This change fixes a small bug about quoting of workerrack column in
NodeListInsertCommand:
Previous: `"..., '%s'", workerRack`
Now: `"..., %s", quote_literal_cstr(workerRack)`
So far we've reloaded them frequently. Besides avoiding that cost -
noticeable for some workloads with large shard counts - it makes it
easier to add information to ShardPlacements that help us make
placement_connection.c colocation aware.
Doing so requires adding a mapping from shardId to the cache
entries. For that metadata_cache.c now maintains an additional
hashtable. That hashtable only references shard intervals in the
dist table cache.
Previously the function was getting too large. Thus this splits the
function into separate parts for looking up the cache entry and
building the cache contents.
CloseNodeConnections() is supposed to close connections to a given node.
However, before this commit it lacks to actually call PQFinish() on the
connections. Using CloseConnection() handles closing and all other necessary
actions.
With this change we start to error out on router planner queries where a common table
expression with data-modifying statement is present. We already do not support if
there is a data-modifying statement using result of the CTE, now we also error out
if CTE itself is data-modifying statement.
Remove the router specific transaction and shard management, and
replace it with the new placement connection API. This mostly leaves
behaviour alone, except that it is now, inside a transaction, legal to
select from a shard to which no pre-existing connection exists.
To simplify code the code handling task executions for select and
modify has been split into two - the previous coding was starting to
get confusing due to the amount of only conditionally applicable code.
Modification connections & transactions are now always established in
parallel, not just for reference tables.
Currently there are several places in citus that map placements to
connections and that manage placement health. Centralize this
knowledge. Because of the centralized knowledge about which
connection has previously been used for which shard/placement, this
also provides the basis for relaxing restrictions around combining
various forms of DDL/DML.
Connections for a placement can now be acquired using
GetPlacementConnection(). If the connection is used for DML or DDL the
FOR_DDL/DML flags should be used respectively. If an individual
remote transaction fails (but the transaction on the master succeeds)
and FOR_DDL/DML have been specified, the placement is marked as
invalid, unless that'd mark all placements for a shard as invalid.
With this change, we start to replicate all reference tables to the new node when new node
is added to the cluster with master_add_node command. We also update replication factor
of reference table's colocation group.
Since we will now replicate reference tables each time we add node, we need to ensure
that test space is clean in terms of reference tables before any add node operation.
For this purpose we had to change order of multi_drop_extension test which caused
change of some of the colocation ids.
With this change we introduce new UDF, upgrade_to_reference_table, which can be used to
upgrade existing broadcast tables reference tables. For upgrading, we require that given
table contains only one shard.
Renamed FindShardIntervalIndex() to ShardIndex() and added binary search
capability. It used to assume that hash partition tables are always
uniformly distributed which is not true if upcoming tenant isolation
feature is applied. This commit also reduces code duplication.
Router planner already handles cases when all shards
are pruned out. This is about missing test cases. Notice that
"column is null" and "column = null" have different shard
pruning behavior.
We have one replication of reference table for each node. Therefore all problems with
replication factor > 1 also applies to reference table. As a solution we will not allow
foreign keys on reference tables. It is not possible to define foreign key from, to or
between reference tables.
Eren Basak
Andres Freund
Önder Kalacı
Murat Tuncer
Murat Tuncer
Burak Yücesoy
Marco Slot
Marco Slot
Jason Petersen
Metin Döşlü
Metin Doslu