On Thu, Dec 15, 2022 at 08:35:53PM -0800, Jeff Davis wrote:
> But why make CLUSTER more like VACUUM? Shouldn't we make
> VACUUM/CLUSTER/ANALYZE more like INSERT/SELECT/REINDEX?
Hm. Since VACUUM may happen across multiple transactions, it is careful to
re-check the privileges for each relation. For example, vacuum_rel()
contains this comment:
/*
* Check if relation needs to be skipped based on privileges. This
check
* happens also when building the relation list to vacuum for a manual
* operation, and needs to be done additionally here as VACUUM could
* happen across multiple transactions where privileges could have
changed
* in-between. Make sure to only generate logs for VACUUM in this case.
*/
I do wonder whether this is something we really need to be concerned about.
In the worst case, you might be able to VACUUM a table with a concurrent
owner change.
The logic for gathering all relations to process (i.e.,
get_all_vacuum_rels() and get_tables_to_cluster()) would also need to be
adjusted to handle partitioned tables. Right now, we gather all the
partitions and checks privileges on each. I think this would be easy to
change.
--
Nathan Bossart
Amazon Web Services: https://aws.amazon.com
