Here's a breakdown of what occurred in the build failure where this appears to manifest:

1. Shard sends _configsvrCommitChunkSplit to config server.
2. Config server finish executing the command.
3. Config server steps down from primary and closes all incoming connections.
4. Shard gets back a network error and retries.
5. Shard sends the commit command and get a preCondition failure because it has already been successfully applied earlier.
6. Shard tries to inspect the state by refreshing the chunk state (by querying) but gets back the state before the commit.