is there any way to deal with hydra eating upwards of 8GB of memory when evaluating my flake? I would like it to build all my systems but 8GB of memory is way too much for that, I think the issue Nix not really GCing that well if I try to `nix eval` all the systems at once, one by one when they're perfectly fine though…

won't help in my case as cgroup still count it towards the overall usage and it'll get killed anyway, if i increase the limit for the whole container i would then run the risk of OOMing my system due to a really memory intensive build that does actually need the ram actively

But it is true that i do not actually have any swap currently, i should probably get some but it won't help in this case

I could get an SLC ssd for both L2 cache and another one for swap but wow are they expensive

In reply to @magic_rb:matrix.redalder.org

won't help in my case as cgroup still count it towards the overall usage and it'll get killed anyway, if i increase the limit for the whole container i would then run the risk of OOMing my system due to a really memory intensive build that does actually need the ram actively

In reply to @magic_rb:matrix.redalder.org

But it is true that i do not actually have any swap currently, i should probably get some but it won't help in this case

I would have to move the nix-deamon out of the container into a separate container in that case

which may be a reasonable thing to do, but itll increase complexity a lot

yeah i do

nomad would still have issues with the ram amounts, i would have to keep memory down to 4GB then increase memory max above 8 to 12 which is a really bad idea because nomad schedules based on memory not memory max

i may just have to split it up, nix-daemon, hydra-http and hydra-evaluator

i dont have space for swap anywhere currently unfortunately, when i clean up the 1TB skyhawks i can put it there but i need to move data of them onto the 4tb wd reds

i would still like to somehow make hydra not eval everything at once, but i would have to know a lot more c++ to do that

In reply to @magic_rb:matrix.redalder.org

i would still like to somehow make hydra not eval everything at once, but i would have to know a lot more c++ to do that

for flakes? how would that work

In reply to Linux Hackerman
If you're on nixos, I'd suggest trying zramSwap.enable = true in any case

well when i push a change hydra will start evaluating everything at once anyway, it has no notion of "hey maybe evaluations can choke the system too"

ah ok so that may be a path forward

In reply to @magic_rb:matrix.redalder.org

ah ok so that may be a path forward