Ceph

@Loïc Dachary: Which two systemd units are you referring to?

@Nathan Weinberg: He is talking about ceph-disk@sdX1.service and ceph-disk@sdX2.service for data/journal.

Boris Ranto writes:

This could be fixed if we added 1 as an acceptable exit code for ceph-disk command, e.g. adding this line to the service file for ceph-disk:

SuccessExitStatus=1

We'd be just ignoring the error, though.

That would be fine with me (forcing the exit code to always be a success). The problem is that it will also hide real errors. Some would say that there are so many level of indirections when / if ceph-disk prepare fails that it won't really make a difference (ceph-disk prepare creates a partition -> fires a udev event -> it triggers a systemd unit -> it runs ceph-disk activate fails because the journal is not there yet -> ceph-disk prepare creates the journal partition -> fires a udev event -> it triggers the systemd unit -> it runs ceph-disk activate does the job).

It's a judgement call really and I'm not sure what's best.

Yeah, I was not too thrilled about it either. We might want to reserve a special exit code for saying 'not all devices ready' and ignore just that one?

btw: If journal and data are on the same drive, this will be ok after reboot as both devices show up at pretty much the same time. However, it is still reproducible when installing the cluster as you first create and activate the journal partition and only then, few seconds later, you create the data partition. It would help a bit to create both partitions at the same time (or tell kernel to activate them at the same time, not one by one).