Open t-bltg opened 2 years ago
It seems like an upstream issue, or it might be because of https://github.com/aminya/CompileBot.jl/commit/a88f7e3c8539bf8f30948c93cf475e226c2b89b8
Have you tried this locally?
I've re-run CI against CompileBot@1.0.8
, and it still fails.
The scripts hangs at [ Info: Processsing the generated precompile signatures
. I think the memory usage grows endlessly and the process crashes with ProcessSignaled(9)
(no more memory available), after ~2h
(it should only take ~5min
to generate the precompilation statements, as is done on julia 1.6
and 1.7
).
The process ends up without errors: [ Info: precompile signatures were written to [...]/Plots.jl/deps/SnoopCompile/precompile/1.8
locally using 1.8.0-beta3
, the process taking up to around 15Gb
RAM.
It seems that the github actions nodes only have MemTotal: 7113128 kB
~ 7GB
RAM memory, and 4GB
swap. What, in the precompilation files generation process, is taking that much memory ?
Clearly going further with the latest run (I've added 10GB
swap in CI), but 5h 51m
(build cancelled automatically) is a massive regression ...
Try excluding some tests from the test script and see if you can find the faulty test. https://aminya.github.io/CompileBot.jl/dev/#Create-the-script-that-runs-snoop_bot-1
Might be related to julia being able to precompile more methods?
See e.g. failing job at https://github.com/JuliaPlots/Plots.jl/runs/6023176404?check_suite_focus=true.
Are there any ways to debug this ?