In the Shard class, we have a runAggregation function that starts a aggregation process. The retryability was implemented with the fact that getMore was not retryable. We should also have retryability at the fetcher level. This would allow to properly retry on retryable errors, while handling other classes of error in Shard.

We just have to be careful stripping the RetryableError label from the fetcher as the retries have already been performed.

We should also consider simply switching to the ARS for the shard class as noted in the technical design of SPM-4003