Hi Team, I have a couple of queries regarding Apache Pinot: 1. Does Pinot support segment compression formats other than

tar.gz

, such as zstd or Snappy? 2. I created an index on a column (

col1

) and ingested data. Suppose a segment contains 50 records, and I run a query with the condition

col1 = 'xyz'

. In this case, does Pinot load the entire segment into memory and then filter the records, or does it directly fetch only the matching data from the segment?

1. Pinot has its own compression techniques for storing data and index. For no-dict column it can be configured to use zstd / snappy. The tar.gz format is for storing in deepstore only. 2. segments are memory mapped and never really entirely loaded in memory. If you have index on the column it will fetch only the marching data

thanks! 🙇

1. In OSS yes, but there is a tiered storage solution in closed source where it does not keep all data. 2. it is a distributed system, it needs to provide fault tolerance, disaster recovery etc 3. Yes controller notifies server using Helix. 4. What if all servers go down? Or you want to scale down and then scale back up later. What if server loses its disk? 5. I don’t recall, tarDie might be temporary location to download tar files. 6. did you delete from all replicas? 7. it is to memory map segment on disk, but not in direct memory. You can google how memory mapping works

did you delete from all replicas?

Yes, i deleted all the replicas

There is no way queries return results if servers don’t have data.

If it is returning the result after deleting data from the server’s local directory, what could be the possible reason?

I am saying if data is truely delete and unmapped from server, there is absolutely no way query can return results.

Does it cache data?

How did you delete the segments?

I deleted all the files from this directory:

pinot.server.instance.dataDir

, and reloaded the segments without restarting the server.

You did \rm -fr? Or you called the delete api?

You did rm -rf

yes, i did this

And then you called reload api - this would have re-downloaded the segments from deepstore to server.

But I can't see the data in the server's local directory. If it actually downloads the data from deep store, it should keep it in the server's local directory.

I think you might be looking in the wrong place

> 2025/08/23 233451.426 INFO [QueryLogger] [jersey-server-managed-async-executor-6] requestId=1452998218000000006,table=user_batch,timeMs=22,docs=16/7998,entries=0/80,segments(queried/processed/matched/consumingQueried/consumingProcessed/consumingMatched/unavailable):9/4/4/0/0/0/0,consumingFreshnessTimeMs=0,servers=1/1,groupLimitReached=false,brokerReduceTimeMs=0,exceptions=0,serverStats=(Server=SubmitDelayMs,ResponseDelayMs,ResponseSize,DeserializationTimeMs,RequestSentDelayMs);localhost_O=2,15,694,1,2,offlineThreadCpuTimeNs(total/thread/sysActivity/resSer):0/0/0/0,realtimeThreadCpuTimeNs(total/thread/sysActivity/resSer):0/0/0/0,clientIp=unknown,query=select * from user_batch limit 10 > 2025/08/23 233503.307 INFO [QueryLogger] [jersey-server-managed-async-executor-7] requestId=1452998218000000007,table=user_batch,timeMs=9,docs=16/7998,entries=0/80,segments(queried/processed/matched/consumingQueried/consumingProcessed/consumingMatched/unavailable):9/4/4/0/0/0/0,consumingFreshnessTimeMs=0,servers=1/1,groupLimitReached=false,brokerReduceTimeMs=0,exceptions=0,serverStats=(Server=SubmitDelayMs,ResponseDelayMs,ResponseSize,DeserializationTimeMs,RequestSentDelayMs);localhost_O=0,7,685,0,1,offlineThreadCpuTimeNs(total/thread/sysActivity/resSer):0/0/0/0,realtimeThreadCpuTimeNs(total/thread/sysActivity/resSer):0/0/0/0,clientIp=unknown,query=select * from user_batch limit 10 > 2025/08/23 233503.829 INFO [QueryLogger] [jersey-server-managed-async-executor-7] requestId=1452998218000000008,table=user_batch,timeMs=9,docs=16/7998,entries=0/80,segments(queried/processed/matched/consumingQueried/consumingProcessed/consumingMatched/unavailable):9/4/4/0/0/0/0,consumingFreshnessTimeMs=0,servers=1/1,groupLimitReached=false,brokerReduceTimeMs=0,exceptions=0,serverStats=(Server=SubmitDelayMs,ResponseDelayMs,ResponseSize,DeserializationTimeMs,RequestSentDelayMs);localhost_O=0,6,694,1,1,offlineThreadCpuTimeNs(total/thread/sysActivity/resSer):0/0/0/0,realtimeThreadCpuTimeNs(total/thread/sysActivity/resSer):0/0/0/0,clientIp=unknown,query=select * from user_batch limit 10 > 2025/08/23 233504.208 INFO [QueryLogger] [jersey-server-managed-async-executor-7] requestId=1452998218000000009,table=user_batch,timeMs=8,docs=16/799WARN [SegmentLocalFSDirectory] [grizzly-http-server-10] Empty list of files for path: /tmp/pinot/data/server/index/user_batch_OFFLINE/prod_user_batch-1755846294280-000_batch_v1, segmentDirectory: /tmp/pinot/data/server/index/user_batch_OFFLINE/prod_user_batch-1755846294280-000_batch_v1/v3 > 2025/08/23 233220.604 WARN [SegmentLocalFSDirectory] [grizzly-http-server-10] Empty list of files for path: /tmp/pinot/data/server/index/user_batch_OFFLINE/prod_user_batch-1755846294947-000_batch_v1, segmentDirectory: /tmp/pinot/data/server/index/user_batch_OFFLINE/prod_user_batch-1755846294947-000_batch_v1/v3 Here, I can see a warning in the logs

Empty list of files for path: /tmp/pinot/data/server/index/user_batch_OFFLINE/prod_user_batch-1755846294280-000_batch_v1, segmentDirectory: /tmp/pinot/data/server/index/user_batch_OFFLINE/prod_user_batch-1755846294280-000_batch_v1/v3

Is this your configured dataDir?

/tmp/pinot/data/server/index

yes

If so, never have dataDir on

/tmp

Yes, the tmp location (the default directory location in

server.conf

) is set on my local machine only for POC. The tmp files will be deleted once I restart the system. In production, it will point to a different location. During my POC, I deleted the files and then checked the behavior to see whether it would download the files from deep store or not, and I observed different behavior.

That is very basic functionality in Pinot, without that a distributed system won't work.