[Catalog-sig] PyPI down :(

"Martin v. Löwis" martin at v.loewis.de
Sat Jan 19 00:37:35 CET 2008

> On Fri, Jan 18, 2008 at 01:47:27PM -0200, Daniel (ajax) Diniz wrote:
>> There's been a problem with your request
> Thanks for your report.  I've restarted the web server and that 
> seemed to clear the problem.

I've studied this a bit. Apparently, at 9:52 UTC, the OOM killer killed
a process (apache2), then over the next minutes it killed more processes
(apache2, moin.fcgi, pypi.fcgi, and eventually, at 10:04, it killed
"postmaster". That apparently resolved the out-of-memory situation, but
it also meant that PyPI broke. The postgres log file has all kinds of
messages on stopped and crashed servers, rolled-back transactions etc,
but eventually, postgres recovered. Not so PyPI, which (for some reason)
failed to reconnect to the server. Looking at the tracebacks, it seems
that psycopg now reports OperationalError for broken connections, and
not (anymore?) InterfaceError.


More information about the Catalog-SIG mailing list