Is it practical to use Mongo's capped collection a memory cache?

Question

Is it practical to use Mongo's capped collections as poor man's memcache assuming it will be kept in memory most of the time?

Answer 1

You can definitely use capped collections for this purpose. But it's important to know the basic limitations.

1. Data will expire based on insert order not time-to-expire
2. Data that is frequently accessed may still be pushed out of memory
3. _id columns are not defined on Capped Collections by default, you will need to ensure those indexes.
4. The objects cannot be modified in a way that changes the size of the object. For example, you could increment an existing integer, but you cannot add a field or modify a string value on the entry.
5. The capped collections cannot be sharded.

Because of #1 & #4 & #5, you are definitely losing some the core Memcache functionality.

There is a long-outstanding JIRA ticket for TTL-based capped collections which is probably exactly what you want.

Of course, the big question in this whole debate is "where is the extra RAM". Many people who use MongoDB as their primary store simply drop Memcache. If you have a bunch of extra RAM sitting around, why not just use it store the actual data instead of copies of that data?

Answer 2

Yes. This is a perfectly acceptable use of a MongoDB capped collection. I am assuming you will have many more reads than writes, so make sure you use an index as well.

MongoDB Capped Collections: Applications