<samueldr>
hmm, doubling evaluator_max_heap_size to 22000000000 still gives Too many heap sections: Increase MAXHINCR or MAX_HEAP_SECTS :/
<samueldr>
let's double it again
init_6 has joined #nixos-dev
jtojnar has quit [Read error: Connection reset by peer]
<gchristensen>
should we ditch the nixpkgs-unstable channel, and just have nixos-unstable (can still update nixpkgs-unstable, but have it always equal nixos-unstable)
<samueldr>
I wouldn't care, but I'm not one using it
<samueldr>
are there recent cases where nixpkgs-unstable really lagged behind?
<samueldr>
btw, tried a couple other values for evaluator_max_heap_size, at 8x its initial value (88000000000) it still doesn't evaluate successfully
<gchristensen>
don't know
<LnL>
so nixos would block for darwin?
<gchristensen>
yeah, LnL
<samueldr>
sure doesn't help that I'm not confident with how it's used
<LnL>
I wouldn't complain about having people pay better attention to stuff there, but I'm not sure it's the best idea
<gchristensen>
aye
<gchristensen>
I don't like that people describe a channel called "unstable"
<gchristensen>
is why I ask :)
<LnL>
samueldr: if I understand it correctly it will also try to allocate blocks that are too large if you use a high value
<LnL>
so you kind of have to find a good balance
<LnL>
but I could be totally wrong about that
<samueldr>
no idea, but the values used are the ones configured in nixos' hydra, and then I gradually increased the max heap size
<samueldr>
I have 64GiB ram to play with, 56GiB available ram here :/
<gchristensen>
how much ram do you have?...oh
<samueldr>
enough!
<gchristensen>
there are other boehm environment variables you can play with
<samueldr>
considering it uses (according to ram) 8.16GiB
<samueldr>
yeah, but is it something that should be done blindly, while this is something that would be required to be replicated "in prod"?
<samueldr>
I'd much rather let someone that has a better grasp of the situation look at it :/
<LnL>
try GC_INITIAL_HEAP_SIZE=8G
<gchristensen>
oh yeah, thats a good one
<samueldr>
LnL: it's by using the hydra evaluator
<samueldr>
~/Projects/nixos/hydra/result/bin/hydra-eval-jobs --option evaluator_initial_heap_size 10000000000 --option eva
<samueldr>
luator_max_heap_size 88000000000
<gchristensen>
hydra-eval-jobs takes those as options?
<samueldr>
I hope the initial heap size there is what's used
<samueldr>
according to clever it might
<gchristensen>
let's find out
<samueldr>
at least, I saw different results without
<samueldr>
(I haven't verified yet, but now think I should)
<LnL>
yeah looks like it
<gchristensen>
auto initialHeapSize = config->getStrOption("evaluator_initial_heap_size", "");
<samueldr>
plausible, annoying though that other than "it looks funny" I don't really know how to report that
<samueldr>
or even if my interpretation is right
<gchristensen>
maybe git log will shed some light
<gchristensen>
not a lot
<samueldr>
I'm checking what static means in `static size_t maxHeaps` just to be sure
pie__ has quit [Remote host closed the connection]
pie___ has joined #nixos-dev
init_6 has quit []
worldofpeace has joined #nixos-dev
jtojnar has joined #nixos-dev
init_6 has joined #nixos-dev
<samueldr>
eek, with GC_INITIAL_HEAP_SIZE=10G it got to 17GiB of use (according to time) but still failed with Too many heap sections (but I did miss your GC_INITIAL_HEAP_SECTIONS) recommendation when I started it
<gchristensen>
oh I meant SIZE... I forget what SECTIONS does
<timokau[m]>
IIRC the restart won't propagate automatically right? If that's the case, please also restart sage
drakonis has joined #nixos-dev
Drakonis__ has joined #nixos-dev
drakonis1 has joined #nixos-dev
drakonis_ has quit [Ping timeout: 240 seconds]
drakonis has quit [Ping timeout: 252 seconds]
Drakonis__ has quit [Ping timeout: 252 seconds]
phreedom has quit [Remote host closed the connection]
drakonis1 has quit [Read error: Connection reset by peer]
phreedom has joined #nixos-dev
<samueldr>
I think I restarted the right sage build
<timokau[m]>
It apparently did restart, but immediately failed again. Not sure if that's hydra goofyness because suitesparse wasn't done yet or a real issue.