Druid ingest data decrease while publishing segments

Our druid generate segments per hour ( segmentGranularity = 1H), thus ingesting data had a shock every hour.

The druid ingest metrics shows the ingest events processed, just as follows:

And the data ingest in our business system is more obvious, just as follows.

Is there anyone can help me to fix this? Our business system is very sensitive to real-time data, every shock in the ingest may cause alarm.



Hey Xinxin,

Are the dips in ingestion happening between subsequent indexing tasks or during the run of an index task? If during the run, could you post the logs for your index task when this happens? Specifically I’m wondering if there are any logs messages similar to ‘Ingestion was throttled for [%,d] millis because persists were pending.’ in there.

Hi David,

Our task logs did not have “Ingesting was throttled for …”.

But we tried to change half datasources’ segmentGranularity from HOUR to DAY, and this problem was not as serious as before. But we are still looking for improvement.

Thanks a lot,

在 2016年10月27日星期四 UTC+8上午8:04:22,David Lim写道:

Hi Xinxin,
Are you using tranquility to create druid tasks ?

If using tranquility try tuning task.warmingPeriod in your tranquility configs (https://github.com/druid-io/tranquility/blob/master/docs/configuration.md) which can hopefully help in your case.

Hi Nishant,
I did not use tranquility to create druid tasks, just kafka indexing service.

在 2016年11月18日星期五 UTC+8下午3:28:44,Nishant Bangarwa写道:

Hey Xinxin, is this still an issue for you or have you found a way to make it work with the rest of your systems?