Potential race condition in shard map updating

If the mapping query returns an error on all databases, there's a possibility that the shard data structures are written to and read concurrently. Since the reading and writing of it is not thread-safe, there is a potential race condition that leads to undefined behavior.

In practice this is unlikely to happen as the query will never fail due to anything in the database and the only way for it to happen is for the connection to break. However, it is possible to construct a configuration where a query will end up producing a result (e.g. using a regexfilter to cause syntax errors).