mongos currently applies the RetryableError label to any IngressRequestRateLimitExceeded errors it propagates back to the client, even ones it may have parsed from a remote response from the shards. This is a problem for two reasons:

• mongos may have already retried the request that produced this response a few times, and attaching the RetryableError label may multiply the total number of retry attempts, worsening overload.
• The request that failed might not be idempotent, and retrying might produce data corruption.

We should ensure that this label is only applied in the proper circumstances.