>> I'm not sure about this. I think pgbench will not start transactions
>> until all clients establish connections to PostgreSQL.
> I think that is true if "!is_connect", all client connections are
> performed at the beginning of threadRun, but under -C each client has
> its own connect/deconnect integrated within doCustom, so it is done in
> parallel to other clients having their transactions processed, hence
> the issue with the formula.


I have tested with pgpool-II which is set to accept up to 2
connections, then run pgbench with -C and -c 32. pgbench was blocked
as expected and I attached gdb and got stack trace:

(gdb) bt
#0  0x00007f48d5f17110 in __poll_nocancel ()
    at ../sysdeps/unix/syscall-template.S:81
#1  0x00007f48d6724056 in pqSocketCheck ()
   from /usr/local/src/pgsql/current/lib/libpq.so.5
#2  0x00007f48d6724940 in pqWaitTimed ()
   from /usr/local/src/pgsql/current/lib/libpq.so.5
#3  0x00007f48d671f3e2 in connectDBComplete ()
   from /usr/local/src/pgsql/current/lib/libpq.so.5
#4  0x00007f48d671fbcf in PQconnectdbParams ()
   from /usr/local/src/pgsql/current/lib/libpq.so.5
#5  0x0000000000402b2b in doConnect () at pgbench.c:650
#6  0x0000000000404591 in doCustom (thread=0x25c2f40, st=0x25c2770, 
    conn_time=0x25c2f90, logfile=0x0, agg=0x7fffba224260) at pgbench.c:1353
#7  0x000000000040a1d5 in threadRun (arg=0x25c2f40) at pgbench.c:3581
#8  0x0000000000409ab4 in main (argc=12, argv=0x7fffba224668) at pgbench.c:3455

As you can see, one of threads wants to connect to PostgreSQL
(actually pgpool-II) and waits for reply.

In threadRun() around line 3581:

        /* send start up queries in async manner */
        for (i = 0; i < nstate; i++)
                CState     *st = &state[i];
                Command   **commands = sql_files[st->use_file];
                int                     prev_ecnt = st->ecnt;

                st->use_file = getrand(thread, 0, num_files - 1);
                if (!doCustom(thread, st, &thread->conn_time, logfile, &aggs))
                        remains--;                      /* I've aborted */

                if (st->ecnt > prev_ecnt && commands[st->state]->type == 
                        fprintf(stderr, "client %d aborted in state %d; 
execution of meta-command failed\n",
                                        i, st->state);
                        remains--;                      /* I've aborted */
                        st->con = NULL;

Here doCustome() is called with st->con == NULL. In doCustom() around
line 1353:

        if (st->con == NULL)
                instr_time      start,

                if ((st->con = doConnect()) == NULL)

doConnect() blocks until PostgreSQL (pgpool-II) allows to be

Because outer loop in threadRun() wants to loop over until all threads
succefully connects to PostgreSQL, pgbench is blocked here.

        /* send start up queries in async manner */
        for (i = 0; i < nstate; i++)

>> I'm going to commit your patch if there's no objection.
> This seems fine with me.
> The formula change, and just this one, should probably be backported
> somehow, as this is a bug, wherever pgbench resides in older
> versions. It is just 's/nthreads/nclients/' in the printResult formula
> for computing tps_exclude.

Yeah, that's definitely a bug but I'm afraid the fix will change the
TPS number and may break the backward compatibility. Since we have
lived with bug for years, I hesitate to back port to older stable

Best regards,
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php

Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:

Reply via email to