#951 unwraps single `TaskGroup` failures in parallel reader Pull request opened by samhita-alla • This changes

ObstoreParallelReader

to surface the real underlying exception when a Python 3.11+

asyncio.TaskGroup

fails with exactly one sub-exception. Previously, a single worker failure during parallel downloads could bubble up as

ExceptionGroup: unhandled errors in a TaskGroup

which obscured the actual storage error in runtime logs and tracebacks. • Broadens worker-side exception logging. • Fixes a pre-3.11 bug: it waited only on

outq.get()

, so if a worker failed before queueing a result, the consumer could block forever. Changed that branch to wait on both queued results and fill/worker task completion. flyteorg/flyte-sdk