clarify guarantees around sensor cursors and failure recovery

[dagsterbot]

Dagster Documentation Gap

This issue was generated from the slack conversation at: https://dagster.slack.com/archives/C01U954MEER/p1646426665408129?thread_ts=1646426665.408129&cid=C01U954MEER

Conversation excerpt

U032Z1Y5L7R: 👋 Question regarding using cursors with sensors as described here:
https://docs.dagster.io/concepts/partitions-schedules-sensors/sensors#sensor-optimizations-using-cursors
What are the transactional guarantees around update_cursor() and yielded run requests? Is there anything we need to watch out for that could result in missed runs while using cursors (e.g. the cursor is updated but the run never actually happens)? Or is it guaranteed that the cursor update will never be recorded without the run request also?
U016C4E5CP8: Hey Danny - the cursor isn't updated until after the end of the tick that creates the runs (so calling update_cursor doesn't immediately update the cursor right there, it's passed back through to the daemon and written once its gone ahead and launched the runs)
U032Z1Y5L7R: Perfect. I figured that was the case but didn’t see anything that explicitly defined the semantics and didn’t want to assume and silently drop data. Thanks, <@U016C4E5CP8>.
U016C4E5CP8: we should make that clearer, yeah. <@U018K0G2Y85> docs clarify guarantees around sensor cursors and failure recovery

---

Message from the maintainers:

Are you looking for the same documentation content? Give it a 👍. We factor engagement into prioritization.