/*
* worker_apply_shard_ddl_command extends table, index, or constraint names in
* the given DDL command. The function then applies this extended DDL command
* against the database.
*/
Datum
worker_apply_shard_ddl_command(PG_FUNCTION_ARGS)
{
uint64 shardId = PG_GETARG_INT64(0);
text *schemaNameText = PG_GETARG_TEXT_P(1);
text *ddlCommandText = PG_GETARG_TEXT_P(2);
char *schemaName = text_to_cstring(schemaNameText);
const char *ddlCommand = text_to_cstring(ddlCommandText);
Node *ddlCommandNode = ParseTreeNode(ddlCommand);
/* extend names in ddl command and apply extended command */
RelayEventExtendNames(ddlCommandNode, schemaName, shardId);
ProcessUtility(ddlCommandNode, ddlCommand, PROCESS_UTILITY_TOPLEVEL,
NULL, None_Receiver, NULL);
PG_RETURN_VOID();
}
/*
* worker_apply_inter_shard_ddl_command extends table, index, or constraint names in
* the given DDL command. The function then applies this extended DDL command
* against the database.
*/
Datum
worker_apply_inter_shard_ddl_command(PG_FUNCTION_ARGS)
{
uint64 leftShardId = PG_GETARG_INT64(0);
text *leftShardSchemaNameText = PG_GETARG_TEXT_P(1);
uint64 rightShardId = PG_GETARG_INT64(2);
text *rightShardSchemaNameText = PG_GETARG_TEXT_P(3);
text *ddlCommandText = PG_GETARG_TEXT_P(4);
char *leftShardSchemaName = text_to_cstring(leftShardSchemaNameText);
char *rightShardSchemaName = text_to_cstring(rightShardSchemaNameText);
const char *ddlCommand = text_to_cstring(ddlCommandText);
Node *ddlCommandNode = ParseTreeNode(ddlCommand);
/* extend names in ddl command and apply extended command */
RelayEventExtendNamesForInterShardCommands(ddlCommandNode, leftShardId,
leftShardSchemaName, rightShardId,
rightShardSchemaName);
ProcessUtility(ddlCommandNode, ddlCommand, PROCESS_UTILITY_TOPLEVEL, NULL,
None_Receiver, NULL);
PG_RETURN_VOID();
}
/*
* FetchRegularTable fetches the given table's data using the copy out command.
* The function then fetches the DDL commands necessary to create this table's
* replica, and locally applies these DDL commands. Last, the function copies
* the fetched table data into the created table; and on success, returns true.
* On failure due to connectivity issues with remote node, the function returns
* false. On other types of failures, the function errors out.
*/
static bool
FetchRegularTable(const char *nodeName, uint32 nodePort, const char *tableName)
{
StringInfo localFilePath = NULL;
StringInfo remoteCopyCommand = NULL;
List *ddlCommandList = NIL;
ListCell *ddlCommandCell = NULL;
CopyStmt *localCopyCommand = NULL;
RangeVar *localTable = NULL;
uint64 shardId = 0;
bool received = false;
StringInfo queryString = NULL;
const char *tableOwner = NULL;
Oid tableOwnerId = InvalidOid;
Oid savedUserId = InvalidOid;
int savedSecurityContext = 0;
List *tableNameList = NIL;
/* copy remote table's data to this node in an idempotent manner */
shardId = ExtractShardId(tableName);
localFilePath = makeStringInfo();
appendStringInfo(localFilePath, "base/%s/%s" UINT64_FORMAT,
PG_JOB_CACHE_DIR, TABLE_FILE_PREFIX, shardId);
remoteCopyCommand = makeStringInfo();
appendStringInfo(remoteCopyCommand, COPY_OUT_COMMAND, tableName);
received = ReceiveRegularFile(nodeName, nodePort, remoteCopyCommand, localFilePath);
if (!received)
{
return false;
}
/* fetch the ddl commands needed to create the table */
tableOwner = RemoteTableOwner(nodeName, nodePort, tableName);
if (tableOwner == NULL)
{
return false;
}
tableOwnerId = get_role_oid(tableOwner, false);
/* fetch the ddl commands needed to create the table */
ddlCommandList = TableDDLCommandList(nodeName, nodePort, tableName);
if (ddlCommandList == NIL)
{
return false;
}
/*
* Apply DDL commands against the database. Note that on failure from here
* on, we immediately error out instead of returning false. Have to do
* this as the table's owner to ensure the local table is created with
* compatible permissions.
*/
GetUserIdAndSecContext(&savedUserId, &savedSecurityContext);
SetUserIdAndSecContext(tableOwnerId, SECURITY_LOCAL_USERID_CHANGE);
foreach(ddlCommandCell, ddlCommandList)
{
StringInfo ddlCommand = (StringInfo) lfirst(ddlCommandCell);
Node *ddlCommandNode = ParseTreeNode(ddlCommand->data);
ProcessUtility(ddlCommandNode, ddlCommand->data, PROCESS_UTILITY_TOPLEVEL,
NULL, None_Receiver, NULL);
CommandCounterIncrement();
}
/*
* 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);
}