shunping opened a new pull request, #38472: URL: https://github.com/apache/beam/pull/38472
In `executePipeline`, bundle execution goroutines previously used the parent job context instead of the errgroup context (`egctx`). When a bundle failed, `egctx` was cancelled but other concurrent bundle execution goroutines did not detect cancellation. This caused `eg.Wait()` to block indefinitely, preventing Prism from reporting `JobState_FAILED` promptly and leading to 60-second gRPC `DEADLINE_EXCEEDED` errors on the client side. With the proposed fix, the following steps will be triggered to ensure both the server and the client exit gracefully. 1. As soon as any bundle fails, `eg` cancels `egctx`. 1. All other concurrent bundle execution goroutines immediately detect `<-egctx.Done()` and exit. 1. `eg.Wait()` unblocks immediately, allowing `RunPipeline` to call `j.Failed(err)`. 1. Prism immediately broadcasts `JobState_FAILED` over `GetStateStream`, and the Python SDK client gracefully receives and raises the underlying worker exception without encountering a deadline exceeded error. -- 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]
