumn, distributionMethod,
colocateWithTableName, viaDeprecatedAPI);
relation_close(relation, NoLock);
PG_RETURN_VOID();
}
/*
* CreateReferenceTable creates a distributed table with the given relationId. The
* created table has one shard and replication factor is set to the active worker
* count. In fact, the above is the definition of a reference table in Citus.
*/
Datum
create_reference_table(PG_FUNCTION_ARGS)
{
Oid relationId = PG_GETARG_OID(0);
Relation relation = NULL;
char *colocateWithTableName = NULL;
List *workerNodeList = NIL;
int workerCount = 0;
Var *distributionColumn = NULL;
bool viaDeprecatedAPI = false;
EnsureCoordinator();
CheckCitusVersion(ERROR);
/*
* Ensure schema exists on each worker node. We can not run this function
* transactionally, since we may create shards over separate sessions and
* shard creation depends on the schema being present and visible from all
* sessions.
*/
EnsureSchemaExistsOnAllNodes(relationId);
/*
* Lock target relation with an exclusive lock - there's no way to make
* sense of this table until we've committed, and we don't want multiple
* backends manipulating this relation.
*/
relation = relation_open(relationId, ExclusiveLock);
/*
* We should do this check here since the codes in the following lines rely
* on this relation to have a supported relation kind. More extensive checks
* will be performed in CreateDistributedTable.
*/
EnsureRelationKindSupported(relationId);
workerNodeList = ActivePrimaryNodeList();
workerCount = list_length(workerNodeList);
/* if there are no workers, error out */
if (workerCount == 0)
{
char *relationName = get_rel_name(relationId);
ereport(ERROR, (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("cannot create reference table \"%s\"", relationName),
errdetail("There are no active worker nodes.")));
}
CreateDistributedTable(relationId, distributionColumn, DISTRIBUTE_BY_NONE,
colocateWithTableName, viaDeprecatedAPI);
relation_close(relation, NoLock);
PG_RETURN_VOID();
}
/*
* master_drop_sequences attempts to drop a list of sequences on worker nodes.
* The "IF EXISTS" clause is used to permit dropping sequences even if they may not
* exist. If the commands fail on the workers, the operation is rolled back.
* If ddl propagation (citus.enable_ddl_propagation) is set to off, then the function
* returns without doing anything.
*/
Datum
master_drop_sequences(PG_FUNCTION_ARGS)
{
ArrayType *sequenceNamesArray = PG_GETARG_ARRAYTYPE_P(0);
ArrayIterator sequenceIterator = NULL;
Datum sequenceText = 0;
bool isNull = false;
StringInfo dropSeqCommand = makeStringInfo();
bool coordinator = IsCoordinator();
CheckCitusVersion(ERROR);
/* do nothing if DDL propagation is switched off or this is not the coordinator */
if (!EnableDDLPropagation || !coordinator)
{
PG_RETURN_VOID();
}
/* iterate over sequence names to build single command to DROP them all */
sequenceIterator = array_create_iterator(sequenceNamesArray, 0, NULL);
while (array_iterate(sequenceIterator, &sequenceText, &isNull))
{
if (isNull)
{
ereport(ERROR, (errmsg("unexpected NULL sequence name"),
errcode(ERRCODE_INVALID_PARAMETER_VALUE)));
}
/* append command portion if we haven't added any sequence names yet */
if (dropSeqCommand->len == 0)
{
appendStringInfoString(dropSeqCommand, "DROP SEQUENCE IF EXISTS");
}
else
{
/* otherwise, add a comma to separate subsequent sequence names */
appendStringInfoChar(dropSeqCommand, ',');
}
appendStringInfo(dropSeqCommand, " %s", TextDatumGetCString(sequenceText));
}
if (dropSeqCommand->len != 0)
{
appendStringInfoString(dropSeqCommand, " CASCADE");
SendCommandToWorkers(WORKERS_WITH_METADATA, DISABLE_DDL_PROPAGATION);
SendCommandToWorkers(WORKERS_WITH_METADATA, dropSeqCommand->data);
}
PG_RETURN_VOID();
}
/*
* get_current_transaction_id returns a tuple with (databaseId, processId,
* initiatorNodeIdentifier, transactionNumber, timestamp) that exists in the
* shared memory associated with this backend. Note that if the backend
* is not in a transaction, the function returns uninitialized data where
* transactionNumber equals to 0.
*/
Datum
get_current_transaction_id(PG_FUNCTION_ARGS)
{
TupleDesc tupleDescriptor = NULL;
HeapTuple heapTuple = NULL;
Datum values[5];
bool isNulls[5];
DistributedTransactionId *distributedTransctionId = NULL;
CheckCitusVersion(ERROR);
/* build a tuple descriptor for our result type */
if (get_call_result_type(fcinfo, NULL, &tupleDescriptor) != TYPEFUNC_COMPOSITE)
{
elog(ERROR, "return type must be a row type");
}
/* MyBackendData should always be avaliable, just out of paranoia */
if (!MyBackendData)
{
ereport(ERROR, (errmsg("backend is not ready for distributed transactions")));
}
distributedTransctionId = GetCurrentDistributedTransactionId();
memset(values, 0, sizeof(values));
memset(isNulls, false, sizeof(isNulls));
/* first two fields do not change for this backend, so get directly */
values[0] = ObjectIdGetDatum(MyDatabaseId);
values[1] = Int32GetDatum(MyProcPid);
values[2] = Int32GetDatum(distributedTransctionId->initiatorNodeIdentifier);
values[3] = UInt64GetDatum(distributedTransctionId->transactionNumber);
/* provide a better output */
if (distributedTransctionId->initiatorNodeIdentifier != 0)
{
values[4] = TimestampTzGetDatum(distributedTransctionId->timestamp);
}
else
{
isNulls[4] = true;
}
heapTuple = heap_form_tuple(tupleDescriptor, values, isNulls);
PG_RETURN_DATUM(HeapTupleGetDatum(heapTuple));
}
/*
* master_get_new_placementid is a user facing wrapper function around
* GetNextPlacementId() which allocates and returns a unique placement id for the
* placement to be created.
*
* NB: This can be called by any user; for now we have decided that that's
* ok. We might want to restrict this to users part of a specific role or such
* at some later point.
*/
Datum
master_get_new_placementid(PG_FUNCTION_ARGS)
{
uint64 placementId = 0;
Datum placementIdDatum = 0;
EnsureCoordinator();
CheckCitusVersion(ERROR);
placementId = GetNextPlacementId();
placementIdDatum = Int64GetDatum(placementId);
PG_RETURN_DATUM(placementIdDatum);
}
/*
* master_get_new_shardid is a user facing wrapper function around GetNextShardId()
* which allocates and returns a unique shardId for the shard to be created.
*
* NB: This can be called by any user; for now we have decided that that's
* ok. We might want to restrict this to users part of a specific role or such
* at some later point.
*/
Datum
master_get_new_shardid(PG_FUNCTION_ARGS)
{
uint64 shardId = 0;
Datum shardIdDatum = 0;
EnsureCoordinator();
CheckCitusVersion(ERROR);
shardId = GetNextShardId();
shardIdDatum = Int64GetDatum(shardId);
PG_RETURN_DATUM(shardIdDatum);
}
/*
* citus_total_relation_size accepts a table name and returns a distributed table
* and its indexes' total relation size.
*/
Datum
citus_total_relation_size(PG_FUNCTION_ARGS)
{
Oid relationId = PG_GETARG_OID(0);
uint64 totalRelationSize = 0;
char *tableSizeFunction = PG_TOTAL_RELATION_SIZE_FUNCTION;
CheckCitusVersion(ERROR);
if (CStoreTable(relationId))
{
tableSizeFunction = CSTORE_TABLE_SIZE_FUNCTION;
}
totalRelationSize = DistributedTableSize(relationId, tableSizeFunction);
PG_RETURN_INT64(totalRelationSize);
}
/*
* master_create_worker_shards is a user facing function to create worker shards
* for the given relation in round robin order.
*/
Datum
master_create_worker_shards(PG_FUNCTION_ARGS)
{
text *tableNameText = PG_GETARG_TEXT_P(0);
int32 shardCount = PG_GETARG_INT32(1);
int32 replicationFactor = PG_GETARG_INT32(2);
Oid distributedTableId = ResolveRelationId(tableNameText);
/* do not add any data */
bool useExclusiveConnections = false;
EnsureCoordinator();
CheckCitusVersion(ERROR);
CreateShardsWithRoundRobinPolicy(distributedTableId, shardCount, replicationFactor,
useExclusiveConnections);
PG_RETURN_VOID();
}
/*
* master_drop_all_shards attempts to drop all shards for a given relation.
* Unlike master_apply_delete_command, this function can be called even
* if the table has already been dropped.
*/
Datum
master_drop_all_shards(PG_FUNCTION_ARGS)
{
Oid relationId = PG_GETARG_OID(0);
text *schemaNameText = PG_GETARG_TEXT_P(1);
text *relationNameText = PG_GETARG_TEXT_P(2);
List *shardIntervalList = NIL;
int droppedShardCount = 0;
char *schemaName = text_to_cstring(schemaNameText);
char *relationName = text_to_cstring(relationNameText);
CheckCitusVersion(ERROR);
/*
* The SQL_DROP trigger calls this function even for tables that are
* not distributed. In that case, silently ignore and return -1.
*/
if (!IsDistributedTable(relationId) || !EnableDDLPropagation)
{
PG_RETURN_INT32(-1);
}
EnsureCoordinator();
CheckTableSchemaNameForDrop(relationId, &schemaName, &relationName);
/*
* master_drop_all_shards is typically called from the DROP TABLE trigger,
* but could be called by a user directly. Make sure we have an
* AccessExlusiveLock to prevent any other commands from running on this table
* concurrently.
*/
LockRelationOid(relationId, AccessExclusiveLock);
shardIntervalList = LoadShardIntervalList(relationId);
droppedShardCount = DropShards(relationId, schemaName, relationName,
shardIntervalList);
PG_RETURN_INT32(droppedShardCount);
}
/*
* assign_distributed_transaction_id updates the shared memory allocated for this backend
* and sets initiatorNodeIdentifier, transactionNumber, timestamp fields with the given
* inputs. Also, the function sets the database id and process id via the information that
* Postgres provides.
*
* This function is only intended for internal use for managing distributed transactions.
* Users should not use this function for any purpose.
*/
Datum
assign_distributed_transaction_id(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
/* MyBackendData should always be avaliable, just out of paranoia */
if (!MyBackendData)
{
ereport(ERROR, (errmsg("backend is not ready for distributed transactions")));
}
/*
* Note that we don't need to lock shared memory (i.e., LockBackendSharedMemory()) here
* since this function is executed after AssignDistributedTransactionId() issued on the
* initiator node, which already takes the required lock to enforce the consistency.
*/
SpinLockAcquire(&MyBackendData->mutex);
/* if an id is already assigned, release the lock and error */
if (MyBackendData->transactionId.transactionNumber != 0)
{
SpinLockRelease(&MyBackendData->mutex);
ereport(ERROR, (errmsg("the backend has already been assigned a "
"transaction id")));
}
MyBackendData->databaseId = MyDatabaseId;
MyBackendData->transactionId.initiatorNodeIdentifier = PG_GETARG_INT32(0);
MyBackendData->transactionId.transactionNumber = PG_GETARG_INT64(1);
MyBackendData->transactionId.timestamp = PG_GETARG_TIMESTAMPTZ(2);
MyBackendData->transactionId.transactionOriginator = false;
SpinLockRelease(&MyBackendData->mutex);
PG_RETURN_VOID();
}
/*
* master_apply_delete_command takes in a delete command, finds shards that
* match the criteria defined in the delete command, drops the found shards from
* the worker nodes, and updates the corresponding metadata on the master node.
* This function drops a shard if and only if all rows in the shard satisfy
* the conditions in the delete command. Note that this function only accepts
* conditions on the partition key and if no condition is provided then all
* shards are deleted.
*
* We mark shard placements that we couldn't drop as to be deleted later. If a
* shard satisfies the given conditions, we delete it from shard metadata table
* even though related shard placements are not deleted.
*/
Datum
master_apply_delete_command(PG_FUNCTION_ARGS)
{
text *queryText = PG_GETARG_TEXT_P(0);
char *queryString = text_to_cstring(queryText);
char *relationName = NULL;
char *schemaName = NULL;
Oid relationId = InvalidOid;
List *shardIntervalList = NIL;
List *deletableShardIntervalList = NIL;
List *queryTreeList = NIL;
Query *deleteQuery = NULL;
Node *whereClause = NULL;
Node *deleteCriteria = NULL;
Node *queryTreeNode = NULL;
DeleteStmt *deleteStatement = NULL;
int droppedShardCount = 0;
LOCKMODE lockMode = 0;
char partitionMethod = 0;
bool failOK = false;
#if (PG_VERSION_NUM >= 100000)
RawStmt *rawStmt = (RawStmt *) ParseTreeRawStmt(queryString);
queryTreeNode = rawStmt->stmt;
#else
queryTreeNode = ParseTreeNode(queryString);
#endif
EnsureCoordinator();
CheckCitusVersion(ERROR);
if (!IsA(queryTreeNode, DeleteStmt))
{
ereport(ERROR, (errmsg("query \"%s\" is not a delete statement",
queryString)));
}
deleteStatement = (DeleteStmt *) queryTreeNode;
schemaName = deleteStatement->relation->schemaname;
relationName = deleteStatement->relation->relname;
/*
* We take an exclusive lock while dropping shards to prevent concurrent
* writes. We don't want to block SELECTs, which means queries might fail
* if they access a shard that has just been dropped.
*/
lockMode = ExclusiveLock;
relationId = RangeVarGetRelid(deleteStatement->relation, lockMode, failOK);
/* schema-prefix if it is not specified already */
if (schemaName == NULL)
{
Oid schemaId = get_rel_namespace(relationId);
schemaName = get_namespace_name(schemaId);
}
CheckDistributedTable(relationId);
EnsureTablePermissions(relationId, ACL_DELETE);
#if (PG_VERSION_NUM >= 100000)
queryTreeList = pg_analyze_and_rewrite(rawStmt, queryString, NULL, 0, NULL);
#else
queryTreeList = pg_analyze_and_rewrite(queryTreeNode, queryString, NULL, 0);
#endif
deleteQuery = (Query *) linitial(queryTreeList);
CheckTableCount(deleteQuery);
/* get where clause and flatten it */
whereClause = (Node *) deleteQuery->jointree->quals;
deleteCriteria = eval_const_expressions(NULL, whereClause);
partitionMethod = PartitionMethod(relationId);
if (partitionMethod == DISTRIBUTE_BY_HASH)
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("cannot delete from hash distributed table with this "
"command"),
errdetail("Delete statements on hash-partitioned tables "
"are not supported with master_apply_delete_command."),
errhint("Use master_modify_multiple_shards command instead.")));
}
else if (partitionMethod == DISTRIBUTE_BY_NONE)
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("cannot delete from distributed table"),
errdetail("Delete statements on reference tables "
"are not supported.")));
}
CheckDeleteCriteria(deleteCriteria);
CheckPartitionColumn(relationId, deleteCriteria);
shardIntervalList = LoadShardIntervalList(relationId);
/* drop all shards if where clause is not present */
if (deleteCriteria == NULL)
{
deletableShardIntervalList = shardIntervalList;
ereport(DEBUG2, (errmsg("dropping all shards for \"%s\"", relationName)));
}
else
{
deletableShardIntervalList = ShardsMatchingDeleteCriteria(relationId,
shardIntervalList,
deleteCriteria);
}
droppedShardCount = DropShards(relationId, schemaName, relationName,
deletableShardIntervalList);
PG_RETURN_INT32(droppedShardCount);
}
/*
* get_all_active_transactions returns all the avaliable information about all
* the active backends.
*/
Datum
get_all_active_transactions(PG_FUNCTION_ARGS)
{
ReturnSetInfo *returnSetInfo = (ReturnSetInfo *) fcinfo->resultinfo;
TupleDesc tupleDescriptor = NULL;
Tuplestorestate *tupleStore = NULL;
MemoryContext perQueryContext = NULL;
MemoryContext oldContext = NULL;
int backendIndex = 0;
Datum values[5];
bool isNulls[5];
CheckCitusVersion(ERROR);
/* check to see if caller supports us returning a tuplestore */
if (returnSetInfo == NULL || !IsA(returnSetInfo, ReturnSetInfo))
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("set-valued function called in context " \
"that cannot accept a set")));
}
if (!(returnSetInfo->allowedModes & SFRM_Materialize))
{
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("materialize mode required, but it is not " \
"allowed in this context")));
}
/* build a tuple descriptor for our result type */
if (get_call_result_type(fcinfo, NULL, &tupleDescriptor) != TYPEFUNC_COMPOSITE)
{
elog(ERROR, "return type must be a row type");
}
perQueryContext = returnSetInfo->econtext->ecxt_per_query_memory;
oldContext = MemoryContextSwitchTo(perQueryContext);
tupleStore = tuplestore_begin_heap(true, false, work_mem);
returnSetInfo->returnMode = SFRM_Materialize;
returnSetInfo->setResult = tupleStore;
returnSetInfo->setDesc = tupleDescriptor;
MemoryContextSwitchTo(oldContext);
/*
* We don't want to initialize memory while spinlock is held so we
* prefer to do it here. This initialization is done only for the first
* row.
*/
memset(values, 0, sizeof(values));
memset(isNulls, false, sizeof(isNulls));
/* we're reading all distributed transactions, prevent new backends */
LockBackendSharedMemory(LW_SHARED);
for (backendIndex = 0; backendIndex < MaxBackends; ++backendIndex)
{
BackendData *currentBackend =
&backendManagementShmemData->backends[backendIndex];
SpinLockAcquire(¤tBackend->mutex);
/* we're only interested in active backends */
if (currentBackend->transactionId.transactionNumber == 0)
{
SpinLockRelease(¤tBackend->mutex);
continue;
}
values[0] = ObjectIdGetDatum(currentBackend->databaseId);
values[1] = Int32GetDatum(ProcGlobal->allProcs[backendIndex].pid);
values[2] = Int32GetDatum(currentBackend->transactionId.initiatorNodeIdentifier);
values[3] = UInt64GetDatum(currentBackend->transactionId.transactionNumber);
values[4] = TimestampTzGetDatum(currentBackend->transactionId.timestamp);
SpinLockRelease(¤tBackend->mutex);
tuplestore_putvalues(tupleStore, tupleDescriptor, values, isNulls);
/*
* We don't want to initialize memory while spinlock is held so we
* prefer to do it here. This initialization is done for the rows
* starting from the second one.
*/
memset(values, 0, sizeof(values));
memset(isNulls, false, sizeof(isNulls));
}
UnlockBackendSharedMemory();
/* clean up and return the tuplestore */
tuplestore_donestoring(tupleStore);
PG_RETURN_VOID();
}
/*
* master_get_table_metadata takes in a relation name, and returns partition
* related metadata for the relation. These metadata are grouped and returned in
* a tuple, and are used by the caller when creating new shards. The function
* errors if given relation does not exist, or is not partitioned.
*/
Datum
master_get_table_metadata(PG_FUNCTION_ARGS)
{
text *relationName = PG_GETARG_TEXT_P(0);
Oid relationId = ResolveRelationId(relationName);
DistTableCacheEntry *partitionEntry = NULL;
char *partitionKeyString = NULL;
TypeFuncClass resultTypeClass = 0;
Datum partitionKeyExpr = 0;
Datum partitionKey = 0;
Datum metadataDatum = 0;
HeapTuple metadataTuple = NULL;
TupleDesc metadataDescriptor = NULL;
uint64 shardMaxSizeInBytes = 0;
char shardStorageType = 0;
Datum values[TABLE_METADATA_FIELDS];
bool isNulls[TABLE_METADATA_FIELDS];
CheckCitusVersion(ERROR);
/* find partition tuple for partitioned relation */
partitionEntry = DistributedTableCacheEntry(relationId);
/* create tuple descriptor for return value */
resultTypeClass = get_call_result_type(fcinfo, NULL, &metadataDescriptor);
if (resultTypeClass != TYPEFUNC_COMPOSITE)
{
ereport(ERROR, (errmsg("return type must be a row type")));
}
/* form heap tuple for table metadata */
memset(values, 0, sizeof(values));
memset(isNulls, false, sizeof(isNulls));
partitionKeyString = partitionEntry->partitionKeyString;
/* reference tables do not have partition key */
if (partitionKeyString == NULL)
{
partitionKey = PointerGetDatum(NULL);
isNulls[3] = true;
}
else
{
/* get decompiled expression tree for partition key */
partitionKeyExpr =
PointerGetDatum(cstring_to_text(partitionEntry->partitionKeyString));
partitionKey = DirectFunctionCall2(pg_get_expr, partitionKeyExpr,
ObjectIdGetDatum(relationId));
}
shardMaxSizeInBytes = (int64) ShardMaxSize * 1024L;
/* get storage type */
shardStorageType = ShardStorageType(relationId);
values[0] = ObjectIdGetDatum(relationId);
values[1] = shardStorageType;
values[2] = partitionEntry->partitionMethod;
values[3] = partitionKey;
values[4] = Int32GetDatum(ShardReplicationFactor);
values[5] = Int64GetDatum(shardMaxSizeInBytes);
values[6] = Int32GetDatum(ShardPlacementPolicy);
metadataTuple = heap_form_tuple(metadataDescriptor, values, isNulls);
metadataDatum = HeapTupleGetDatum(metadataTuple);
PG_RETURN_DATUM(metadataDatum);
}
/*
* master_get_active_worker_nodes returns a set of active worker host names and
* port numbers in deterministic order. Currently we assume that all worker
* nodes in pg_dist_node are active.
*/
Datum
master_get_active_worker_nodes(PG_FUNCTION_ARGS)
{
FuncCallContext *functionContext = NULL;
uint32 workerNodeIndex = 0;
uint32 workerNodeCount = 0;
CheckCitusVersion(ERROR);
if (SRF_IS_FIRSTCALL())
{
MemoryContext oldContext = NULL;
List *workerNodeList = NIL;
uint32 workerNodeCount = 0;
TupleDesc tupleDescriptor = NULL;
bool hasOid = false;
/* create a function context for cross-call persistence */
functionContext = SRF_FIRSTCALL_INIT();
/* switch to memory context appropriate for multiple function calls */
oldContext = MemoryContextSwitchTo(functionContext->multi_call_memory_ctx);
workerNodeList = ActiveReadableNodeList();
workerNodeCount = (uint32) list_length(workerNodeList);
functionContext->user_fctx = workerNodeList;
functionContext->max_calls = workerNodeCount;
/*
* This tuple descriptor must match the output parameters declared for
* the function in pg_proc.
*/
tupleDescriptor = CreateTemplateTupleDesc(WORKER_NODE_FIELDS, hasOid);
TupleDescInitEntry(tupleDescriptor, (AttrNumber) 1, "node_name",
TEXTOID, -1, 0);
TupleDescInitEntry(tupleDescriptor, (AttrNumber) 2, "node_port",
INT8OID, -1, 0);
functionContext->tuple_desc = BlessTupleDesc(tupleDescriptor);
MemoryContextSwitchTo(oldContext);
}
functionContext = SRF_PERCALL_SETUP();
workerNodeIndex = functionContext->call_cntr;
workerNodeCount = functionContext->max_calls;
if (workerNodeIndex < workerNodeCount)
{
List *workerNodeList = functionContext->user_fctx;
WorkerNode *workerNode = list_nth(workerNodeList, workerNodeIndex);
Datum workerNodeDatum = WorkerNodeGetDatum(workerNode,
functionContext->tuple_desc);
SRF_RETURN_NEXT(functionContext, workerNodeDatum);
}
else
{
SRF_RETURN_DONE(functionContext);
}
}
/*
* master_get_table_ddl_events takes in a relation name, and returns the set of
* DDL commands needed to reconstruct the relation. The returned DDL commands
* are similar in flavor to schema definitions that pgdump returns. The function
* errors if given relation does not exist.
*/
Datum
master_get_table_ddl_events(PG_FUNCTION_ARGS)
{
FuncCallContext *functionContext = NULL;
ListCell *tableDDLEventCell = NULL;
CheckCitusVersion(ERROR);
/*
* On the very first call to this function, we first use the given relation
* name to get to the relation. We then recreate the list of DDL statements
* issued for this relation, and save the first statement's position in the
* function context.
*/
if (SRF_IS_FIRSTCALL())
{
text *relationName = PG_GETARG_TEXT_P(0);
Oid relationId = ResolveRelationId(relationName);
bool includeSequenceDefaults = true;
MemoryContext oldContext = NULL;
List *tableDDLEventList = NIL;
/* create a function context for cross-call persistence */
functionContext = SRF_FIRSTCALL_INIT();
/* switch to memory context appropriate for multiple function calls */
oldContext = MemoryContextSwitchTo(functionContext->multi_call_memory_ctx);
/* allocate DDL statements, and then save position in DDL statements */
tableDDLEventList = GetTableDDLEvents(relationId, includeSequenceDefaults);
tableDDLEventCell = list_head(tableDDLEventList);
functionContext->user_fctx = tableDDLEventCell;
MemoryContextSwitchTo(oldContext);
}
/*
* On every call to this function, we get the current position in the
* statement list. We then iterate to the next position in the list and
* return the current statement, if we have not yet reached the end of
* list.
*/
functionContext = SRF_PERCALL_SETUP();
tableDDLEventCell = (ListCell *) functionContext->user_fctx;
if (tableDDLEventCell != NULL)
{
char *ddlStatement = (char *) lfirst(tableDDLEventCell);
text *ddlStatementText = cstring_to_text(ddlStatement);
functionContext->user_fctx = lnext(tableDDLEventCell);
SRF_RETURN_NEXT(functionContext, PointerGetDatum(ddlStatementText));
}
else
{
SRF_RETURN_DONE(functionContext);
}
}
