On 15/07/10 07:04, ankit kamal wrote:
> Hi All
>
>
> I am using a cluster setup with two nodes in it. Replication between
> two nodes is being done through slony.
>
> Postgres version is 8.1.2 and slony version is 1.1.5 .
>
> On Master node an error "CDT FATAL: invalid frontend message type 69"
> encountered at 10:51 and postgres crashed.
Would "CDT" be your timezone ("Central Daylight Time" I think)? It's not
part of any PG error message.
The other part is an error message that gets produced when the database
receives a message (69 = "E" = "Execute") for a protocol version later
than the client has said it's using. In this case, probably PQexecParams
or similar (see the libpq chapter of the manual).
Either some client code thought it was talking to an old version of
PostgreSQL (that only supported protocol 2 rather than 3), there was a
network error (and a message got garbled) or you've hit a bug.
It is a good idea to upgrade to the latest 8.1.x release of PostgreSQL
and also perhaps a more recent slony - both have received bug-fixes
since the versions you are running. Particularly upgrade PostgreSQL -
you're missing almost 20 sets of bug-fixes.
> There were no postgres logs logged from 10:51 to 11:11, and after that,
> error "CDT FATAL: sorry, too many clients already" was logged in
> postgres logs upto postgres shutdown. The problem resolved after the
> postgres was restarted on master node.
>
> I could not find this error defined anywhere.
Well, "too many clients" just you have reached your maximum number of
connections. It would have been useful to check what was connected, and
what each connection was doing. Always worth seeing what "ps" reports
the machine is doing and what the "pg_stat_activity" view says
PostgreSQL is doing.
> Any ideas of what it means, how to track the cause and cure?
>
> Is there any way to reproduce this error?
Not without knowing more about how it happened, which we could only
investigate before you restarted the master.
--
Richard Huxton
Archonet Ltd
