On Fri, Feb 14, 2014 at 12:48 AM, john gale <john@smadness.com> wrote:
> Does this sound related to anything that might have been attempted to be
> fixed in 9.3.2 ? Does the sample backtrace look like it could be a Postgres
> issue, or could it be a Mac OS X / unix network issue ?
It sounds like a network issue to me. Is there a stateful firewall or
NAT router between the client and the database? Is it possible it's
timing out mappings?
The traces indicate the database is blocked trying to write data to
the client and the client is waiting for more data from the server. To
avoid having connections pile up indefinitely you could try enabling
tcp keepalives which are ironically named because their main purpose
is to kill connections that are stuck in a state like this for too
long. But that won't fix the underlying problem.
--
greg