The upstream issue was closed but not fixed AFAIK. I've never been able to reproduce the problems locally either but hydra builds regularly failed.

Failure on aarch64-linux (full log)

Attempted: rustc

tar: A lone zero block at 9623
install -d -m 755 '/nix/store/k9h9x4b1ncdy205dcx6qq7q9cg18xv60-git-2.18.0/lib/perl5/site_perl/5.28.0'
(cd perl/build/lib && tar cf - .) | \
(cd '/nix/store/k9h9x4b1ncdy205dcx6qq7q9cg18xv60-git-2.18.0/lib/perl5/site_perl/5.28.0' && umask 022 && tar xof -)
tar: Skipping to next header
tar: Exiting with failure status due to previous errors
make: *** [Makefile:2733: install] Error 2
builder for '/nix/store/0c80jp9nlf16f2kywfsy7sfhlh3kgpvl-git-2.18.0.drv' failed with exit code 2
cannot build derivation '/nix/store/k9p6bn9s3h3r241hnrcldi3mz7w8m1bw-rustc-1.27.0.drv': 1 dependencies couldn't be built
error: build of '/nix/store/k9p6bn9s3h3r241hnrcldi3mz7w8m1bw-rustc-1.27.0.drv' failed

Timed out, unknown build status on x86_64-linux (full log)

Attempted: rustc

   Compiling rustc_mir v0.0.0 (file:///build/rustc-1.27.0-src/src/librustc_mir)
   Compiling rustc_typeck v0.0.0 (file:///build/rustc-1.27.0-src/src/librustc_typeck)
   Compiling rustc_traits v0.0.0 (file:///build/rustc-1.27.0-src/src/librustc_traits)
   Compiling rustc_allocator v0.0.0 (file:///build/rustc-1.27.0-src/src/librustc_allocator)
   Compiling rustc_resolve v0.0.0 (file:///build/rustc-1.27.0-src/src/librustc_resolve)
   Compiling rustc_save_analysis v0.0.0 (file:///build/rustc-1.27.0-src/src/librustc_save_analysis)
   Compiling rustc_privacy v0.0.0 (file:///build/rustc-1.27.0-src/src/librustc_privacy)
   Compiling rustc_plugin v0.0.0 (file:///build/rustc-1.27.0-src/src/librustc_plugin)
building of '/nix/store/31iqgzrqi9jdnkp60zynym7q91l68grm-rustc-1.27.0.drv' timed out after 1800 seconds
error: build of '/nix/store/31iqgzrqi9jdnkp60zynym7q91l68grm-rustc-1.27.0.drv' failed

This was the pull request, where we tried to enable it earlier: #21742

I did build the entire rebuild on one of my machines during the last day. It finished some time this morning. rustc was compiled without any visible issue. I had a load of around 48 to 600 during the entire process.

I also just build rust on it's own again. I think we should give this a try. None of us able to produce an error and if hydra does we are doing something wrong...

Dealing with intermittent failures that have a large dependency tree like rustc is a pain in hydra. There are still thinks like literal sleeps in the rust makefiles, I'm almost certain we're going to revert this anyway. So this seems a bit pointless to me, even tho I'd like to have it enabled, since there's no indication that the problems might be resolved.

I suspect lower the io bandwidth in combination with concurrent builds are triggering problems that can't be reproduced on a machine with SSD or something similar like that.

If we're not sure about this we could put it in its own little Hydra jobset and restart the build multiple times to see what happens.

(triage) Anyone knows who to poke to put this PR in its Hydra jobset so that we can give log files to upstream and they can fix the issue?

I don't think this is an issue anymore, rustc-1.36 was able to max out my cores during the last build.

@andir @oxij @LnL7

In fact, this was not an issue even once on any of my builders since I opened this PR. I'm living happily with thing change applied for almost a year now.

@oxij are you ok with closing this PR then?

I think he meant that the concerns about the problems with this PR are now invalid.
I.e. we can probably merge this PR.

@GrahamcOfBorg eval
@GrahamcOfBorg build rustc

I've just tried nix-build --check -E 'let pkgs = import ./. { overlays = []; }; in pkgs.rustc' on ed755a6 ; looks like at least the time before the build gets oom-killed does use all my cores, though intermittently, probably due to limits in what's parallelizable.

Maybe upstream just forced parallel builds on?

seems to fan in and out depending on workload. At least most of it is already parallel

either way, seems to take about as long as specifying enableParallelBuilding = true; or not