|
Currently, we send WriteWithoutShardKeyWithId in their own batch when ordered=false. This may cause an overall increase in latency due to several requests waiting to be processed serially as we increase the throughput of client requests, when compared to the performance with UpdateOneWithoutShardKeyWithId feature flag off.
One solution here will require some code refactoring and promotion of non-retryable writes to retryable writes, enabling us to retry with idempotency in the case of stale shard/db version errors.
|