Include the URL of your launchpad blueprint:
Currently we enforce locking at the stack level, so only one operation can be in progress at a time on a stack. This is not fine-grained enough, as it prevents us from starting a new update while awaiting the result of a previous one. Phase one of convergence is to remove this restriction by locking at the level of individual resources.
Make rows in the Resource table lockable, by ensuring that state changes are atomic. We’ll also need to store the ID of the engine that currently holds the lock, so that we can use this to detect when an engine has died and clean up appropriately.
We’ll use the “UPDATE ... WHERE ...” form discussed in http://www.joinfu.com/2015/01/understanding-reservations-concurrency-locking-in-nova/ to ensure atomic updates.
The existing StackLock code does almost exactly what we want already, but the downside is that it uses a separate table in the database to do so. Using that rather than applying new semantics to the writes we are already doing would make convergence even more database-intensive than it already is.