How to handle secondary keys in partitioning?

[John Smith]

Say I have a table "Message" and I want to partition it by "recipient_id".

That way when I want to get all messages sent to a certain user, I only need to query a single partition.

But sometimes I also want to query for all messages sent by a certain user. In that case I'd need to send query to every single partition and then merge the results before sorting them.

Is that going to be very slow (much slower than without using partition) because I'd need to run the query N times (N being the number of partitions) and I can't use index in sorting the final results?

So does it make sense to use partition in a scenario like this where I have secondary indexes & queries that are "cross-partition"?

And what if I want to use Spider Engine to turn the partitions into separate DB shards? Would these cross-shard queries (such as querying for messages from a certain sender) be even slower?

This must be a pretty common scenario. Is there any tricks in dealing with cross-partition queries/indexes like this?

Thanks.