Is there documentation somewhere on how to upgrade the dagster daemon? How are currently-running jobs handled, etc?

Hi Duncan, not sure there's a doc on this specifically - if you're using any run launcher other than the default one, the runs don't actually happen as part of the daemon process, so you should be able to shut down the daemon without interrupting any ongoing runs. Schedules and sensors (and the run queue) should also pick up where they left off once the daemon is back up and running.

I see. Do the runs not need to communicate with the daemon in order to signal that they have completed and record that fact?

The way that they communicate that back to the daemon is by writing updates to the database which the daemon then reads to figure out what to do next - there's no direct communication

So that’s done in-process with the user code, on the target that is doing the running?

Each run writes events and updates to the database within the run process, yeah

I see. That’s smart.

schedules for partitioned jobs will catch up on any partitions that were missed (a mini-backfill, like you said). Schedules for jobs that are not on partitions will just wait for the next schedule time to happen

Cool. Tangentially: what happens if the jobs can’t reach the database (perhaps you need downtime for an upgrade).

In that case I believe the jobs would fail - pause and retry would be nice, but not a feature we have yet

Right