[Pgpool-general] SQL sessions die when a node fails

[J. Carlos Muro]

Hi! If I shutdown a postgresql node while I am in a sql session, it will
stop working. Is that normal?

I have the next scenario:
  vapp1a: server with pgpool, listening port 5431
  vpgsql1a: server with postgresql, backend 0
  vpgsql1b: server with postgresql, backend 1

Setups for pgpool:
    replication_mode = true
    load_balance_mode = true
    replicate_select = true

Pgpool start logs look this way:
  Jun 25 18:31:49 vapp1a pgpool: 2009-06-25 18:31:41 DEBUG: pid 30820:
num_backends: 2 num_backends: 2 total_weight: 2.000000
  Jun 25 18:31:49 vapp1a pgpool: 2009-06-25 18:31:41 DEBUG: pid 30820:
backend 0 weight: 1073741823.500000
  Jun 25 18:31:49 vapp1a pgpool: 2009-06-25 18:31:41 DEBUG: pid 30820:
backend 1 weight: 1073741823.500000
  [....]
  Jun 25 18:31:51 app1a pgpool: 2009-06-25 18:31:42 LOG:   pid 30820: pgpool
successfully started
  Jun 25 18:31:51 app1a pgpool: 2009-06-25 18:31:42 DEBUG: pid 30861: I am
PCP 30861

I log-in to the database through  pgpool and execute some inserts:
  pgpool at vapp1:~$ psql -h localhost -p 5431 -U auser mybase
  mybase=> create table test (id integer);
  mybase=> insert into test (1);

I see that row on both vpgsql1a and vpgsql1b. I won't exit the psql session
yet...

Now I stop backend 1 and get the next pgpool logs:
  Jun 25 18:36:58 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30860:
detect_error: kind: E
  Jun 25 18:36:58 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30860:
detect_stop_postmaster_error: receive admin shutdown error from a node.
  Jun 25 18:36:58 vapp1a pgpool: 2009-06-25 18:36:58 LOG:   pid 30860:
notice_backend_error: 1 fail over request from pid 30860
  Jun 25 18:36:58 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30820:
failover_handler called
  Jun 25 18:36:58 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30820:
failover_handler: starting to select new master node
  Jun 25 18:36:59 vapp1a pgpool: 2009-06-25 18:36:58 LOG:   pid 30820:
starting degeneration. shutdown host vpgsql1b(5432)
  Jun 25 18:36:59 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30820:
failover_handler: kill 30829
  [....]
  Jun 25 18:37:00 vapp1a pgpool: 2009-06-25 18:36:58 LOG:   pid 30820:
failover_handler: set new master node: 0
  Jun 25 18:37:00 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30852: child
receives shutdown request signal 3
  [....]
  Jun 25 18:37:04 vapp1a pgpool: 2009-06-25 18:36:58 LOG:   pid 30820:
failover done. shutdown host vpgsql1b(5432)
  Jun 25 18:37:04 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30820:
reap_handler called
  Jun 25 18:37:04 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30820:
reap_handler: call wait3
  Jun 25 18:37:04 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30820: child
30829 exits with status 0 by signal 0
  [....]
  Jun 25 18:37:06 vapp1a pgpool: 2009-06-25 18:36:58 DEBUG: pid 30820:
reap_handler: normally exited

In the psql session I try to insert a new row, getting:
  mybase=> insert into kk values (2);
  server closed the connection unexpectedly
          This probably means the server terminated abnormally
          before or while processing the request.
  The connection to the server was lost. Attempting reset: Succeeded.

Now I exit psql, and log-in again, now successfully inserting a new row:
  pgpool at vapp1:~$ psql -h localhost -p 5431 -U auser mybase
  mybase=> create table test (id integer);
  mybase=> insert into test (2);
  INSERT 0 1

Is that normal?
Thanks in advance,
J. Carlos Muro
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://pgfoundry.org/pipermail/pgpool-general/attachments/20090625/05170ae6/attachment.html>