#general
- r
rough-minister-58256
07/02/2015, 5:49 PM(e.g. within a task as well as at the RoundEngine level itself) - r
rough-minister-58256
07/02/2015, 5:51 PM@enough-analyst-54434 @bored-art-40741 ^ - b
bored-art-40741
07/02/2015, 5:52 PMHopefully we can discuss that a bit, but right now we're just trying to get some valid schedules on the board - b
bored-art-40741
07/02/2015, 5:52 PMI'll keep the channel updated - r
rough-minister-58256
07/02/2015, 5:52 PMrgr that - h
happy-kitchen-89482
07/02/2015, 6:07 PMre that - we should strongly desire that tasks operate exclusively on a set of input products (with no out-of-band "context"), and produce a set of output products (with no side effects, other than caching/memoizing for performance). Then parallelization becomes much more straightforward - either a remote node computes an input product (or has already computed and cached it in the past) or the product is shipped to it, whichever performs better in each case. - h
happy-kitchen-89482
07/02/2015, 6:08 PMNot to get all functional, but tasks do need to be composable and have no side-effects. - w
witty-crayon-22786
07/06/2015, 3:55 PM+1 - a
ancient-stone-50795
07/06/2015, 9:22 PMIf no one objects, I'm going to publish 0.0.34 in a bit: https://rbcommons.com/s/twitter/r/2436/ - a
ancient-stone-50795
07/06/2015, 9:43 PM - h
happy-kitchen-89482
07/06/2015, 9:44 PMI need to change that tag line - h
happy-kitchen-89482
07/06/2015, 9:44 PM"bottom-up" and "pants" sounds too much like a bad pun - a
ancient-stone-50795
07/06/2015, 9:50 PMHeh. - b
bored-art-40741
07/06/2015, 10:00 PMAgreed - w
witty-crayon-22786
07/07/2015, 4:03 PMplanning to have a newhire work on this one to reduce disk usage on our iron clusters: https://github.com/pantsbuild/pants/issues/1762 - p
proud-alligator-68508
07/07/2015, 5:56 PMwe run tmpwatch, fwiw - p
proud-alligator-68508
07/07/2015, 5:56 PMI didn’t feel like re-inventing it inside pants, and it was a one-liner:tmpwatch --all 14d .local_artifact_cache - p
proud-alligator-68508
07/07/2015, 5:59 PMwe also run that in cron on our remote artifact caches - w
witty-crayon-22786
07/07/2015, 6:04 PMhow do you clean up on developer laptops? - w
witty-crayon-22786
07/07/2015, 6:05 PM(no tmpwatch there by default) - w
witty-crayon-22786
07/07/2015, 6:05 PMalso, linux boxes usually have atime off - w
witty-crayon-22786
07/07/2015, 6:05 PManyway, rather than different solutions for each of the 3 environments, something native isn't too hard. - p
proud-alligator-68508
07/07/2015, 6:43 PMfair enough. - p
proud-alligator-68508
07/07/2015, 6:43 PMi noticed you mention running in in the background there - p
proud-alligator-68508
07/07/2015, 6:44 PMnote that HFS+ is so terrible at deletes that we often see severely degraded performance during cleans / prunes, so we’d probably want it to only be explicitly triggered, and run in the foreground, by default. - b
bored-art-40741
07/07/2015, 6:45 PMYou don't need atime, just relatime - b
bored-art-40741
07/07/2015, 6:46 PMWhich is not nearly as expensive - p
proud-alligator-68508
07/07/2015, 6:46 PMeg We used to default to clean-async, but had to switch to blocking because of how often people came complaining of terrible build perf when it turned out it was just OSX bogged down with a background delete - p
proud-alligator-68508
07/07/2015, 6:47 PMalso, = @bored-art-40741, we run our cache boxes and build machines with relatime, which is more than enough for pruning “old” files but doesn’t hit every access with the write cost. - p
proud-alligator-68508
07/07/2015, 6:49 PMI think the default is to only re-rwite the atime if it is something like a few minutes old