Created attachment 946094 [details] vdsm+engine logs Description of problem: This is a clone of BZ #1141658,though consequences are far worse. The rebooted host never come back due to PPC wrong boot configuration, hence, makes the whole setup useless. We also need to make sure that the sanlock fix for rhel7 BZ #1142454 will be available for the rhev-m/IBM/fedora OS as well. Version-Release number of selected component (if applicable): How reproducible: 100% see BZ #1141658 for more description
(In reply to Ori from comment #0) It is not clear what is "wrong PPC boot configuration", and how is this related to vdsm. If the root cause of this bug is the same as in bug 1141658, then this is a duplicate. If this bug describe another issue (wrong boot configuration?), then it is not a vdsm bug. Also I don't see how reconstruct master is related to bug 1141658. That bug is about blocking access to storage, which leads to reboot of the machine, because sanlock cannot terminate or kill vdsm (because selinux policy is incorrect). To make it more clear, please attach these logs: - /var/log/sanlock.log - /var/log/audit/audit.log* Also missing: - vdsm version - steps to reproduce - reproducible: 100% - how may times did you reproduced this?
Nir, This bug is related to BZ #1141658, but this pertains to PPC (RHEV for IBM PPC), we require separation between the bug given the different platforms and multiple validations. Regarding the "wrong PPC boot configuration", the reboot caused by the selinux blocking the SAN locking mechanism ends up with the OS not booting up (wrong petitboot configuration). This indeed is not vdsm related but is just my observation. Per comment https://bugzilla.redhat.com/show_bug.cgi?id=1141658#c4, the reconstruct master is related to this bug. This is 100% reproducible as stated in the bug, I reproduced this 4 times. I will attach the logs requested with the 5th try. Thanks, Ori.
(In reply to Gilad Lazarovich from comment #2) > Per comment https://bugzilla.redhat.com/show_bug.cgi?id=1141658#c4, the > reconstruct master is related to this bug. The issue is not related to in any way to reconstruct master. The issue (bug 1141658) is: 1. blocking access to storage 2. sanlock try to stop the spm and fail (selinux policy bug) 3. sanlock reboot the machine (expected behavior) Reconstruct master did not happen because the machine was rebooted. > This is 100% reproducible as stated in the bug, I reproduced this 4 times. > I will attach the logs requested with the 5th try. Do not forget the steps to reproduce. Currently we can do nothing with this bug.
Sorry Nir, PPC OS is taking it's first QA steps which means that at any given time, some Engineers are running tests on the system, I cannot make any changes to the SElinux,in addition, from the web https://brewweb.devel.redhat.com/taskinfo?taskID=8095125 it's not clear whether the packages supports PPC, I don't want to break any running tests.
(In reply to Ori from comment #7) > Sorry Nir, PPC OS is taking it's first QA steps which means that at any > given time, some Engineers are running tests on the system, I cannot make > any changes to the SElinux,in addition, from the web > https://brewweb.devel.redhat.com/taskinfo?taskID=8095125 it's not clear > whether the packages supports PPC, I don't want to break any running tests. Then this bug will have to wait until you have a release including this selinux policy.
(In reply to Ori from comment #7) > Sorry Nir, PPC OS is taking it's first QA steps which means that at any > given time, some Engineers are running tests on the system, I cannot make > any changes to the SElinux,in addition, from the web > https://brewweb.devel.redhat.com/taskinfo?taskID=8095125 it's not clear > whether the packages supports PPC, I don't want to break any running tests. It's a noarch package - It supports PPC. Please schedule yourself a window on the PPC machine and test with this package.
will do it next time I'll have those hosts.
this is already in 3.4.3, ON_QA for retest…
BZ #1156017 have been marked as blocker due to lack of storage resources on host, I need 2 domains with different ip's to verify the correct behavior, right now the creation of any nfs Storage domain is impossible.
verified on powerKVM latest ver. the whole operation takes longer than usual though: normal time is about 15 minutes where reconstruct on PPC setup took about 22 minutes.
released Oct 31