It has been discovered a case where a stale router running an updateOne without a shard-key can return internal transient errors (related to the internal transaction used) instead of automatically retrying

[js_test:repro]   "errmsg" : "Write results unavailable from failing to target a host in the shard repro-rs0 :: caused by :: Command error committing internal transaction :: caused by :: The collection was dropped", 

I left attached a repro, but the overall flow can be summarized as:

• 1) On mongos0 create and shard collection
• 2) On mongos1 drop collection.
• 3) On mongos1 recreate the collection as unsharded and insert one document.
• 4) On mongos0 (stale, still thinks collection is sharded) run updateOne without shard key.

The updateOne returns the error above