[Pgcluster-general] pgcluster does not work in degraded mode

[At.Mitani]

Hi Filip,

> First I am trying to test pgcluster in  "degraded" mode:
> 3) fire up pgbench -i (db population mode)
>     $ pgbench -h lb_3 -p 5433 -U pgcluster -i -s100 pgbench100
> 4) while pgbench is running, I put cluster_1 down with
>     cluster_1 $ pg_ctl stop -m fast
> 
> pgbench fails ( I take it as normal behaviour ) with:
> ...
> 260000 tuples done.
> 270000 tuples done.
> server closed the connection unexpectedly
>         This probably means the server terminated abnormally
>         before or while processing the request.
> PQendcopy failed

This is a specification of pgbench.
pgbench uses one session during create initial table data.
Therefore, initial transaction rollbacked when first session was disconnected.

> But, this is not the case. First probe works:
> $ psql  -h lb_3 -p 5433 -U pgcluster postgres -qAtc 'select 1'
> 1
> but next  probe fails:
> $ psql  -h lb_3 -p 5433 -U pgcluster postgres -qAtc 'select 1'
> psql: expected authentication request from server, but received S
> this is not acceptable, at least R/O traffic should work...

Yha... I also got same situation with two sets of cluster DBs.
But it not occured in three sets of cluster DBs.
It's a terrible bug!

I'm going to create patch tonight.

Thank you for report us!

Best Regards,
-- 
At.Mitani <mitani at sraw.co.jp>