On Fri, Jun 06, 2014 at 01:29:44PM +0200, Michael Scherer wrote:
Due to CVE on openssl and on kernel, I did upgrade various piece of the infrastructure ( foreman, lists, stats, monitoring ), which implied a few reboots ( due to kernel lagging behind, which is not that great with local root exploit ). As this is friday and I assumed most of the Tel Aviv office was not working, i hope this kept the disruption to a minimum. However, if something is broken, please tell it so we can fix.
Nice work.
This also got me thinking. In order to bring a bit more order, what about having a fixed schedule for upgrade ?
In my previous position, we were doing that once per month ( except during end of quarter freeze ), with mandatory reboot ( cause if something do not boot, you want to know it when you have a planned outage, not when everyone is running around updating stuff ). Fedora has a rather complex procedure to decide what to upgrade, hilighted on http://infrastructure.fedoraproject.org/infra/docs/massupgrade.txt
At my previous employer we had something similar. I also wrote a puppet custom fact reboot_needed which checks if the running kernel matches the default kernel that would be booted.
So we could adopt a schedule ( once per month, unless there is something critical, in which case we do it ASAP, with warning on the list and irc ).
+1
The schedule should of course take in account "business need", which is "release schedule of ovirt".
So what about "first friday of the month, unless exception" ?
We should also make sure that we don't reboot ALL servers at once. So if we have multiple centos 6 jenkins slaves, try to just reboot one at a time. Also would be nice if the slave did proper scheduling in jenkins so no jobs are running.
And by update, i mean "yum upgrade -y". Cleaning the list of repo on various servers is also IMHO another task to discuss, to make sure the task can be safely executed. ( having something like mcollective/ansible/func is also needed, but that's more a convenience than a requirement at this stage ).
We sometimes have pinned versions on jenkins build slaves. That means we should either do a proper yum versionlock or find something else. Note that I'm all in favor of being able to to a blind yum upgrades.