Hi everyone. I believe I have an issue in the clus...
# troubleshooting
k
Hi everyone. I believe I have an issue in the cluster but I am not sure. I am getting few segments like this in Zookeeper.
Copy code
"auctionsStats__6__13__20250703T1233Z": {
      "Server_pinot-prod-server-0.pinot-prod-server-headless.pinot.svc.cluster.local_8098": "OFFLINE",
      "Server_pinot-prod-server-1.pinot-prod-server-headless.pinot.svc.cluster.local_8098": "OFFLINE",
      "Server_pinot-prod-server-2.pinot-prod-server-headless.pinot.svc.cluster.local_8098": "OFFLINE"
    },
When I try to download the segments again, I get an error saying they are not in my deepstore. However, queries seem to work normally. Is it expected for segments to be reported as offline and missing in deepstore? What exactly does offline mean as a segment status? Bellow are the latest messages for the above segment:
Copy code
INFO 2025-07-18T05:35:05.820609035Z [resource.labels.containerName: server] 2025/07/18 05:35:05.820 INFO [HttpClient] [auctionsStats__6__13__20250703T1233Z] Sending request: http://pinot-prod-controller-1.pinot-prod-controller-headless.pinot.svc.cluster.local:9000/segmentStoppedConsuming?reason=org.apache.pinot.shaded.org.apache.kafka.common.KafkaException&streamPartitionMsgOffset=0&instance=Server_pinot-prod-server-2.pinot-prod-server-headless.pinot.svc.cluster.local_8098&offset=-1&name=auctionsStats__6__13__20250703T1233Z to controller: pinot-prod-controller-1.pinot-prod-controller-headless.pinot.svc.cluster.local, version: Unknown
INFO 2025-07-18T05:35:05.821542868Z [resource.labels.containerName: server] 2025/07/18 05:35:05.821 INFO [ServerSegmentCompletionProtocolHandler] [auctionsStats__6__13__20250703T1233Z] Controller response {"status":"PROCESSED","streamPartitionMsgOffset":null,"isSplitCommitType":true,"buildTimeSec":-1} for http://pinot-prod-controller-1.pinot-prod-controller-headless.pinot.svc.cluster.local:9000/segmentStoppedConsuming?reason=org.apache.pinot.shaded.org.apache.kafka.common.KafkaException&streamPartitionMsgOffset=0&instance=Server_pinot-prod-server-2.pinot-prod-server-headless.pinot.svc.cluster.local_8098&offset=-1&name=auctionsStats__6__13__20250703T1233Z
INFO 2025-07-18T05:35:05.821571462Z [resource.labels.containerName: server] 2025/07/18 05:35:05.821 INFO [RealtimeSegmentDataManager_auctionsStats__6__13__20250703T1233Z] [auctionsStats__6__13__20250703T1233Z] Got response {"status":"PROCESSED","streamPartitionMsgOffset":null,"isSplitCommitType":true,"buildTimeSec":-1}
INFO 2025-07-18T05:35:05.983729827Z [resource.labels.containerName: server] 2025/07/18 05:35:05.976 INFO [local_8098 - SegmentOnlineOfflineStateModel] [HelixTaskExecutor-message_handle_thread_7] SegmentOnlineOfflineStateModel.onBecomeOfflineFromConsuming() : ZnRecord=cc787368-9a93-42f3-8588-ebefe88f2a07, {CREATE_TIMESTAMP=1752816905933, ClusterEventName=IdealStateChange, EXECUTE_START_TIMESTAMP=1752816905976, EXE_SESSION_ID=300627ec087008e, FROM_STATE=CONSUMING, MSG_ID=cc787368-9a93-42f3-8588-ebefe88f2a07, MSG_STATE=read, MSG_TYPE=STATE_TRANSITION, PARTITION_NAME=auctionsStats__6__13__20250703T1233Z, READ_TIMESTAMP=1752816905959, RESOURCE_NAME=auctionsStats_REALTIME, RESOURCE_TAG=auctionsStats_REALTIME, RETRY_COUNT=3, SRC_NAME=pinot-prod-controller-2.pinot-prod-controller-headless.pinot.svc.cluster.local_9000, SRC_SESSION_ID=2006281fc800087, STATE_MODEL_DEF=SegmentOnlineOfflineStateModel, STATE_MODEL_FACTORY_NAME=DEFAULT, TGT_NAME=Server_pinot-prod-server-2.pinot-prod-server-headless.pinot.svc.cluster.local_8098, TGT_SESSION_ID=300627ec087008e, TO_STATE=OFFLINE}{}{}, Stat=Stat {_version=0, _creationTime=1752816905946, _modifiedTime=1752816905946, _ephemeralOwner=0}
INFO 2025-07-18T05:35:05.984995178Z [resource.labels.containerName: server] 2025/07/18 05:35:05.983 INFO [HelixInstanceDataManager] [HelixTaskExecutor-message_handle_thread_7] Removing segment: auctionsStats__6__13__20250703T1233Z from table: auctionsStats_REALTIME
INFO 2025-07-18T05:35:05.985038958Z [resource.labels.containerName: server] 2025/07/18 05:35:05.983 INFO [auctionsStats_REALTIME-RealtimeTableDataManager] [HelixTaskExecutor-message_handle_thread_7] Removing segment: auctionsStats__6__13__20250703T1233Z from table: auctionsStats_REALTIME
INFO 2025-07-18T05:35:05.985045952Z [resource.labels.containerName: server] 2025/07/18 05:35:05.983 INFO [auctionsStats_REALTIME-RealtimeTableDataManager] [HelixTaskExecutor-message_handle_thread_7] Closing segment: auctionsStats__6__13__20250703T1233Z of table: auctionsStats_REALTIME
INFO 2025-07-18T05:35:05.985110098Z [resource.labels.containerName: server] 2025/07/18 05:35:05.984 INFO [MutableSegmentImpl_auctionsStats__6__13__20250703T1233Z_auctionsStats] [HelixTaskExecutor-message_handle_thread_7] Trying to close RealtimeSegmentImpl : auctionsStats__6__13__20250703T1233Z
INFO 2025-07-18T05:35:05.985117081Z [resource.labels.containerName: server] 2025/07/18 05:35:05.984 INFO [auctionsStats_REALTIME-6-ConcurrentMapPartitionUpsertMetadataManager] [HelixTaskExecutor-message_handle_thread_7] Skip removing untracked (replaced or empty) segment: auctionsStats__6__13__20250703T1233Z
INFO 2025-07-18T05:35:05.987557288Z [resource.labels.containerName: server] 2025/07/18 05:35:05.987 INFO [MmapMemoryManager] [HelixTaskExecutor-message_handle_thread_7] Deleted file /var/pinot/server/data/index/auctionsStats_REALTIME/consumers/auctionsStats__6__13__20250703T1233Z.0
INFO 2025-07-18T05:35:05.990545309Z [resource.labels.containerName: server] 2025/07/18 05:35:05.990 INFO [auctionsStats_REALTIME-RealtimeTableDataManager] [HelixTaskExecutor-message_handle_thread_7] Closed segment: auctionsStats__6__13__20250703T1233Z of table: auctionsStats_REALTIME
INFO 2025-07-18T05:35:05.990570191Z [resource.labels.containerName: server] 2025/07/18 05:35:05.990 INFO [auctionsStats_REALTIME-RealtimeTableDataManager] [HelixTaskExecutor-message_handle_thread_7] Removed segment: auctionsStats__6__13__20250703T1233Z from table: auctionsStats_REALTIME
INFO 2025-07-18T05:35:05.990578459Z [resource.labels.containerName: server] 2025/07/18 05:35:05.990 INFO [HelixInstanceDataManager] [HelixTaskExecutor-message_handle_thread_7] Removed segment: auctionsStats__6__13__20250703T1233Z from table: auctionsStats_REALTIME
INFO 2025-07-18T06:15:57.880369560Z [resource.labels.containerName: controller] 2025/07/18 06:15:57.880 INFO [PinotLLCRealtimeSegmentManager] [pool-10-thread-7] Repairing segment: auctionsStats__6__13__20250703T1233Z which is OFFLINE for all instances in IdealState
m
Segment should be in consuming, online or error state. Offline is usually a transient state, if it is stuck on there, you have some other issue that is leading to it.
k
Understood. Thanks. I’ll keep monitoring. We switched to a more stable Kafka deployment today, so I hope this will help. The issue might have been related to our Kafka setup.
What can I do with these segments if they’re not in Deep Store? I checked, and the files are missing from google.
m
Are these the latest segments for those partitions? If so, they were never committed and hence just reloading will work
k
They weren't latest segements for the partition. I deleted them it seems to not affect anything in my data.
However now I have new offline segements again. You can see my log files here: It seems like there was a DNS resolution issue and the segment has been removed from one of the nodes and the segment never recovered. Is this expected behaviour?
When I manually reloaded the segment, everything seemed fine again. It became online, and a new segment started consuming data.
m
I have never seen running into these issues. The fact you are running so much into it I feel there is a serious issue with your upstream or with ZK setup.
k
We are running this helm: https://docs.pinot.apache.org/basics/getting-started/kubernetes-quickstart in Google's managed kubernetes. Is this helm production ready by design?
We have almost no traffic. We generate roughly 10K Kafka messages per day.
Our biggest tables is 200MB.
What do you mean by “upstream”? We are running entire Pinot and Kafka clusters in Google’s London data center.
m
I mean if ZK is not running, or you have network connectivity issue or Kafka is not stable etc
Helm charts in OSS are for quick start, you need to customize it per your workload. But since your workload seems very tiny, it should be fine
k
Thanks. I will proceed my investigation. I also believe it is infrastructure issue in its core.
m
Yes, agreed.
k
@Mayank When we switched from our hosted Kafka to a managed one, Pinot became much more stable. I believe the main issue was with the Kafka cluster itself. How we managed to mess up a Kafka cluster is a different story that I’ll investigate separately. Thanks for you assitnance.
m
Glad to hear that. Yeah that’s what I meant by upstream problems.