The tradeoff between online and offline optimization is really interesting, especially for streaming workloads where you basically have to pick between latency and read performance. The decision flow diagram nails it though, batch jobs can just run compaction right after and get the best of both worlds. Have you seen cases where optimized writes still cause SLA issues even with modest data volumes?
The tradeoff between online and offline optimization is really interesting, especially for streaming workloads where you basically have to pick between latency and read performance. The decision flow diagram nails it though, batch jobs can just run compaction right after and get the best of both worlds. Have you seen cases where optimized writes still cause SLA issues even with modest data volumes?
Thanks.
I have used optimized write for 10k messages per minute and it adds like 10sec latency, which is mostly fine.