Sharding and Failover

Failure of a mongod process within a shard (PLANNED)

If a member of a replica set in a shard is down, read and write operations are still permitted.  Replica sets functionality is pending: SERVER-557 .

Failure of all mongod processes within a shard

If all replicas within a shard are down, the data within that shard is (at least temporarily) unavailable.  Queries which can be resolved at other shards continue to work properly.  Queries that require the down shard will return an error.

Failure of a config server

When one of the config servers is down, split and migrate operations do not happen.  Other aspects of the system including conventional reads and writes continue properly.  It is important that the down config server be repaired by the administrator in a reasonable period of time (say, 1 day) so that shards do not become unbalanced due to lack of migrates (which may or may not happen in practice).  Note if repairing the down config server will be difficult, it could simply be retired and removed from the cluster; another can be added back later.