In May 2019, the DataStax Accelerate conference, organized by DataStax, the manufacturer of the commercial Cassandra offshoot, provided a number of interesting practical application examples. Yahoo Japan, for example, reported on how its Cassandra cluster for apps, logs, and statistical data grew from 100 nodes in 2016 to 4,700 nodes in 170 clusters today. However, operating a large installation of this nature also causes problems, especially with regard to the repair processes required after errors and failures that often broke down or got stuck in infinite loops. Without further ado, they developed their own nodetool repair command, which was able to reduce the repair time from 970 to 67 minutes.
The Cassandra operators at Instagram had different problems. In particular, the latencies in their cluster, geographically distributed over continents, caused…