Re: [HACKERS] psql casts aspersions on server reliability - Mailing list pgsql-hackers

From Bruce Momjian
Subject Re: [HACKERS] psql casts aspersions on server reliability
Date
Msg-id ZV957kbJfJ4BiTKU@momjian.us
Whole thread Raw
In response to Re: [HACKERS] psql casts aspersions on server reliability  (Bruce Momjian <bruce@momjian.us>)
Responses Re: [HACKERS] psql casts aspersions on server reliability
List pgsql-hackers
On Wed, Nov 22, 2023 at 10:25:14PM -0500, Bruce Momjian wrote:
> On Wed, Nov 22, 2023 at 07:38:52PM -0500, Tom Lane wrote:
> > Bruce Momjian <bruce@momjian.us> writes:
> > > On Wed, Sep 28, 2016 at 09:14:41AM -0400, Tom Lane wrote:
> > >> I could go along with just dropping the last sentence ("This probably...")
> > >> if the last error we got was FATAL level.  I don't find "unexpectedly"
> > >> to be problematic here: from the point of view of psql, and probably
> > >> of its user, the shutdown *was* unexpected.
> > 
> > > I looked at this thread from 2016 and I think the problem is the
> > > "abnormally" word, since if the server was shutdown by the administrator
> > > (most likely), it isn't abnormal.  Here is a patch to remove
> > > "abnormally".
> > 
> > I do not think this is an improvement.  The places you are changing
> > are reacting to a connection closure.  *If* we had previously gotten a
> > "FATAL: terminating connection due to administrator command" message,
> > then yeah the connection closure is expected; but if not, it isn't.
> > Your patch adds no check for that.  (As I remarked in 2016, we could
> > probably condition this on the elevel being FATAL, rather than
> > checking for specific error messages.)
> 
> Yes, you are correct.  Here is a patch that implements the FATAL test,
> though I am not sure I have the logic correct or backwards, and I don't
> know how to test this.  Thanks.

I developed the attached patch which seems to work better.  In testing
kill -3 on a backend or calling elog(FATAL) in the server for a
session, libpq's 'res' is NULL, meaning we don't have any status to
check for PGRES_FATAL_ERROR.  It is very possible that libpq just isn't
stuctured to have the PGRES_FATAL_ERROR at the point where we issue this
message, and this is not worth improving.

    test=> select pg_sleep(100);
-->    FATAL:  FATAL called
    
    server closed the connection unexpectedly
-->            This probably means the server terminated null
            before or while processing the request.
    The connection to the server was lost. Attempting reset: Succeeded.

-- 
  Bruce Momjian  <bruce@momjian.us>        https://momjian.us
  EDB                                      https://enterprisedb.com

  Only you can decide what is important to you.

Attachment

pgsql-hackers by date:

Previous
From: "Drouvot, Bertrand"
Date:
Subject: Re: Synchronizing slots from primary to standby
Next
From: Alvaro Herrera
Date:
Subject: Re: Catalog domain not-null constraints