Обсуждение: Connections refused during backups

After an upgrade from 8.0.8 to 8.1.8, I am getting a number of our
scripts failing with 'connection refused' at around 4:00am, which seems
to be when the nightly database backup occurs.

I am using "/usr/bin/pg_dump -Fc -b -c -C" to do the deed.

Has there been a change in behavior, i.e. is the whole databases being
exclusively locked or something that would prevent a connection.

Glen Eustace <geustace@godzone.net.nz> writes:
> After an upgrade from 8.0.8 to 8.1.8, I am getting a number of our
> scripts failing with 'connection refused' at around 4:00am, which seems
> to be when the nightly database backup occurs.

"Connection refused" is a kernel-level rejection, and at least in theory
should have nothing to do with what Postgres is doing.  Did you do a
kernel update at the same time?

            regards, tom lane

Tom Lane wrote:

"Connection refused" is a kernel-level rejection, and at least in theory
should have nothing to do with what Postgres is doing.  Did you do a
kernel update at the same time?

            regards, tom lane


My apologies, I shouldn't have used the generic socket error 'Connection
refused'. What I am getting are errors like

"Could not connect to database" from libnss-pgsql

Glen Eustace <geustace@godzone.net.nz> writes:
> My apologies, I shouldn't have used the generic socket error 'Connection
> refused'. What I am getting are errors like
> "Could not connect to database" from libnss-pgsql

Please be *exact* rather than handwaving.  What is the full text of the
client-side message?  What if anything shows up in the postmaster log?

            regards, tom lane

Tom Lane wrote:
> Glen Eustace <geustace@godzone.net.nz> writes:
>> My apologies, I shouldn't have used the generic socket error 'Connection
>> refused'. What I am getting are errors like
>> "Could not connect to database" from libnss-pgsql
>
> Please be *exact* rather than handwaving.  What is the full text of the
> client-side message?  What if anything shows up in the postmaster log?
>
>             regards, tom lane

This is the exact message from libnss-pgsql, there is currently no
timestamp in the log ( which I will fix ), so I cannot place any
messages into the correct timeframe. I'll wait until tomorrow morning
and see if anything is obvious. The only thing I am aware of is that
nss-pgsql uses a 10s connection_timeout, which may be too short when the
pgsql server is busy.

Glen Eustace <geustace@godzone.net.nz> writes:
>>> My apologies, I shouldn't have used the generic socket error 'Connection
>>> refused'. What I am getting are errors like
>>> "Could not connect to database" from libnss-pgsql

> This is the exact message from libnss-pgsql,

While you are at it, you should complain to the libnss-pgsql authors
about the inadequacy of the error message.  It should at least be
repeating the libpq error message ...

            regards, tom lane

Tom Lane wrote:
> Glen Eustace <geustace@godzone.net.nz> writes:
>>>> My apologies, I shouldn't have used the generic socket error 'Connection
>>>> refused'. What I am getting are errors like
>>>> "Could not connect to database" from libnss-pgsql
>
>> This is the exact message from libnss-pgsql,
>
> While you are at it, you should complain to the libnss-pgsql authors
> about the inadequacy of the error message.  It should at least be
> repeating the libpq error message ...
>
>             regards, tom lane

Nothing of interest appears in the postmaster logs so I am broadening my
search for general network incidents/activity as opposed to postgresql
specific.

I have also downloaded the source for libnss-pgsql and will add some
extra code to see if I can get any more useful info from it.

Thanks for the comments/help to date :-)