#general
- b
bored-art-40741
03/25/2016, 9:21 PMAs in, we didn't have any hangs due to locks before, but now we have crashes that don't really make any sense - b
bored-art-40741
03/25/2016, 9:22 PMWe're just going to fork for our internal release and use fasteners, which seems to work fine - b
bored-art-40741
03/25/2016, 9:22 PMI'd tentatively suggest holding the release until we figure this out--it's a major break to end up in a real release - a
ancient-stone-50795
03/25/2016, 9:39 PMThat's weird because we did have hangs and haven't seen them since. - a
ancient-stone-50795
03/25/2016, 9:41 PM(We put the updated lock into production on Wednesday) - b
bored-art-40741
03/25/2016, 9:42 PMI'm just productionizing the good lock now - b
bored-art-40741
03/25/2016, 9:42 PMI'll have an RB out shortly - a
ancient-stone-50795
03/25/2016, 9:42 PMSGTM - a
ancient-stone-50795
03/25/2016, 9:43 PM@bored-art-40741: I was wondering what would happen if you changed implementations the used the same name for the lock file. I was thinking that could lead to odd bugs. - a
ancient-stone-50795
03/25/2016, 9:44 PMThe older code would probably blow away a lock with the same name created by the fasteners library - b
bored-art-40741
03/25/2016, 9:44 PMI also considered that - b
bored-art-40741
03/25/2016, 9:44 PMIt's a plausible explanation. In our fork I also changed the name - b
bored-art-40741
03/25/2016, 9:44 PMAny time we do something like change the impl, we should also definitely change the name - a
ancient-stone-50795
03/25/2016, 9:45 PMWell, the interim situation isn't good if you really need the lock: you'll get 2 instances mucking with the ivy cache at the same time which is what the lock was trying to prevetn. - w
witty-crayon-22786
03/25/2016, 9:48 PM@bored-art-40741, @ancient-stone-50795 : sooo... does that block the release? - w
witty-crayon-22786
03/25/2016, 9:48 PMsounds like it might be a lock name collision perhaps? - a
ancient-stone-50795
03/25/2016, 9:59 PMWe also talked about a revert on Wednesday (but now there would be 2 RBs to revert) - a
ancient-stone-50795
03/25/2016, 10:00 PMI'm not sure why the patch I put in place would cause a collision with the old code - underneath it all it was the same
python library. I did update the rev of the lockfile library to the latest though.lockfile - w
witty-crayon-22786
03/25/2016, 10:18 PM...oof. >1:30 travis queue right now - b
bored-art-40741
03/25/2016, 10:19 PMOkay so it seems like my trick for writing into the file locked lockfile is not going to work. Reopening it seems to clobber the lock itself - r
rough-minister-58256
03/25/2016, 10:20 PMyes, you can’t reopen - r
rough-minister-58256
03/25/2016, 10:20 PMthats a known limitation of that locking style iirc - b
bored-art-40741
03/25/2016, 10:20 PMIt looks like I can append, given the file handle available to me - b
bored-art-40741
03/25/2016, 10:22 PMAt this point, unless someone has a better idea for creating this message in a reasonable way, I suggest pushing off the messaging part of the lock and just going with a vanilla
, which seems to work fine on its ownfasteners.InterProcessLock - b
bored-art-40741
03/25/2016, 10:22 PMI suppose I could drop a message saying "`cat <the lockfile`" for more information about how it's being held - l
limited-country-20626
03/25/2016, 10:25 PMFYI - open sourcing the graph validation stuff is blocked on the release - fsqio cannot be updated until a release is issued that includes 58fd18511bbb4ef67b851fc39366c1280bce05c8 (benjy's ban on option types broke Fsqio) - b
bored-art-40741
03/25/2016, 10:25 PMNope, even appending doesn't work - b
bored-art-40741
03/25/2016, 10:26 PMI guess I can consider the lock on the first file to also be an implied lock on a sibling "message" file - b
bored-art-40741
03/25/2016, 10:34 PMYeah, this seems to work fairly well - b
bored-art-40741
03/25/2016, 10:34 PMBut it leaves these message files floating around, which isn't great