[theomega]

Very interesting article, thanks for publishing.

I have two related questions: 1. I assume the process which reads from Cassandra and puts it back to Redis is parallized if not even distributed. How do you ensure correctness? Implementing 2PC seems extreme overhead. Or do you lock in Redis? 2. What database is used to actually store the view counts? Cassandras Counters are afaik not very reliable...

[kchandra]

1. Redis is atomic, so we use the SETNX operation to ensure that only one write succeeds.

2. We have HLLs in Redis, so we just issue a PFCOUNT and store the result of that in Cassandra as an integer value. We don't use counters in Cassandra.