!sBfrWMVsLoSyFTCkNv:nixos.org

OfBorg

164 Members
Number of builds and evals in queue: <TBD>60 Servers

You have reached the beginning of time (for this room).


SenderMessageTime
12 Feb 2024
@infinisil:matrix.org@infinisil:matrix.org * Regarding https://github.com/NixOS/nixpkgs/pull/269403, I actually don't think it would improve ofborg eval speed at all00:53:21
@infinisil:matrix.org@infinisil:matrix.orgIt gives a 5x speedup if you have 31 additional cores without anything to do00:53:55
@infinisil:matrix.org@infinisil:matrix.orgWell I guess it depends on how booked-out ofborg generally is00:54:42
@infinisil:matrix.org@infinisil:matrix.orgBecause if it's already using 32 cores for 32 separate evaluations, that PR would not help at all (in fact it would make it a lot worse, because each parallel unit of work duplicates work..)00:55:42
@cole-h:matrix.orgcole-hofborg does not use 32 cores for 32 evaluations 😆 It only has enough memory to run 3 evaluations on one machine most times. (256GiB total, each eval-outpaths takes ~60+GiB of memory.)15:03:16
@cole-h:matrix.orgcole-hSo, provided there's no huge build taking place on the same box, the other cores are usually sitting around twiddling their thumbs.15:04:13
@raitobezarius:matrix.orgraitobezarius
In reply to @infinisil:matrix.org
Because if it's already using 32 cores for 32 separate evaluations, that PR would not help at all (in fact it would make it a lot worse, because each parallel unit of work duplicates work..)
well it's not like Nix is smart and knows how to deal with values appropriately
15:51:22
@raitobezarius:matrix.orgraitobezarius so not sure if that would be that wasteful 15:51:26
@raitobezarius:matrix.orgraitobezariusmeasurements are important here15:51:28
@raitobezarius:matrix.orgraitobezariusbecause the GC is really in a bad shape15:51:36
@infinisil:matrix.org@infinisil:matrix.orgI see, so it sounds like it would still help a bunch16:29:41
@trofi:matrix.orgtrofi It would improve in a sense that that PR does not really list a superset of all hydra attributes. For example it skips all recursive attributes like pythonPackages or haskellPackages. 21:42:19
@trofi:matrix.orgtrofi But I agree nix's RAM usage for tree traversals is unusable on large trees. https://github.com/NixOS/nix/issues/9671 feels like being a bit more eager about evicting unreferenced values would help a lot. nix-env -qa is a bit trickier as package list is built and sorted outside nix expressions. 21:44:08
14 Feb 2024
@keiichi:matrix.orgtetoHow can I see the actual ofborg queue ? I would like to see where my PR (288761) darwin job is to get an idea how long it will take. I see there are 133 jobs19:35:26
@7c6f434c:nitro.chat7c6f434cBased on another PR and the stats, I think the two Darwin queues are comparable, so one of them reaching that PR is a good sign…19:40:14
@cole-h:matrix.orgcole-h You can't, sorry.  19:40:31
@keiichi:matrix.orgtetoI suspected so :'( ty19:59:35
16 Feb 2024
@a-n-n-a-l-e-e:matrix.org@a-n-n-a-l-e-e:matrix.orgnot sure if this is the right channel but: i think i'm seeing some failures on ofborg aarch64-darwin that aren't on hydra. hydra builds are passing but ofborg builds are failing. i created ofborg builds by creating a PR and adding a comment in postPatch to trigger a rebuild but seem to be failing 100% while running unit tests. I noticed issues when trying to upgrade the same package and which is why i backgracked to a known good build (green hydra) but am still seeing failures.09:11:01
@a-n-n-a-l-e-e:matrix.org@a-n-n-a-l-e-e:matrix.orghttps://hydra.nixos.org/job/nixpkgs/trunk/python311Packages.duckdb.aarch64-darwin/all09:11:22
@a-n-n-a-l-e-e:matrix.org@a-n-n-a-l-e-e:matrix.orgfailures: https://logs.ofborg.org/?key=nixos/nixpkgs.289199&attempt_id=1451caa6-78d6-4b52-80b1-abcbaaa29298 (all aarch64-darwins are fails) from PR https://github.com/NixOS/nixpkgs/pull/289199/checks?check_run_id=2164600442309:12:26
17 Feb 2024
@a-n-n-a-l-e-e:matrix.org@a-n-n-a-l-e-e:matrix.orgrace in unit test -- thought i disabled it but did not. race not triggered on hydra as i guess hydra is machines are slower.06:14:16
20 Feb 2024
@janik0:matrix.org@janik0:matrix.orgdid the ofborg storage run full some time ago? https://github.com/NixOS/nixpkgs/pull/27637616:59:07
@cole-h:matrix.orgcole-hIt did last month, yes, but that should be resolved now.17:00:15
@lily:lily.flowersLily Foster
In reply to @cole-h:matrix.org
It did last month, yes, but that should be resolved now.
nix should be automatically garbage collecting with the min-free option, right? i forget if we had something like that set on the ofborg builders
17:01:08

Show newer messages


Back to Room ListRoom Version: 6