Mongodb count with queries have poor performance

Question

Counting the number of records in a collection that is matched by a query even on an indexed field takes too much time. For example lets say there is a collection consists of 10000 records, and there is an index on the creationDate field of this collection. Getting the last ten records from the collection is faster than counting the number of records created on the last day. It takes more than 5 seconds, sometimes even up to 70 seconds to return the result of the count query. Do you have any idea how to solve this problem, what is the best way to solve this issue?

Btw we also use morphia, and we saw that getting the count through morphia is even slower, so for count queries, we transform the morphia query to the java driver query. Did anyone encounter a similar situation, why does morphia response even slower? Does this happen only for count queries or is it slow in general compared to using only java driver?

Help, suggestions or work-arounds would be really appreciated, our application relies heavily on count queries and the slowness of system is really annoying for us right now.

Thanks in advance.

Answer 1

Wait for a fix.

As Asya Kamsky commented out, count performance is really bad on 2.2. The only workaround we've found is to avoid them as much as possible.

(There are other things that are unexplainably slow with mongodb - such as aggregate queries - most of those have associated JIRA issues and are being worked on/scheduled)

Answer 2

While this might not be the final answer, let's get started and evolve this further:

1. Your indexes should always fit into RAM, otherwise you will get really bad performance.

2. To evaluate how much RAM is used, you can either use 10gen's MMS or check with various tools. For a description plus possible reasons for low (resident) memory usage see http://www.kchodorow.com/blog/2012/05/10/thursday-5-diagnosing-high-readahead/. Or you simply haven't accessed enough data yet in which case you can use MongoDB's touch (but I doubt that since you're already having performance issues).

Besides adding RAM and making sure you use all the available RAM, you could also drop unused indexes or use compound indexes where possible.