Hacker News Reader

Interesting that they're not using EBS to provide durable storage for Cassandra, but instead using S3 (along with S3-backed, or "ephemeral storage", AMIs). I wonder if that means they're batching up and writing their database logs to S3, plus running enough instances across AZs that it's generally okay to keep everything in memory even when an instance or two fails.

Anyone have any experience running a NoSQL datastore in this fashion?

View on HN

They're probably using the local ephemeral drives for Cassandra storage rather than S3. I'm guessing they're then moving snapshots into S3 or elsewhere.

http://www.mail-archive.com/user@cassandra.apache.org/msg110...

That's exactly what we do.

Keep in mind that these ain't banking records. If a queue reordering occasionally fails or a rating doesn't save, it's probably ok.

Reply on news.ycombinator.com