Coordinator state persistence

[benjamb]

As things are, restarting the coordinator will result in any locks or reservations that were being tracked to be lost (despite the fact that some of this state may have been written to disk (e.g. locks in places.yaml)).

While easy enough to avoid this by waiting for a period where no devices are in use, it would be ideal if I could upgrade to a new container image without causing this state to be lost entirely.

I see this topic was touched upon briefly as part of https://github.com/labgrid-project/labgrid/issues/293#issuecomment-507594892, but I'm unsure if this particular comment implies that this behaviour is by design.

[jluebbe]

This is by design, as during a coordinator restart we would miss all events from the exporters happening during the restart, which could lead to an inconsistent state on any client. Changing that without a different communication infrastructure seems difficult.

[Bastian-Krause]

@benjamb Is this something you strictly need? If not, I'd like to close this.

[benjamb]

@jluebbe Thanks for your response.

@Bastian-Krause This is not strictly required at this time.

[Bastian-Krause]

Okay, I'm closing this for now. If you consider working on a solution for this in the future, please reopen this issue beforehand, so we can discuss the design.