KAFKA-18966: Change controller_num_nodes_override to override_num_isolated_controllers #19194
Conversation
github-actions
bot
added
triage
chia7712
changed the title
|
@CalvinConfluent Could you please share the tests to me? I'd like to test it on my local to ensure it does reduce the test time. |
|
@chia7712 Thanks for checking. It is the transactions_test.py where Combined_kraft and hard_bounce are used. The transaction timeout is 40s, in my environment, the controller restart time(broker restart + controller starting handle ISR updates) is very close to 40s which makes the test flaky. |
CalvinConfluent
changed the title
| @@ -330,7 +327,7 @@ def __init__(self, context, num_nodes, zk, security_protocol=SecurityConfig.PLAI | |||
| if self.quorum_info.has_brokers: | |||
| num_nodes_broker_role = num_nodes | |||
| if self.quorum_info.has_controllers: | |||
| self.num_nodes_controller_role = self.num_kraft_controllers(num_nodes_broker_role, controller_num_nodes_override) | |||
| self.num_nodes_controller_role = self.num_kraft_controllers(num_nodes_broker_role, 0) | |||
There was a problem hiding this comment.
Pardon me, can assigning the value to 0 address the "Propose to set the controller node to at least 3"? It seems to me, it still depend on the num_nodes_broker_role, right?
if num_nodes_broker_role < 3:
return 1
if num_nodes_broker_role < 5:
return 3
return 5
https://github.com/apache/kafka/blob/trunk/tests/kafkatest/services/kafka/kafka.py#L495
There was a problem hiding this comment.
Yes, if the available nodes are less than 3, it only assigns 1 controller node. I can update the description.
There was a problem hiding this comment.
I may be mistaken, but I understand this pull request to be intended to increase the number of controller nodes, thereby preventing a complete controller service outage during a rolling bounce. Is that correct?
|
@chia7712 Can you help take another look? |
Bounce the broker that hosting the controller can cause unnecessary downtime for the partition. Avoid creating only 1 controller node in the combined kraft mode. For more details please check https://issues.apache.org/jira/browse/KAFKA-18966