Realtime indexing data looks incorrect when task.partitions is not 1


I am using druid for real time analysis service. And I meet some problems after I modify task.partitions from 1 to 2.

Druid creates 2 tasks per hour as expected. But I have found data for every LAST minute per hour is incorrect while it matches with real log files perfectly on other minutes. Any suggestion?

The important related setting:

“granularitySpec”: {

“segmentGranularity”: “hour”,

“type”: “uniform”,

“queryGranularity”: “minute”


“tuningConfig”: {

“maxRowsInMemory”: “300000”,

“type”: “realtime”,

“windowPeriod”: “PT10M”,

“intermediatePersistPeriod”: “PT1M”


“properties”: {

“task.partitions” : “2”,

“topicPattern.priority”: “1”,

“topicPattern”: “testData”