[GitHub] [druid] haragr opened a new issue #12135: High querytime for topN queries with injective=false namespace after upgrade to druid 0.22

[GitBox <gi...@apache.org>]

haragr opened a new issue #12135:
URL: https://github.com/apache/druid/issues/12135


   After upgrade to 0.22.0 version, we started observing significantly higher query and cpu times for namespaces with injective=false compared to 0.18.2. No config change was made from our end and we ensured that data remains the same.
   
   For v18, the querytime was around 450ms.  
   For v22, the querytime was around 80,000ms
   
   The time difference was at historical service. Downgrading only the historicals fixed the issue.
   
   Datasource1 details (For 1 day of data)-> Total size is around 270G  
   Segment count is around 550  
   Rows per segment = 802,500  
   Dimension1 cardinality is around 1000
   
   Namespace1 details ->  
   Count = 294,101  
   Injective = false
   
   Query ->
   
   `{"dimension": {"type": "lookup","dimension": "dimension1","outputName": "Result","name": "namespace1","injective": false,"retainMissingValue": true},"threshold": 50,"metric": {"type": "numeric","metric": "metric1"},"context": {"queryId": "queryID123","useCache":false,"vectorize":false,"debug":true,"timeout":200000},"queryType": "topN","dataSource": "datSource1","granularity": "All","filter": {"type": "and","fields": [{"type": "or","fields": [{"type": "in","dimension": "dimension1","extractionFn": {"type": "registeredLookup","lookup": "namespace1","retainMissingValue": true},"values": ["123 - test string ","234 - test string ",null]}]}]},"intervals": ["2021-12-05T00:00:00Z/2021-12-06T00:00:00Z"],"aggregations": [{"type": "filtered","aggregator": {"type": "longSum","name": "metric1","fieldName": "Views"},"filter": {"type": "and","fields": [{"type": "and","fields": [{"type": "not","field": {"type": "selector","dimension": "dimension2","value": "0"}}]}]}}],"postAggregations": []}`
   
   Below are the flame graphs for druid historical service when only the query in question was running.
   
   [flame_graph.zip](https://github.com/apache/druid/files/7838014/flame_graph.zip)
   
   
   In the flamegraphs, we noticied a major portion was spent in query processing in v0.22.0.  
   If we follow the topN_datasource_1 stack to the top , we notice that the classes used for string aggregation are different across versions.  
   In v0.22.0 -
   
   > Lorg/apache/druid/query/topn/types/StringTopNColumnAggregatesProcessor:::scanAndAggregateWithCardinalityUnknown
   
   In v0.18.1 -
   
   > Lorg/apache/druid/query/topn/types/StringTopNColumnAggregatesProcessor:::scanAndAggregateWithCardinalityKnown
   
   Also in v18 , there is usage of class
   
   > Lorg/apache/druid/query/topn/HeapBasedTopNAlgorithm:::scanAndAggregate
   
   which is missing in v22.
   
   There was no change in underlying data between the tests.
   
   Can anyone please have a look and see if this is a genuine issue with druid.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@druid.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@druid.apache.org
For additional commands, e-mail: commits-help@druid.apache.org