Is there a config to increase the timeout of broker queries? We seem to get the following error if the

timeUsedMs

is >= 10s:

{
  "exceptions": [],
  "numServersQueried": 3,
  "numServersResponded": 0,
  "numSegmentsQueried": 0,
  "numSegmentsProcessed": 0,
  "numSegmentsMatched": 0,
  "numConsumingSegmentsQueried": 0,
  "numDocsScanned": 0,
  "numEntriesScannedInFilter": 0,
  "numEntriesScannedPostFilter": 0,
  "numGroupsLimitReached": false,
  "totalDocs": 0,
  "timeUsedMs": 10000,
  "segmentStatistics": [],
  "traceInfo": {},
  "minConsumingFreshnessTimeMs": 0
}

If the query takes below 10s it's fine, even 9.7 seconds for example

so, 9.7 seconds is a long time for the query. What is the best way to figure out what is the bottleneck?

In the query result set you can see information related to the execution plan. You can see how many segments were scanned, and that will give you some details on whether or not your indexing configuration can be improved. 9.7 seconds is on the highest end of a response that I have seen.

Can you also paste your query here?

It's something like:

select foo, bar, count(*) from pinot_table where timestamp >= <now - 5 minutes> group by foo, bar

So one of the issues was that we used the old spec for star tree indexes. I recreated and verified we have star tree indexes on the table now (and inverted indexes) - how do I rebuild the realtime segments so that the star tree indexes are created?

pinot-0.3.0 doesn't seem to build star tree indexes - the setStarTreeIndexConfigs in IndexConfig is unused. So it's only for offline segments, right?

Looks like it is built for realtime tables in master though.

Is that correct?

yes, it was added later to build it on the fly

checking if its enabled in real-time

@Kishore G @Elon I don't think star-tree will be added on the fly (during segment reload). Building star-tree on the server side could cost a lot of CPU and memory, thus hurt the server performance or cause it to run out of memory. Maybe leveraging Minion to do this is safer.

Hi Jackie, I agree with you for external analytics use cases. But for internal use cases, it would be ok to generate it on the fly. we should enable this via config and let the user's decide if they want to enable it and provide enough resources.

Sure, we can add another segment preprocessor to handle this

Thanks for all the help yesterday, we got the query down to 30-50ms from 8s!

qq - if we add/remove inverted or sorted indexes do the realtime segments get updated or is there some rebuild command (for realtime)?

sorted index will only be applicable to new segments. inverted index can be created by running reload segments api

I hit an error when trying to use multi-value columns with star tree index. It looks like these are unsupported. Any timelines for when these will be supported?

Caused by: java.lang.IllegalStateException: Multi value column is not supported
	at org.apache.pinot.core.data.readers.PinotSegmentColumnReader.getDictionaryId(PinotSegmentColumnReader.java:143) ~[pinot-all-0.4.0-SNAPSHOT-jar-with-dependencies.jar:0.4.0-SNAPSHOT-c152f18f273042668a6d91d5d15c7ca018f31b4a]
	at org.apache.pinot.core.startree.v2.builder.BaseSingleTreeBuilder.getSegmentRecordDimensions(BaseSingleTreeBuilder.java:228) ~[pinot-all-0.4.0-SNAPSHOT-jar-with-dependencies.jar:0.4.0-SNAPSHOT-c152f18f273042668a6d91d5d15c7ca018f31b4a]

Do you need it?

What kind of behavior are you looking for with multi value column

I probably don't need it yet. I wanted to be able to have a list of experiment groups on a row and aggregate by the groups. I also want to have support for custom dimensions but this isn't designed yet (I have N separate fields currently on the event).

It'd make the experiment group work easier.

Is there a way I can treat an index more as a summary table? E.g. support getting aggregate stats for a small number of sub dimensions? I have about 20 optional dimensions that I might want to combine with the required dimensions. I don't need all of the dimensions all of the time.

E.g. I have dimensions like "device", a bunch for geo, I have some custom dimensions. I'm fine not supporting aggregate stats for multiple dimensions.

I dint follow, do you have a query?

I am guessing you are referring star tree

Yea, I'll send directly.