Search code examples
azure-cosmosdbpartition

Partition key for DocumentDB

I have a question about DocumentDB partition key choise. I have data with UserId, DeviceId and WhateverId. UserId parameter will be in queries always, so I have chosen UserId as a partition key. But I have a lot of data for one user (millions of entities) and when I made a quety like "SELECT * FROM c WHERE c.DeviceId = @DeviceId" with partition key specified it takes a lot of time(about 6 minutes for about 220 000 returned entities). Maybe it would be more efficient to choose for example DeviceId as a partition key and make queries against a few partitions in parallel (specifying EnableCrossPartitionQuery = true and MaxDegreeOfParallelism = partition count)? Or maybe it is a good idea to use separate collection for every user?

Solution

  • It might help a little but I don't think a partition for each user will solve your problem because you essentially have that under the covers.

    You could experiment with the partition key to improve the parrallism but, at best that would give you 2x to 5x improvement in my experience. Is that enough?

    For more dramatic improvements you usually have to resort to selective denormalization and/or caching.