Hi Community, I have a question about the task retry mechanism in Flyte Connectors. We are migrating our Flyte workflows from Databricks Plugin to Databricks Connector. In Databricks Plugin, we mapped task failures to

PhaseInfoRetryableFailure

task phase (see Issue #5210) , which enabled us to use the Flyte task retry mechanism for Databricks tasks. Unfortunately, I could not find an equivalent task execution phase in Databricks Connector. Is there a managed solution for retrying failed tasks executed by connectors? @full-toddler-5766

You can update the connector to retry

@freezing-airport-6809 Are you suggesting that the entire retry logic be moved to the connector level? Retry sounds like a common functionality to me. I assumed a common retry logic was implemented in

flytepropeller

, connectors simply feed propeller with the proper task phase and propeller manages the retry attempts. Am I wrong? Would it still be possible to track the retry attempts on the Flyte task if we moved the entire retry logic to the connector level?

No

yeah, that's fine

We will tal and let you know

I had the same idea, but I could not find any TaskExecutionPhase here that could be used for mapping retryable failures. Maybe I missed something.

the retryable failures exist because we use flyteadmin's state before to decide what phase to return in the databricks plugin

oh, I see

this is a good question, I think its worth to add, but need to think about how the whole system will be implemented, or will this affect anything?

all right, should I open a GitHub issue for this?

yes plz

ok, thanks for your quick feedback. will raise an issue