Do not stop the restore process after the "point of no return".

[pivanof]

Once vttablet shuts down MySQL and starts deleting all files, it won't be able
to recover if the context is cancelled (e.g. because the RPC got cancelled).
That will render the tablet completely unusable. To avoid that let's just not
interrupt the restore process once it passed that point of no return.

[michael-berlin]

Note that we already have poll_for_vars in test/utils.py.

After your change, this function sounds to be very similar to the other one? Maybe merge them?
(poll_for_vars is mostly used in worker.py.)

[pivanof]

It turned out that there's also utils.wait_for_tablet_type() that I can use instead of creating a new function. So I just reverted these changes and modified the test to use utils.wait_for_tablet_type().

[alainjobart]

For the record, the longer term fix we had talked about was to change the RPCs to a few RPCs:
XXXStart (starts the action in the background, returns an action UUID).
XXXStatus (streaming RPC that sends progress for a UUID).
XXXCancel (to really interrupt it, whatever the consequences).
and apply it for long-running operations, like Backup and Restore. The added bonus of that approach is to get a progress, and be able to display it in an automation UI (that can re-connect to a running action if restarting a workflow from a checkpoint, and the UUID was check-pointed).

Short of that, this approach looks good to me, after addressing Michael's comments.

[alainjobart]

LGTM