Thanks for updating the patch!
On Fri, Jul 10, 2026 at 7:00 PM Daniel Gustafsson <[email protected]> wrote:
> > Shouldn't "the process which operate" be either "the process which operates"
> > or "the processes which operate"? Since multiple data checksums workers
> > can run, the latter seems more appropriate.
>
> Currently the processing is limited to a single worker, so I think the
> proposed
> wording is more appropriate.
So "operate" of "the process which operate" should be "operates"
(i.e., "s" is necessary for "operate")?
> > While looking at the related code, I also found that
> > B_DATACHECKSUMSWORKER_LAUNCHER and B_DATACHECKSUMSWORKER_WORKER
> > were added as BackendType entries in miscadmin.h. Since those entries appear
> > under the comment for auxiliary processes, it could give the impression that
> > the data checksums launcher and workers are auxiliary processes. To avoid
> > that confusion, would it make sense to add a comment such as:
> >
> > -------------------------
> > B_WAL_WRITER,
> >
> > + /*
> > + * XXXXXXXXXX
> > + */
> > B_DATACHECKSUMSWORKER_LAUNCHER,
> > B_DATACHECKSUMSWORKER_WORKER,
> > -------------------------
>
> Fair point, see attached.
+ /* Background worker processes */
Thanks for adding the comment! I'm a bit concerned that it's misleading,
since enum BackendType already contains background worker entries
such as B_BG_WORKER and B_SLOTSYNC_WORKER. So, how about making it
more specific? For example:
------------
Data checksums processes are dynamic background workers, but they
use dedicated backend types for pgstat I/O accounting.
------------
> > BTW, I'm also wondering why dedicated BackendType entries were added for
> > the data checksums processes, while other background workers don't have
> > their own BackendType values.
>
> They are need for the pgstat system though aren't they, or am I missing
> something?
As I understand it, dedicated backend types are not inherently required
by the pgstat infrastructure. They are needed here so that pg_stat_io
can account for I/O performed by the data checksums launcher and
workers separately from generic background workers. Otherwise,
their I/O would simply be counted under the background worker backend
type, making it impossible to distinguish their I/O statistics.
Is this my understanding right??
Regards,
--
Fujii Masao
