heyihong commented on PR #54094: URL: https://github.com/apache/spark/pull/54094#issuecomment-3891074300
> I need a bit more context. What's the current protobuf protocol for observed metrics and how this PR changes it? Yes, below is the summary of the `Success path`: 1. The driver runs the plan; observations produce Rows (or the Python accumulator produces a seq of values). 2. The server converts those to ObservedMetrics: name, keys, values (each value as Expression.Literal). 3. The result is sent as ExecutePlanResponse.observed_metrics (repeated). 4. The client treats each item as success: keys and values have the same length; it decodes each Literal and builds either a Row (Scala) or PlanObservedMetrics / pairs (Python). For this PR, success path is unchanged; it only adds reporting of observation collection failures via root_error_idx and errors, and surfaces those failures on both Scala and Python clients. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
