Automatic patching saves a lot of manual work, but updates can also restart services or change behavior unexpectedly. Do you let production servers update themselves, stage patches first, or review everything manually?
How much do you trust automatic updates on production servers?
Answers
Automate the rollout, but don't make production the test bed. Patch a test environment that's as close to production as practical, run functional and regression checks, then promote the same updates after a set delay. If a check fails, stop and investigate.
A weekly schedule can work for some hosts, while others might need a slower cadence. The key is knowing which packages were applied and having monitoring that catches trouble during and after deployment.
Reply to this answer
For a large fleet, doing every update by hand creates its own problems. Unattended security and regular updates can be a reasonable choice if the alternative is machines going unpatched for long stretches.
That doesn't mean pretending packages never break. Decide which repositories you trust, keep an eye on update results, and have a plan for failures involving things like the bootloader or storage stack.
Reply to this answer
I'd split security fixes from bigger version changes. Apply security updates automatically if the risk makes sense for the service, then gate feature and version upgrades through testing.
Some distributions backport security fixes to existing package versions, which can reduce the amount of change involved. Still check what your distribution actually ships. A security update isn't a guarantee that nothing will behave differently.
Reply to this answer
I prefer scheduled automation over completely unattended updates. Use a patch job that logs what happened, coordinates reboots, and reports failures. For virtual machines, a snapshot before patching may help with recovery, provided you've checked what it covers and how to restore it.
Monitoring maintenance mode can cut down on noisy alerts during the window, but make sure checks are active again afterward.
Reply to this answer
No automatic production installs for systems where a surprise restart or failed boot would be hard to recover from. Review and test the updates first, then have someone approve the rollout. That's slower, but it gives the team a chance to check the change and recovery plan before anything touches production.
Reply to this answer
Add your answer
Answers are reviewed before publication.
_f39+drummarker.5:
How do you handle servers that don't have a matching test environment?