|
we are running mongo cluster with 5 shards, after several month of production, we find it didn't split chunks in proper ranges, some chunks never contain data while others are very heavy. so we decided to apply pre-split on one of online database. as the range of shard key is fixed, so we choose 4 moderate points as middle point. we didn't stop balancer when doing pre-split. and we think this is some what easy and non fatal.
but when we start writing to these empty and pre-splitted collections, the were still trying to autosplit, and we got "cannot find range min to max" error message. this may lead to slow insert performance and last about 8 hours. after the message "time to load chunks ", these collections start to auto split successfully. and we lost some data because of slow insert.
so, I want to know are there some essential step did i miss after pre-split? or when does "time to load chunks" happens?
|