alter_distributed_table seems to OOM with many tables

[onderkalaci]

CREATE TABLE  time_series_events (event_time timestamp, event int, user_id int)
PARTITION BY RANGE (event_time);

SELECT create_time_partitions(table_name:= 'time_series_events',
    partition_interval:= '1 day',
    end_at:= '2025-10-30',
    start_from:= '2021-10-10');

SELECT create_distributed_table('time_series_events', 'user_id');

SELECT alter_distributed_table('time_series_events', shard_count:=512, cascade_to_colocated:=yes);

After some time, this gives OOM

[onderkalaci]

One potential problem could be that we keep

def ConvertTable:
 
    foreach_oid(partitionRelationId, partitionList)
        attachPartitionCommands = lappend(attachPartitionCommands, attachPartitionCommand);

and then keep attachPartitionCommands until all the partitions are converted.

This is NOT the only problem, but one problem. We should aggressively pfree (or better MemoryContextCreate/Delete) per table conversion

[hanefi]

For mitigation, we considered the following list of steps:

1. Detach all the partitions.
2. Change shard counts for the detached partitions
3. Change shard count for the partitioned table
4. Attach all the partitions.

Edit: I got a transient error that vanished on a clean Citus build. I used to have an error message that is not possible to reproduce, so I deleted it.