You are browsing a read-only backup copy of Wikitech. The live site can be found at wikitech.wikimedia.org

Portal:Cloud VPS/Admin/Runbooks/Check for VMs leaked by the nova-fullstack test

From Wikitech-static
< Portal:Cloud VPS‎ | Admin‎ | Runbooks
Revision as of 16:14, 8 November 2021 by imported>Andrew Bogott (Added section about intermittent test failures)
Jump to navigation Jump to search
The procedures in this runbook require admin permissions to complete.

Error / Incident

Usually an email/icinga alert with the subject ** PROBLEM alert - <hostname>/Check for VMs leaked by the nova-fullstack test is CRITICAL **

This happens when there's an error in the creation and/or deletion of the nova-fullstack VM. To see the logs, you have to login to the instance that triggered the alert (if you are looking at the https://alerts.wikimedia.org, it is shown as a blue tag). For example cloudcontrol1003:

Debugging

Quick check

 ssh cloudcontrol1003.wikimedia.org

There you can check the systemctl status: 

 dcaro@cloudcontrol1003:~$ sudo systemctl status nova-fullstack.service

It might be that the service re-triggered since then and now it's running. Logs are in Kibana: https://logstash.wikimedia.org/app/dashboards#/view/13149940-ff6c-11eb-85b7-9d1831ce7631

Another possible source of information is the Grafana dasboard.[1]

Debugging service errors

TODO: Add here any service errors you encountered and how you fixed them.

Per-VM debugging

If there's nothing in the logs (sometimes the log has been rotated), you can check the servers for the admin-monitoring project: 

 dcaro@cloudcontrol1003:~$ sudo wmcs-openstack --os-project-id admin-monitoring server list
 +--------------------------------------+---------------------------+--------+----------------------------------------+--------------------------------------------+-----------------------+
 | ID                                   | Name                      | Status | Networks                               | Image                                      | Flavor                |
 +--------------------------------------+---------------------------+--------+----------------------------------------+--------------------------------------------+-----------------------+
 | d603b2e0-7b8b-462f-b74d-c782c2d34fea | fullstackd-20210110160929 | BUILD  |                                        | debian-10.0-buster (deprecated 2021-02-22) |                       |
 | 33766360-bbbe-4bef-8294-65fca6722e20 | fullstackd-20210415002301 | ACTIVE | lan-flat-cloudinstances2b=172.16.4.230 | debian-10.0-buster                         | g3.cores1.ram2.disk20 |
 | 697ebb69-0394-4e29-82fc-530153a38a1b | fullstackd-20210414162903 | ACTIVE | lan-flat-cloudinstances2b=172.16.5.251 | debian-10.0-buster                         | g3.cores1.ram2.disk20 |
 | 1812e03b-c978-43a5-a07e-6e3e240a9bf0 | fullstackd-20210414123145 | ACTIVE | lan-flat-cloudinstances2b=172.16.4.184 | debian-10.0-buster                         | g3.cores1.ram2.disk20 |
 | d70a111d-6f23-40e1-8c01-846dedb5f2ca | fullstackd-20210414110500 | ACTIVE | lan-flat-cloudinstances2b=172.16.2.117 | debian-10.0-buster                         | g3.cores1.ram2.disk20 |
 | 0b476a1a-a75e-4b56-bf51-8f9d43ec9201 | fullstackd-20210413182752 | ACTIVE | lan-flat-cloudinstances2b=172.16.4.198 | debian-10.0-buster                         | g3.cores1.ram2.disk20 |
 +--------------------------------------+---------------------------+--------+----------------------------------------+--------------------------------------------+-----------------------+

There we see that there's a few instances stuck from April 14th, and one from the 13th, and a building instance from January.

Very old failed instances

Usually the best of action there is just to delete the server, as any traces will be lost already, in this case fullstackd-20210110160929. 

 sudo wmcs-openstack --os-project-id admin-monitoring server delete fullstackd-20210110160929

If it replies with: 

 No server with a name or ID of 'fullstackd-20210110160929' exists.

Then you have a stuck server entry, follow this: Portal:Cloud VPS/Admin/Troubleshooting#Instance Troubleshooting

In this case, it was an old build request that got lost (quite uncommon): Portal:Cloud VPS/Admin/Troubleshooting#Deleting an orphaned build request

Steady, intermittent failures

Sometimes the fullstack test will exhibit a slow leak, with one failed VM every few hours or every few days. That suggests an intermittent failure. Here are some things that have caused that in the past:

  • Race conditions in the instance startup script, usually related to getting clean puppet runs (puppet/modules/openstack/templates/nova/vendordata.txt.erb). This usually shows up in the cloud-init logs on the failed instances, /var/log/cloud-init*.
  • Partial outage of OpenStack services, resulting either from one service on one cloudcontrol being out, or an haproxy configuration error. First place to look for that is in /var/log/haproxy/haproxy.log
  • Bad tokens in a service cache, causing intermittent auth failures between openstack services. To clear the caches, restart memcached on all cloudcontrols.
New instances

Now we can dig a bit deeper, and check for openstack logs containing those instance ids, let's pick 697ebb69-0394-4e29-82fc-530153a38a1b You can now go to kibana[2] and search for any openstack entry that has that server id in it, for example:

https://logstash.wikimedia.org/goto/33a4ec64e36b40d00fc000ca37adb130

Remember to set the time span accordingly to the instance creation date

In this case there seems not to be anything interesting there for that instance.

Going to the new instance

As a last resort, we can try sshing to the instance and running puppet, see if anything is broken there: 

 ssh fullstackd-20210415002301.admin-monitoring.eqiad1.wikimedia.cloud
 dcaro@fullstackd-20210415002301:~$ sudo run-puppet-agent
  Info: Using configured environment 'production'
  Info: Retrieving pluginfacts
  Info: Retrieving plugin
  Info: Retrieving locales
  Info: Loading facts
  Info: Caching catalog for fullstackd-20210415002301.admin-monitoring.eqiad1.wikimedia.cloud
  Info: Applying configuration version '(dd0bf90505) Manuel Arostegui - mariadb: Productionzie db1182'
  Notice: The LDAP client stack for this host is: sssd/sudo
  Notice: /Stage[main]/Profile::Ldap::Client::Labs/Notify[LDAP client stack]/message: defined 'message' as 'The LDAP client stack for this host is: sssd/sudo'
  Notice: Applied catalog in 5.74 seconds

In this case nothing failed.

Cleanup

As we did not find anything in the logs (some might be lost already), and the service is running ok currently, all that's left to do is to cleanup the VMs, see if deleting them fails, and freeing up the resources if not. 

 dcaro@cloudcontrol1003:~$ sudo wmcs-openstack --os-project-id admin-monitoring server delete 33766360-bbbe-4bef-8294-65fca6722e20

Everything worked, so repeat with the other servers.

Support contacts

Usually anyone in the WMCS team should be able to help/debug the issue, the main subject matter expert would be Andrew Bogott.

Related information

References

  1. ↑ Nova fullstack Grafana dashboard
  2. ↑ Wikimedia Kibana
Retrieved from "https://wikitech-static.wikimedia.org/w/index.php?title=Portal:Cloud_VPS/Admin/Runbooks/Check_for_VMs_leaked_by_the_nova-fullstack_test&oldid=584713"