Hello, torque 3.x fails with cpuset support enabled fails to run tasks on 3.2.x kernels (and presumably on all 3.x kernels). The reason for this is changed path names for cpuset files, now they are the same as in cgroup filesystem, even if fs was mounted with -t cpuset. I described this in more detail in the torque bug_#195. Patch is available, it was tested on 3.2.14 vanilla kernel.
Created attachment 311579 [details, diff] torque-3.0.5-cpusets.patch It applies for 3.0.4 too. I just bumped my local version.
Created attachment 311581 [details] emerge --info
Sorry for the massive delay but is this still an issue?
Hello, yes, this is still an issue even with torque-3.0.6. Ustream suggests another solution: mount cpuset with noprefix option, but this does not work if cgroups is already mounted, which portage does; see discussion here: http://www.clusterresources.com/bugzilla/show_bug.cgi?id=195 So I still use this patch with 3.0.6. I haven't tried 4.x branch and reading torque-users mail list suggests me that it is still too buggy for production, though some people are already using it in. Anyway in will be useful to see 4.x in portage.
(In reply to Andrew Savchenko from comment #4) > Hello, > > yes, this is still an issue even with torque-3.0.6. > Ustream suggests another solution: mount cpuset with noprefix option, but > this does not work if cgroups is already mounted, which portage does; see > discussion here: > http://www.clusterresources.com/bugzilla/show_bug.cgi?id=195 > So I still use this patch with 3.0.6. > > I haven't tried 4.x branch and reading torque-users mail list suggests me > that it is still too buggy for production, though some people are already > using it in. Anyway in will be useful to see 4.x in portage. Well I just pushed 4.1.5.1 which hopefully has most of the bugs ironed out of it by now. When you have time, would you mind checking it out? And trust me, given how long I sometimes take to respond, I won't hold it against you if you don't test :)
Hello, at this moment our system is used in production, so it is a bit premature to test torque-4 there considering severity of changes made since 3.x branch. I'm planning test it on our playground, though it is not on the top of the list now, so it will take some (considerable) time. Nevertheless I'll report back when and if I'll have any results.
If anyone here can test cpusets with torque-4 or torque-6, and report back if it is still an issue, I would appreciate it. For now I'm closing as RESO/OBSOLETE since v3 has been removed from the gentoo repo for some time now.
