ok, so that means pr 37705 was incomplete. i think it's trying to set the limit too high, it fails once, then it falls back to the default. it should probably make two attempts: set it to the recommended level, and if that fails and we're still below the minimum, try one more time to set it to the minimum
After we address this, the bincheck workaround should be reverted.
also assigned @bobvawter in case you should be routing this instead of ben
Urgh. Seems like we should be taking the minimum of kern.maxfilesperproc, kern.maxfiles, and ulimit -n.
@benesch i just reached the same conclusion and have a PR almost ready
Most helpful comment
@benesch i just reached the same conclusion and have a PR almost ready