Ways to debug backPressure

We started seeing a lot of backPressure from ingestion metric from Druid middle manager a lot. Is there a way to debug the issue?

We’re using Druid 0.9.0 with Tranquility connected to Storm 0.9.4

After digging through the log I found that ingestion/persists/time is as high as 6 minutes. I’m not sure if that’s normal?

From the log, I also see something like this.

2016-07-12T20:51:45,842 INFO [sparrow-firehose-web-incremental-persist] io.druid.segment.IndexMerger - Starting dimension[sVe] with cardinality[391]
2016-07-12T20:51:46,984 INFO [sparrow-firehose-web-incremental-persist] io.druid.segment.IndexMerger - Completed dimension[sVe] in 1,142 millis.


1,142 millis is the highest I can see most of the dimensions are within 10-100ms. Is that normal?

We recently updated the metrics docs to clarify what backPressure actually means. https://github.com/druid-io/druid/pull/3243

How long does a handoff take right now?

Oh thanks. So that means, the bigger the segment the longer it takes to persist?

Yes, the more data in the in-memory buffer, the longer it takes to index and create the segment.