Proven Provenzano created KAFKA-14860:
-----------------------------------------
Summary: New controllers created for migration should validate
connectivity to brokers before starting migration
Key: KAFKA-14860
URL: https://issues.apache.org/jira/browse/KAFKA-14860
Project: Kafka
Issue Type: Bug
Components: kraft
Reporter: Proven Provenzano
Assignee: Proven Provenzano
The process of migration starts with creating controllers and adding them to
the Kafka cluster at which point they wait for the brokers to register in a
state ready for migrations. Brokers are then updated in a rolling manner to
enable migrations. When all brokers have enabled migration, then migration
starts.
After the metadata is migrated, the controller tries to connect back to the
brokers to signal that metadata migration is complete. If the controller cannot
create this connection (because of authentication issues) it will continuously
retry (no timeout). All metadata operations initiated by the brokers that have
not been contacted will fail. Rollback from this state back to ZK with no
controllers though is a manual step.
I want the controllers to create the connection back to the brokers after all
brokers have registered but before metadata migration is started. This will
prevent metadata operations from failing if the controller cannot connect to
any broker and allow for simple rollback of the cluster by disabling migration
in the controllers and restarting them in a rolling manner.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
