Track instance state during slave attach

[Iccha Sethi]

Currently what happens is that the slave goes to state active, and then the slave attach takes place. If the later fails, the instance state changes from ACTIVE to FAILED. This is confusing customer experience seeing their active instance going to failed state. We must set instance to active state only if attach succeeds.

[Nikhil Manchanda (slicknik)]

Any updates on this?

[Peter Stachowski (peterstac)]

I thought this was going to be fixed with replication v2, however it still seems to be an issue (even though attaching the slave is now part of the create process). I believe it's because it's an asynchronous call. The only way to not have it flip to ACTIVE first is to actually wait until it either finishes or times out. That change might be trivial (I think it will be, but could be wrong). I'll take a look at it from that angle (hopefully within the next day or two).

[Amrith Kumar (amrith)]

any word on this Peter?

[Peter Stachowski (peterstac)]

I think this was fixed with the manager refactor in the mitaka cycle. At least I haven't noticed the issue since, so I'll mark it resolved and we can open a new ticket if it happens again.

[Amrith Kumar (amrith)]

thanks peter, I'm setting this up to expire in 60 days.

[Launchpad Janitor (janitor)]

[Expired for OpenStack DBaaS (Trove) because there has been no activity for 60 days.]