Now we want to take a pressure test to our system to find the bottleneck in our system and take some optimizations to our system, our system using Kafka indexing instead of realtime node,the whole architecture is like this:
Now we want to simulate a high concurrency queries and add streaming data to our system to find the bottleneck in our system, and change some parameters to see if it can optimize our system.We plan to optimize parameters in 3 parts:
1.Increase taskcount: we had found that when we have 300+ tasks in midmanagers, when they pass a duration, they will query to mysql in the table named pendingSegments. If we dont’t add index to this table, it will take a long time in query, so we increase taskcount to make a task run fast, but it can add task nums. so we want to increase taskcount to add tasks to see if the whole system have other bottleneck.
2.Increase replications: increase replications to see if it can speed the query.
3.Change Kafka partition:we found different datasource has different size, so we want to change different Kafka’s topic partition, if the datasource’size is small, decrease the partition, to make the segment size to average.
Hope all to give some suggestions to the 3 plans above, or give us some other suggestions to optimize the druid, and if you have some experience in pressure test the druid system, hope to give us some suggestions, sincerely thanks for your answer.