Consider a configsvr stepdown occurs after having repaired all chunks, but before having notified the shards through _flushRoutingTableCacheUpdates. So if the command fails due to the stepdown, the router retries but this time the configsvr determines that all chunks have already been repaired, thus does not add them to the changedShardIds list and thus does not call _flushRoutingTableCacheUpdates on the shards. This causes shards to not repair their chunk 'history' on config.cache.chunks.<nss>