Hi all. We noticed some unexpected behavior with cloud events where details were missing for non-successful phase transitions. It looks like this code block is the cause. We were specifically looking for the launch plan ID. Was this an optimization done to reduce load on the DB? If so, this looks like a good candidate for a read replica. Are folks open changing the behavior or making it configurable?

cc @thankful-minister-83577 since it related to artifact

Still curious if the reason is db pressure because Flyte may benefit from utilizing read replicas. Not sure it does yet

the reason was mainly due to necessity. for our purposes, we only needed it once, on successful termination of the node, so it felt like it was a waste to send it for all the other phases. db load was a concern, as was the handling of a read replica.

I see. For our use case we specifically care about non-successful state transitions so I might add a PR to make it configurable. With regards to using the read replica I can see replication lag being an issue. I'll see how/if the DB load has changed for us and re-evaluate

yeah it may just work - i think most (all?) of the information it uses is long term information, things like the task template and such, it should already be there. if there’s any recent information though, then that may well be missing.