canonical/lxd

Clean up after cluster join errors

masnax opened this issue · 0 comments

#12571 addressed an issue where dqlite would report back that it was busy, and LXD would error out its attempt to join a cluster. This was a problem in the first place because LXD didn't handle an error at this point very well, and would leave the joining node unrecoverable.

What we should do here is include some revert hooks to clean up a failed attempt at joining a cluster so that the joiner returns back to a state where it can be requested to join a cluster or be initialized.