Tom:
> Looks like the disconnect was because pgbouncer restarted. If that
> wasn't supposed to happen then you should take it up with the
> pgbouncer folk.
The restart of pgbouncer was intentional, although made by someone else,
so the disconnect is ok. What's not ok is the "UPDATE 153" message after
message with connection lost and the fact that the UPDATE was committed
to database without explicit COMMIT. Maybe pgbouncer issued the commit?
Stefan:
> hmm - and you are really sure that the update got commited in the
> end(even if you got the "UPDATE 153" it should have been rollbacked as
> soon as the connection got dropped)?
Quite sure. I've seen it on my colleague's screen (afterwards) and saved
it to text file. Before the "BEGIN; UPDATE" there was "SELECT * FROM
table" which showed the state of table and just after the unsuccessful
"ROLLBACK" the "SELECT * from table" was issued again and showed updated
table. No one else worked with the table and the change is definitely
committed. The .psql_history contains exactly this as well.
Kuba