fix[javalib]:Runtime#availableProcessors now respects cpuset on small Linux systems #3969
Conversation
|
All but one of the tests are Green. There is one macOS test which appears to be not making progress. I suspect that it will This PR is Linux only, so the macOS test "hang" is almost certainly an unrelated intermittent fault. I still have a bit of documenting to do in the issue itself. No rest for the wicked. |
There was a problem hiding this comment.
Looks great, thank you so much! @amsen20 please check if it does the right thing :)
| * logical processes configured for the system (_SC_NPROCESSORS). | ||
| * | ||
| * Note that OpenMP environment variables are neither consulted nor used. | ||
| * This consistent with Java. |
There was a problem hiding this comment.
| * This consistent with Java. | |
| * This is consistent with Java. |
|
It works for me, thanks :) |
|
natsukagami, amsen20 Thank you both for your review & suggestion/correction. It is nice to know that the proposed solution actually works in the use case and not just on my machines. If either of you think there is a short term need for either > 1024 or for FreeBSD, please create an Issue. Somehow, I think we will all be back here again. |
|
There is one failing Windows Test. I appears to be an intermittent failure, Otherwise all Green & ready for review. |
Weakly fixes #3963
Runtime#availableProcessorson small Linux systems now reports the number of logical processors allowby the process
cpuset.This number may be smaller than that reported by
sysconf(_SC_NPROCESSORS_ONLN)due to variousLinux mechanisms: cgroups, taskset, cpuset, etc.
'small' is defined as less than or equal to, currently, 1024. Above that limit
sysconf(_SC_NPROCESSORS_ONLN)isreported, same as before this PR.
See Issue for discussion and manual testing instructions (next sprint). Most notably, OpenMP variables are not considered.