This message was deleted.

There are metrics to measure the performance and there are metrics to troubleshoot slow performance. Lag and even throughput are the best indicators of performance throughput. For troubleshooting the slow performance in case it doesn't meet the bar you think it should, you can do the following Flame graphs - As you start doing this testing, you can also take some flame graphs on a task to see where the task is spending most of its time in. You can do it on one task. Application Metrics - There are metrics such as

task/action/runtime

,

task/pending/time

that should be monitored. We often see that in a high-load situation, the metadata store is not properly tuned and a slow metadata store starts impacting ingestion performance. The value of

task/action/runtime

is a good proxy of db performance. (https://github.com/apache/druid/issues/12701) System metrics - JVM gc, heap utilization, CPU settings will tell you if you are not giving enough hardware to the application. Usually, there is no need to do any of it. If you are starting afresh, use the 27 release, which has a lot more improvements when it comes to metadata store activity.

@Abhishek Agarwal Thanks for the info. Let me go through it. Then I will get back to you