Kafka Input Plugin: Shutdown semantics

[kaihil]

• Version: Logstash 6.2.1, Kafka 1.1.0
• Operating System: macOS 10.13.5 (17F77) / Docker Version 18.03.1-ce-mac65 (24312)
• Config File:

input {
    kafka {
        bootstrap_servers => "kafka:9092"
        topics => "example-topic"
        codec => "json"
        auto_offset_reset => "latest"
    }
}

• Steps to Reproduce:

I have a question regarding the shutdown semantics of Logstash's Kafka Input Plugin:

When I run it with default settings, send a batch of 10.000 message into kafka and restart* logstash while it is processing the message, I end up with more than 10.000 (e.g. 10.212) messages in ES.

So at a first glance it looks like it just continues at the last commit done via the auto_commit_interval_ms (which may be up to 5seconds old). But I would assume that the plugin properly closes the consumer and sends a last offset commit to avoid double processing of messages.

Now I wonder how this is supposed to work?

Thank you in advance!
Kai

*) I restart logstash with a SIGTERM first, wait until it is properly shut down and then wait a minute to start it again.

[numbnut]

Hi,
any news regarding this issue?