Our analytics department is redesigning our real-time streaming pipelines. When evaluating wide-column storage engines for time-series metrics, which database would you never use again because of configuration complexity? Is Apache Cassandra worth the operational headache for tracking high-volume datasets?
3 answers
Apache Cassandra is a technology I would heavily avoid deploying again unless I had a dedicated infrastructure team. The query model is highly restrictive because you must design your tables precisely around your specific output queries. If the analytics team suddenly requires a new data breakdown, you often have to build entirely new tables and duplicate your big data pipelines. Furthermore, managing the Java Virtual Machine garbage collection configurations on large production clusters takes constant manual intervention. Modern managed cloud engines give you the same distributed scaling advantages without the painful open-source management burden.
Do hosted managed variants of Cassandra eliminate these painful infrastructure tuning bottlenecks completely, or do the structural query restrictions still cripple team development velocity?
Cassandra requires too much administrative overhead. The constant tombstone compaction issues and manual cluster ring rebalancing will drain your engineering hours.
I agree with Keith. The tombstone buildup from frequent data updates completely ruined our read latencies until we manually rewrote our entire data deletion strategy.
Raymond, hosting providers handle the server provisioning, but they cannot fix the restrictive query model. Your data scientists will still get frustrated by the total lack of ad-hoc relational joins, which severely slows down unexpected data analysis requests.