replaying trx and certification failure causes assert in connection end

[Seppo Jaakola]

This issues surfaced when reproducing lp:490759. Replaying transaction can end in certification failure, and it turns out that connection end procedure will then cause an assert.

Here is debug log for the case:

091206 17:18:56 [Note] [Debug] WSREP: mm_galera.c:mm_galera_replay_trx():2325: replaying applier starting
091206 17:18:56 [Note] [Debug] WSREP: mm_galera.c:process_query_write_set():975: remote trx seqno: 25585 25707 last_seen_trx: 25582 25585, cert: 102
091206 17:18:56 [Note] [Debug] WSREP: mm_galera.c:process_query_write_set():994: trx certification failed: (25585 25707) last_seen: 25582
091206 17:18:56 [Note] [Debug] WSREP: galera_job_queue.c:job_queue_remove_worker():116: job queue released, workers now: 1
091206 17:18:56 [Note] [Debug] WSREP: mm_galera.c:mm_galera_replay_trx():2379: replaying over for applier in to grab: 3, seqno: 25585 25707
091206 17:18:56 [Note] [Debug] WSREP: mm_galera.c:mm_galera_pre_commit():1880: local trx certified, seqno: 25586 25708 last_seen_trx: 25582
091206 17:18:56 [Note] [Debug] WSREP: mm_galera.c:mm_galera_pre_commit():1880: local trx certified, seqno: 25587 25709 last_seen_trx: 25582
091206 17:18:56 [Note] WSREP: trx_replay successful for: 2935085968
mysqld: protocol.cc:473: void net_end_statement(THD*): Assertion `0' failed.

Diagnostic area is not set:

(gdb) p thd->main_da
$5 = {is_sent = false, can_overwrite_status = false,
  m_message = "Rows matched: 1 Changed: 1 Warnings: 0\000ry restarting transaction\000", '�' <repeats 445 times>, m_sql_errno = 1213, m_server_status = 3,
  m_affected_rows = 1, m_last_insert_id = 0, m_total_warn_count = 0,
  m_status = Diagnostics_area::DA_EMPTY}

[Seppo Jaakola (seppo-jaakola)]

Re-opened this case, problem manifested with lengthy test run.
A fix was committed in galera library and testing continues.