@Rocky how to fix this 2026-03-04 124602.197+07:00 WARN (Routine load scheduler|41) [KafkaUtil$ProxyAPI.sendProxyRequest():262] Failed to send get kafka partition info request to BE TNetworkAddress(hostname:starrocks-lakehouse-be-0.starrocks-lakehouse-be-se arch.starrocks-lakehouse.svc.cluster.local, port:8060), err: Ocurrs time out with specfied time 12 SECONDS 2026-03-04 124602.197+07:00 WARN (pool-21-thread-108|14081) [KafkaUtil$ProxyAPI.sendProxyRequest():262] Failed to send get kafka partition info request to BE TNetworkAddress(hostname:starrocks-lakehouse-cn-2.starrocks-lakehouse-cn-sea rch.starrocks-lakehouse.svc.cluster.local, port:8060), err: Ocurrs time out with specfied time 12 SECOND

The error you are seeing indicates that the StarRocks Frontend (FE) timed out while waiting for a Backend (BE) to retrieve partition information from Kafka. By default, this operation has a hardcoded or configured timeout (often around 10-12 seconds). Here are the steps to troubleshoot and fix this issue: 1. Check Connectivity from BE to Kafka The BE nodes are the ones actually communicating with Kafka. Ensure every BE node can reach your Kafka brokers. • Log into the BE nodes mentioned in the error (

starrocks-lakehouse-be-0

and

starrocks-lakehouse-cn-2

). • Try to ping the Kafka brokers or use a tool like

telnet

or

nc

to check the Kafka port (usually 9092). • Use a tool like

kcat

(formerly

kafkacat

) from the BE pod to see if you can fetch metadata manually:

bash
  kcat -L -b <kafka_broker_list>

2. Increase the Routine Load Timeout If your Kafka cluster is large or under high load, it might take longer than 12 seconds to return metadata. You can increase the timeout setting in the FE configuration. You can check the current configuration by running:

sql
ADMIN SHOW CONFIG LIKE '%routine_load_kafka_timeout_second%';

To increase it (e.g., to 30 seconds), run this on the FE:

sql
ADMIN SET FRONTEND CONFIG ("routine_load_kafka_timeout_second" = "30");

Note: Depending on your StarRocks version, this parameter might need to be set in the

fe.conf

file and requires an FE restart if it's not dynamic. 3. Check Kafka Broker Side • Load: Check if the Kafka brokers are overloaded or if there are a massive number of partitions in the topic. • DNS/Advertised Listeners: Ensure that the

advertised.listeners

configured in Kafka are resolvable and reachable from within your Kubernetes cluster. If Kafka is returning an IP or hostname that the BE cannot route to, the request will hang and timeout. 4. BE Resources Check the BE logs for any signs of "Process overhead" or "Memory limit exceeded" which might cause the BE's internal Kafka client to become unresponsive. • Look at

be/log/be.INFO

or

be/log/be.WARNING

on the affected nodes. 5. Kubernetes Network Policy Since you are running in Kubernetes (

svc.cluster.local

), ensure there are no NetworkPolicies blocking traffic between the StarRocks BE namespace and the Kafka namespace/brokers. Summary Recommendation: Start by increasing

routine_load_kafka_timeout_second

to

30

and verify that the BE pods can successfully run a

kcat

command against the Kafka brokers.