Re: Conflicting updates of command progress

Sami Imseih <samimseih(at)gmail(dot)com> wrote:

> > pgstat_progress_start_command() is called twice: First with
> > cmdtype=PROGRESS_COMMAND_CLUSTER, second with
> > PROGRESS_COMMAND_CREATE_INDEX. The first happens in cluster_rel() the second
> > in cluster_rel() -> rebuild_relation() -> finish_heap_swap() ->
> > reindex_relation() -> reindex_index().
> >
> > It does not matter though, the current design only expects one command.
>
> When I set a breakpoint on pgstat_progress_start_command and ran a
> CLUSTER on a table with a few indexes, I only saw start_command being
> called once for PROGRESS_COMMAND_CLUSTER. Then I went back in the
> code and realized that reindex_index when called via the CLUSTER command
> has progress set to false. So as it stands now, only PROGRESS_COMMAND_CLUSTER
> is effectively started when CLUSTER is called.
>
> ```
> if (progress)
> {
> const int progress_cols[] = {
> PROGRESS_CREATEIDX_COMMAND,
> PROGRESS_CREATEIDX_INDEX_OID
> };
> const int64 progress_vals[] = {
> PROGRESS_CREATEIDX_COMMAND_REINDEX,
> indexId
> };
>
> pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
> heapId);
> pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
> }

Ah, got it. So for now the REPACK CONCURRENTLY patch only needs to turn off
the progress reporting for PROGRESS_COMMAND_CREATE_INDEX properly. Thanks for
checking!

--
Antonin Houska
Web: https://www.cybertec-postgresql.com