Dustin Cote created KAFKA-5675:
----------------------------------
Summary: Possible worker_id duplication in Connect
Key: KAFKA-5675
URL: https://issues.apache.org/jira/browse/KAFKA-5675
Project: Kafka
Issue Type: Bug
Components: KafkaConnect
Affects Versions: 0.10.2.1
Reporter: Dustin Cote
Priority: Minor
It's possible to set non-unique host/port combinations for workers via
*rest.advertised.host.name* and *rest.advertised.host.port* (e.g.
localhost:8083). While this isn't typically advisable, it can result in weird
behavior for containerized deployments where localhost might end up being
mapped to something that is externally facing. The worker_id today appears to
be set as this host/port combination so you end up with duplicate worker_ids
causing long rebalances presumably because task assignment gets confused. It
would be good to either change how the worker_id is generated or find a way to
not let a worker start if a worker with an identical worker_id already exists.
In the short term, we should document the requirement of unique advertised
host/port combinations for workers to avoid debugging a somewhat tricky
scenario.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
