If purge_builds were to delete directories older than a week or so, we would a) clean out old cruft from the /builds/slave directory b) make it so that we're not always running as close to full as possible
Comment on attachment 475827 [details] [diff] [review] delete directories older than 7 days Looks good
This will land on Monday.
So the summary and comments do not match the code I see here... It seems like we _always_ preserve builds newer than 7 days, rather than always delete older than 7 days. [am I wrong] I expected this bug to be a "We want to delete all builds older than 7 days, and/or newer builds if we still do not have the free-space we want" Am I either missing something or wrong?
(In reply to comment #4) > So the summary and comments do not match the code I see here... Nope just confused. > Am I either missing something or wrong? Missed the addition of |dirs.sort()| before the while... sorry for noise.
This is landed and working.
Comment on attachment 475827 [details] [diff] [review] delete directories older than 7 days Landed in e179cf0fd98b
This got backed out because it caused windows 7 test machines to fail (because they timed out when cleaning up).
could we include this change with a bigger threshold at first and then reduce it every week? I believe the machines that burned the day we enabled had many many GBs of old stuff and that is why it timed out as it had too many things to remove. Think of the IX machines that barely need to clobber as their disk always have more than 12GB free and everything is just getting older. Maybe we could ask the script to check if they have spend X amount of mins deleting stuff and then stop if they have passed it. Luckily the machine would stop between X and the buildbot timeout.
Why is this needs-treeclosure? if the patch that broke win7 hasn't been changed?
(In reply to comment #10) > Why is this needs-treeclosure? if the patch that broke win7 hasn't been > changed? We expect that after the win7 slaves cycle through and clean up, that this will work ok going forward. So it needs tree closure because we expect burning the first time stuff is getting run, but it should be ok after.
I'm also fine with Armen's suggestion of increasing the threshold. The max_age default at line 130 needs to be changed. I'd recommend 28 days to start with.
Comment on attachment 475827 [details] [diff] [review] delete directories older than 7 days Checked in with 28 day default: http://hg.mozilla.org/build/tools/rev/7920b159c3e6
Leaving open to crank down the threshold.
Created attachment 486252 [details] [diff] [review] reduce max age to 14 days
Doesn't need a reconfig per se, but needs somebody to watch it for a bit.
Comment on attachment 486252 [details] [diff] [review] reduce max age to 14 days changeset: 883:5a99b6e600c6
14 days is a good number to finish with.