Software Engineering discussion

I was surprised at how fine-grained the consistency levels are in Cassandra. I wonder if these are actually all exercised in practice.

I also wonder how often the fault-tolerance mechanisms are actually triggered in practice. It would be interesting to see the root causes of NoSQL system failures in the field. With our Hadoop cluster, over the last 1/2 year, we have only had two failures. The first was due to too much job history kept around, causing the JobTracker to run out of Java heap space, which we only saw when students rushed to get homework in. This was fixed with an admin parameter change. The second was due to a networking configuration error. I think this data point, if experienced more widely, is a pretty impressive statement about the reliability of both the hardware and software.