A primary shard failover during unshardCollection after the commit decision is persisted can cause DDL locks to be released prematurely, allowing concurrent DDL operations to interfere with resharding cleanup.

More details here. 

The authoritative configsvr should handle conflicts or joins for existing resharding operations. Update the reshardCollectionCoordinator to forward the _configsvrReshardCollection command to the config server and let it handle the case, instead of having unshardCollection take the early return path.