On Mon, 22 Mar 2004, Andrew Kelly wrote:
> We are about 99% certain we have succeeded in duplicating the platform
> and installing the application. So far everything has worked wonderfully
> with one single exception.
> The application provides the option of performing a new search within
> the result set of the previous search. This worked well in the
> application on the previous host, but not on the new host. Now, when one
> tries to search with the results of a previous search, the postmaster
> process drives CPU usage to 100% and the application never returns a
> result. Strangely enough, it doesn't max out both processors at the same
> time, but the states do change. The maxed out CPU will suddenly drop
> under 1% as the second CPU blows out to 100%.
> Only kill seems to release the CPU.
>
> Now, I realize there is not enough info here to debug this, but that is
> mostly why I'm writing. I know nothing about this app and will have to
> plod through the source to learn it. I'm also a cherry to postgresql (to
> sql in general).
> My stomache tells me this is config issue with the host in general,
> maybe a permissions issue in cache space or something, but I'm only
> guessing. The databases were a straightforward dump at the old site and
> restore on the new site, and nothing has changed in the source code, and
> the fact that everything else runs perfectly seems to indicate it was a
> good duplication.
>
> Can anybody give me a hint on where to start looking and what tools to
> use? I'm a flipping nervous wreck suddenly being responsible for
Well, the first thing to look at is to check the two postgresql.conf
files. It's also possible that they were sending options on the command
line if they were really trying to be annoying, so if you have access to
the startup scripts they were using, there might be information there.
I don't have a 7.2 box to check configuration options against, but there
should be something like log_statement which if set to true will log the
statements sent to the database. You should make sure you're doing
something with the log output (either sending it to syslog or making sure
that you send stderr to a file somewhere). That'll let you see what
statements are happening.
If it's consistently happening on a particular query, you can see any
things like NOTICEs that come up and you can use EXPLAIN to get the query
plan for the query. With that and the table schema of the affected tables
we can probably help a bit more. (We'd normally ask for EXPLAIN ANALYZE
but if the query is taking way too long, it will also since it actually
runs the query while a plain EXPLAIN does not).
Finally, 7.2.4 is fairly old. Once you have the thing working, you
might want to play with upgrading. :)
