Song Younghwan created KAFKA-6539: ------------------------------------- Summary: KafkaConsumer endlessly try to connect to a dead broker, ignoring brokers alive Key: KAFKA-6539 URL: https://issues.apache.org/jira/browse/KAFKA-6539 Project: Kafka Issue Type: Bug Components: consumer Affects Versions: 1.0.0 Environment: Java 8 Brokers on CentOS 7.4 Consumers on Windows 10 Reporter: Song Younghwan Attachments: consumer.log
I consider to use Kafka in my company, so currently doing failover test. Conditions: * org.apache.kafka:kafka-clients:1.0.0 * New consumer using bootstrap.servers, a consumer group and a group coordinator * num. brokers = 3 (id #1, #2, #3) * Topic num. partitions = 3, replication factor = 3 * offsets.topic.replication.factor = 3 Reproduction Step: # Run consumers in the same consumer group, each of them subscribe to a topic # Kill (kill -9) #1, #2 broker simultaneously (only #3 online) # Consumers eventually connect to #3 broker # Start #1, #2 broker again after a while (#1, #2, #3 online) # Kill (kill -9) #2, #3 broker simultaneously (only #1 online) # *{color:#FF0000}Now consumers endlessly try to connect to #3 broker only{color}* # Start #2 broker again after a while (#1, #2 online) # *{color:#FF0000}Consumers still blindly try to connect to #3 broker{color}* Expectation: Consumers successfully connect to #1 broker after step 5. Record: I attached a consumer log file with TRACE log level. Related events below: * 12:03:13 kills #1, #2 broker simultaneously * 12:03:42 starts #1, #2 broker again * 12:04:01 kills #2, #3 broker simultaneously * 12:04:42 starts #2 broker again -- This message was sent by Atlassian JIRA (v7.6.3#76005)