The asset reconciliation sensor has materialized a downstream asset twice. Why can this happen?

ah yep -- I'm guessing that screenshot is showing all runs for the instance (i.e. there's no other materialization that might have happened in between those two runs)? definitely seems like a bug. do you mind checking how long the earlier run took to execute? also things like when it got dequeued etc. theoretically, the sensor should take in-progress runs into account, but it's possible that something is going wrong there (as a first guess)

Daily partitioned source assets -> inference_data (daily partitioned) Non-partitioned source asset + inference_data -> predictions (daily partitioned)

ah thanks for the fast response! that does seem to align roughly with the theory that the in-progress run was not being handled properly (as it seems like the second materialization of

predictions

was kicked off right after

inference_data

completed in the original run, rather than waiting for the original run to complete)

So by the time the second run got scheduled predictions in the first run didn’t start yet

just to confirm, which version of dagster are you on?

1.2.7

and is the timestamp of the materialization of

inference_data

in the first run ~2:14AM?

image.png

great thank you -- this gives me a lot to work with 🙏. I'll look into it and get back to you

🙂 thank you too!

ah, I see what's going on here. it's an easy fix but I think it'll need to wait until next week as this week's release is currently being rolled out 😞

Ok, thanks fine. Thank you for the fix!

that's accurate haha -- the main driving factor behind the complexity is performance (simple solutions end up killing the database or choking on large asset graphs blob smile sweat)