Description of problem:

When it's configured an Kafka output and the message is larger than 1MB and configured the "spec.tuning.outputs.maxWrite" option to allow to write a batch longer than 1MB like:

  - kafka:
      url: tcp://kafka.openshift-logging.svc.cluster.local:9092/clo-topic 
      authentication:
        sasl:
          mechanism: PLAIN
          password:
            key: password
            secretName: kafka-vector
          username:
            key: username
            secretName: kafka-vector
      tuning:
        maxWrite: 6M

It's observed the error:

2025-08-25T22:52:12.077585Z ERROR sink{component_kind="sink" component_id=output_kafka_app component_type=kafka}: vector_common::internal_event::service: Service call failed. No retries or retries exhausted. error=Some(KafkaError (Message production error: MessageSizeTooLarge (Broker: Message size too large))) request_id=4 error_type="request_failed" stage="sending" internal_log_rate_limit=true

*Notes*:
1.The kafka server accepts messages longer than 1MB
2.If it's used Fluentd (Logging v5) instead of Vector, it works and the message arrive to Kafka

Version-Release number of selected component (if applicable):

Tested the issue in Logging 5.9.z, 6.1.8, 6.3.z

How reproducible:

Always

Steps to Reproduce:

1. Deploy a Kafka server that accepts messages up 10MB
2. Configure the "clusterLogForwarder" to log forward to the Kafka server and configure the "spec.output.tuning.maxWrite" up 6MB similar to:

   [...]
   spec:
     outputs:
     - kafka:
         url: tcp://kafka.openshift-logging.svc.cluster.local:9092/clo-topic 
         authentication:
           sasl:
             mechanism: PLAIN
             password:
               key: password
               secretName: kafka-vector
             username:
               key: username
               secretName: kafka-vector
         tuning:
           maxWrite: 6M
       name: kafka-app
       type: kafka
     pipelines:
     - inputRefs:
       - application
       name: test-app
       outputRefs:
       - kafka-app
   

3. Create an application that generates a log up to 2MB

   $ oc new-project test-kafka
   $  kubectl create deployment hello-node --image=registry.k8s.io/e2e-test-images/agnhost:2.43 -- /agnhost serve-hostname 
   $ oc -n test-kafka rsh $(oc get pods -n test-kafka -l app=hello-node -o name)
   ~ $ for i in $(seq 1 $((2 * 1024 * 1024 / 5))); do echo -n 'hello'; done > /tmp/2mb_hello_line.txt
   ~ $ du -khs /tmp/2mb_hello_line.txt 
   2.0M    /tmp/2mb_hello_line.txt
   ~ $ cat /tmp/2mb_hello_line.txt > /proc/1/fd/1
   

Actual results:

The collector pod running in the same node that the pod that generates logs up 2MB generates the error under this line and logs don't arrive to Kafka.

2025-08-25T22:52:12.077585Z ERROR sink{component_kind="sink" component_id=output_kafka_app component_type=kafka}: vector_common::internal_event::service: Service call failed. No retries or retries exhausted. error=Some(KafkaError (Message production error: MessageSizeTooLarge (Broker: Message size too large))) request_id=4 error_type="request_failed" stage="sending" internal_log_rate_limit=true

If it's used Logging v5 with Fluentd, logs are received in Kafka showing that the Kafka server accepts messages bigger than 1MB.

Note: The tuning option "maxWrite" [0] is translated to "batch.max_bytes" [1] and this option is passed to "librdkafka_options.batch.size". If instead of using "maxWrite" is moved to "Unmanaged" the ClusterLogForwarder and set by hand "librdkafka_options."batch.size" = "3000000"", it continues failing.

Expected results:

Kafka sink honour the tuning option "maxWrite" and allows to log forward logs until the size set bigger than 1MB.

Additional info: