Machine rebooted though we configured reboot-strategy: off

[zeisss]

Hi

this morning one of our nodes rebooted and updated from 402 to 457 with a reboot.

We have disabled reboots in the cloud-init though:

#cloud-config

coreos:
  update:
    reboot-strategy: off

  <more config>

/etc/coreos/update.conf says the same:

# cat /etc/coreos/update.conf
GROUP=alpha
REBOOT_STRATEGY=off

[sorccu]

The reboot might have been caused by something else. Once an update has been applied, any reboot will activate it.

[zeisss]

Sounds reasonable, as reading the update-engines log doesn't clearly state that it rebooted the machine.
Sadly I have already destroyed the cluster so I cannot check any other logs.

Can we savely disable update-engine.service to be reboot secure?

[crawford]

Can you post the update-engine logs?

Masking update-engine.service will disable automatic updates. It obviously won't protect the machine from rebooting due to other mechanisms. Why do you want to disable updates?

[zeisss]

The logs I could get from journalctl can be found at https://gist.github.com/ZeissS/c4f7f624c8fd2366b174 - the reboot occured at 8:41.

It obviously won't protect the machine from rebooting due to other mechanisms.

Disabling rebooting is not the goal - changing the software is. Neither etcd nor fleet is currently that stable yet so we prefer pinning coreos to a specific release.

Why do you want to disable updates?

Well, coreos just destroyed our cluster with that reboot ^^ Sounds like reason enough for me to (not yet) trust it (we ran into etcd-io/etcd#815 (comment)).

[sorccu]

Well, I can say from personal experience that using alpha is a good way to occasionally destroy your cluster anyway. Instead of disabling updates, why not try sticking to stable (now that it's even newer than what you started with)?

[zeisss]

We definitely do this for the non-testing clusters ;) We still prefer for now to stick to a specific release.

Since disable update-engine is the way to go to prevent updates happening even on reboot and there does not seem to any reason to believe update-engine is the source of the reboot I am closing this now.

Thanks guys!