Open hannes-ucsc opened 1 year ago
The workaround is to recreate the instance.
Nothing we can do here. There isn't even an upstream issue to block this on.
https://github.com/canonical/cloud-init/issues/3386 is the upstream blocker
Assignee to monitor upstream blocker.
After a regularly scheduled GitLab version upgrade, the instance came back without mounting the data volume at
/mnt/gitlab
. The GitLab application the started with vanilla configuration and data, rendering it essentially dysfunctional.It seems that when cloud-init attempted to create the /etc/fstab entry for the volume, the corresponding device node in
/dev
hadn't yet been created by the kernel. When I later logged in, the device node did exist. So this sounds like a race in cloud-init, the kernel or EC2. Rebooting the instance didn't help because I think that part of cloud-init is once-per-instance, not once-per-boot.Compare that to a normal log