www.osm.org and api.osm.org tune Apache HTTP timeouts

[Firefishy]

Currently both www and api share the same HTTP timeout, the HTTP timeouts should tuned appropriately.

• www.openstreetmap.org should have a ~30 seconds HTTP timeout
• api.openstreetmap.org should likely retain the 300 seconds HTTP timeout

[tomhughes]

Maybe do some research before suggesting timeouts that exactly match what we already have?

[tomhughes]

Here's the settings if you want proof... https://github.com/openstreetmap/openstreetmap-website/blob/f138055849e5b09cd94797c3e1705c6357d6b93b/config/settings.yml#L48

[Firefishy]

I know the api timeout is 300s. The main issue is www should not be such a long timeout.

[tomhughes]

It isn't, it's 30 seconds, that's what web_timeout is.

It doesn't really help though because once one of them is hogging all the daemons the other one won't get a look in.

None of it is in any way relevant to anything that happened today as far as I know, but I'll know more once I've actually finished getting to the bottom of what happened instead of just guessing and throwing random suggestions in the ring.

[Firefishy]

I have tweaked the ticket title to include Apache. Currently apache's global timeout is set to Timeout 300. Timeout directive sets many underlying apache timeouts if not explicitly set.

[Firefishy]

https://httpd.apache.org/docs/2.4/mod/core.html#timeout

[tomhughes]

We already have the request timeout configured separately in apache via the reqtimeout module.

I'm still investigating what happened but at least in part timeouts were actually making it worse because we wound up with lots of database queries still running after the web requests that initiated them had timed out.