Because profiling takes a context for the namespace of the operation and write commands do not use setShardVersion, it's possible for a write to be successful but profiling to throw a stale config error. This aborts the rest of the batch.

Issue probably exists in previous versions of mongodb when the database lock of a write is released before profiling, but would require a version change during that interval in order to manifest.