This is actually how we have Apache configured to respond to Gerrit being unavailable - that error page is served with a 503 when Gerrit is really down.
Today I hacked it to always show that page, so even when it was "up" people wouldn't be hitting it -- we were still debugging and restarting things so I didn't want to give false hopes or end up with half-completed transactions.
This can all be improved I think with some Apache config tweaks.
-Chad
On Thu, Oct 6, 2016 at 4:14 PM Amir Ladsgroup ladsgroup@gmail.com wrote:
It was bothering to me but I'm guessing this is one of so so many flaws of gerrit itself and probably not fixable easily (other people are more qualified to comment) but i want to suggest speeding up the process to move to differential which is much better in handling such down times alongside with other benefits.
Best
On Fri, Oct 7, 2016, 2:26 AM Gergo Tisza gtisza@wikimedia.org wrote:
Thanks a lot for the quick recovery!
Would it be possible to use something other than a redirect next time when traffic needs to be blocked? An apache deny rule or a 404 would work, but a redirect means that reloading the page (or reopening the browser) will cause the URL to be lost with little hope of recovery (browsers don't record redirects in the history). That can be very annoying when one uses tabs as bookmarks (bad habit as it is).
On Thu, Oct 6, 2016 at 3:33 PM, Chad Horohoe chorohoe@wikimedia.org wrote:
Hi!
Sorry for the extended downtime! From what we can tell, it appears as though the machine that Gerrit is running on (lead) is having some hardware issues that are making the CPU misbehave. We've worked around it for now, so things should be up (and Zuul is processing CI events just fine).
However, since it appears it's a hardware problem, we're planning to migrate off of lead to a new machine (cobalt). The public IP addresses will not be changing. The plan right now is to do this migration tomorrow with a scheduled downtime at 17:00UTC (10:00 PST).
We'll be keeping a close eye on things in the meantime, so if things deteriorate again we can start the migration sooner.
(and yeah, wikitech incident report to follow, I'm a little burnt out right now though)
Thanks again for bearing with us!
Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l