action #88225


osd infrastructure: Many failed systemd services on various machines

Added by okurz over 3 years ago. Updated over 3 years ago.

Target version:
Start date:
Due date:
% Done:


Estimated time:



hi guys, is disabled since some weeks since we had bigger problems which we already handled in various tickets, e.g. the broken worker issues reg. network, but it shows currently 14 (!) failed systemd services on our hosts. I think the original ticket is still blocked but by a new issue. I will create a new urgent issue to handle the plethora of failed services

Acceptance criteria

  • AC1: Significantly reduced number of failed systemd services
  • AC2: alert is again enabled

Related issues 1 (0 open1 closed)

Related to openQA Infrastructure - action #88474: All workers on powerqaworker-qam-1 are offlineResolvedlivdywan2021-02-08

Actions #1

Updated by okurz over 3 years ago

  • Status changed from New to Workable
  • Assignee set to nicksinger

assigning to nsinger as proposed by him. Thanks, you are awesome! :)

Actions #2

Updated by nicksinger over 3 years ago

starting out with qa-power8-5-kvm we have the most services failing:

● kdump-early.service      loaded failed failed Load kdump kernel early on startup
● kdump.service            loaded failed failed Load kdump kernel and initrd      
● logrotate.service        loaded failed failed Rotate log files                  
● rebootmgr.service        loaded failed failed Reboot Manager                    
● snapper-cleanup.service  loaded failed failed Daily Cleanup of Snapper Snapshots
● snapper-timeline.service loaded failed failed Timeline of Snapper Snapshots     
● iscsid.socket            loaded failed failed Open-iSCSI iscsid Socket          
  1. iscsid complained about iscsid.socket: Failed to listen on sockets: Address already in use. After restarting the service it worked again. However, I raise the question if the whole iscsi setup is still necessary. From a first look I can't make out its use anymore. (also see
  2. The snapper services hit a connection limit of dbus. To avoid this in the future I created /usr/sbin/rebootmgrd showed up 245 times in the process tree. Some very old ones from before we disabled it (with but also at least one from 2021. Oli raised the very valid point that salt still starts the unit (which can't be confirmed by systemctl status rebootmgr). Therefore I will mask them all for all PPC workers with: salt -l error --no-color -C 'G@roles:worker and G@cpuarch:ppc64le' "systemctl disable --now rebootmgr && systemctl mask rebootmgr"
  3. kdump-early (and therefore kdump) fails as the crashkernel cmdline is missing. I assume that this was because somebody did not follow while rebooting the machine. At least there where no additional parameter in /proc/cmdline. Rebooting the machine with the correct kexec-line fixed that as well. Services are both running now.
  4. logrotate failed to rename /var/log/openvswitch/ovs-vswitchd.log to /var/log/openvswitch/ovs-vswitchd.log-20210126: Permission denied. I've created and applied salt -l error --no-color -C 'G@roles:worker' "chown openvswitch:openvswitch /var/log/openvswitch/ && systemctl restart logrotate" as workaround for now. This got rid of almost every other systemd service failing on other servers too :)
Actions #3

Updated by nicksinger over 3 years ago

  • Priority changed from Urgent to Normal

Unfortunately my workaround for logrotate didn't work and I couldn't find any solution yet. Therefore I will leave the alert disabled. However we're just having 1 failing service per worker now so I reduce the urgency at least :)

Actions #4

Updated by nicksinger over 3 years ago

  • Status changed from Workable to Feedback

I've came up with a hopefully working workaround which is implemented in salt:
If the workaround is merged and deployed I'd give enabling the alert another try. Feedback until then :)

Actions #5

Updated by livdywan over 3 years ago

  • Related to action #88474: All workers on powerqaworker-qam-1 are offline added
Actions #6

Updated by nicksinger over 3 years ago

I extended the workaround state another time with . It seems like logrotate doesn't fail anymore on systems where the directory was properly owned by openvswitch. It was still failing on grenache and two arm workers where it was still owned by root. Unfortunately I didn't note down on which machines I chown'ed the log directory. Therefore now a more persistent approach with salt.

Actions #7

Updated by nicksinger over 3 years ago

postfix on openqaworker-arm-3 failed with:

Jan 29 01:42:20 openqaworker-arm-3 postfix[3476]: fatal: parameter inet_interfaces: no local interface found for ::1

I've enabled ipv6 on the loopback interface once again (on all 3 arm workers) by removing the corresponding line in /etc/sysctl.d/99-poo81198.conf while leaving the others disabled (see poo#81198)

Actions #8

Updated by nicksinger over 3 years ago

  • Status changed from Feedback to Resolved

Over the weekend we only collected fails of the openqa-worker service I asked about in . The main Issue (logrotate) and stuff mentioned in all seem to be fixed and the alert is enabled again. Therefore I'm finally closing this now as resolved.

Actions #9

Updated by okurz over 3 years ago

  • Status changed from Resolved to Feedback

hm, seems like you might have missed something in your queries: Same as mentioned in #88225#note-2 there still is iscsid.socket on openqaworker8. Maybe you just looked for ".service" lately?

Actions #10

Updated by nicksinger over 3 years ago

  • Status changed from Feedback to Resolved

Unfortunately ATM I can not reproduce the iscsi issue. But IMHO when it comes back it should be handled in a different ticket as the measures to resolve it might involve removing the service all together from our infrastructure (I don't see where it is needed anymore).


Also available in: Atom PDF