Cassandra count use case

Question

I'm trying to figure out an appropriate use case for Casandra's counter functionality. I thought of a situation and I was wondering if this would be feasible. I'm not quite sure because I'm still experimenting with Cassandra so any advice would be appreciated.

Lets say you had a small video service, you record the log of views in Cassandra while recording what video was played, which user played it, country, referer etc. You obviously want to show a count of how many times that video was played would incrementing a counter every time you insert a play event be a good solution to this? Or would there be a better alternative. Counting all the events on read every time would take a pretty big performance hit and even if you cached the results the cache would be invalidated pretty quickly if you had a busy site.

Any advice would be appreciated!

Answer 1

Counters can be used for whatever you need to count within an application -- both "frontend" data and "backend" one. I personally use them to store user's behaviour information (for backend analysis) and frontend ratings (each operation a user do in my platform give to the user some points). There is no real limitation on use case -- the limitation is given by few technical limitations, the bigger coming to my mind:

1. a counter cf can be made only by counters columns (except PK, obviously)
2. counters can't be reset: to set 0 value to a counter you need to read and calculate before writing (with no guarantee about the fact that someone else updated before you)
3. no ttl and no indexing/deletion

As far as your video service it all depends on how you choose to model data -- if you find a valid model to hit few partitions each time you write/read and you have a good key distribution I don't see any real problem in its implementation.

btw: you tagged Cassandra 2.0 but if you have to use counters you should think about 2.1 for the reasons described here