[jira] [Commented] (KAFKA-14145) Faster propagation of high-watermark in KRaft topic partitions
[ https://issues.apache.org/jira/browse/KAFKA-14145?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17856152#comment-17856152 ] Josep Prat commented on KAFKA-14145: Changing target fix version to 3.9 since this is not a blocker and we are past code freeze > Faster propagation of high-watermark in KRaft topic partitions > -- > > Key: KAFKA-14145 > URL: https://issues.apache.org/jira/browse/KAFKA-14145 > Project: Kafka > Issue Type: Task > Components: kraft >Reporter: Jose Armando Garcia Sancio >Assignee: Jose Armando Garcia Sancio >Priority: Critical > Fix For: 3.8.0 > > > Typically, the HWM is increase after one round of Fetch requests from the > majority of the replicas. The HWM is propagated after another round of Fetch > requests. If the LEO doesn't change the propagation of the HWM can be delay > by one Fetch wait timeout (500ms). > Looking at the KafkaRaftClient implementation we would have to have an index > for both the fetch offset and the last sent high-watermark for that replica. > Another issue here is that we changed the KafkaRaftManager so that it doesn't > set the replica id when it is an observer/broker. Since the HWM is not part > of the Fetch request the leader would have to keep track of this in the > LeaderState. > {code:java} > val nodeId = if (config.processRoles.contains(ControllerRole)) { > OptionalInt.of(config.nodeId) > } else { > OptionalInt.empty() > } {code} > We would need to find a better solution for > https://issues.apache.org/jira/browse/KAFKA-13168 or improve the FETCH > request so that it includes the HWM. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (KAFKA-14145) Faster propagation of high-watermark in KRaft topic partitions
[ https://issues.apache.org/jira/browse/KAFKA-14145?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17800622#comment-17800622 ] Stanislav Kozlovski commented on KAFKA-14145: - Changing target fix version to 3.8 since this is not a blocker and we are cutting a 3.7 RC > Faster propagation of high-watermark in KRaft topic partitions > -- > > Key: KAFKA-14145 > URL: https://issues.apache.org/jira/browse/KAFKA-14145 > Project: Kafka > Issue Type: Task > Components: kraft >Reporter: Jose Armando Garcia Sancio >Assignee: Jose Armando Garcia Sancio >Priority: Critical > Fix For: 3.7.0 > > > Typically, the HWM is increase after one round of Fetch requests from the > majority of the replicas. The HWM is propagated after another round of Fetch > requests. If the LEO doesn't change the propagation of the HWM can be delay > by one Fetch wait timeout (500ms). > Looking at the KafkaRaftClient implementation we would have to have an index > for both the fetch offset and the last sent high-watermark for that replica. > Another issue here is that we changed the KafkaRaftManager so that it doesn't > set the replica id when it is an observer/broker. Since the HWM is not part > of the Fetch request the leader would have to keep track of this in the > LeaderState. > {code:java} > val nodeId = if (config.processRoles.contains(ControllerRole)) { > OptionalInt.of(config.nodeId) > } else { > OptionalInt.empty() > } {code} > We would need to find a better solution for > https://issues.apache.org/jira/browse/KAFKA-13168 or improve the FETCH > request so that it includes the HWM. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (KAFKA-14145) Faster propagation of high-watermark in KRaft topic partitions
[ https://issues.apache.org/jira/browse/KAFKA-14145?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17712957#comment-17712957 ] Mickael Maison commented on KAFKA-14145: We're past feature freeze for 3.5.0 so moving to 3.6.0. > Faster propagation of high-watermark in KRaft topic partitions > -- > > Key: KAFKA-14145 > URL: https://issues.apache.org/jira/browse/KAFKA-14145 > Project: Kafka > Issue Type: Task > Components: kraft >Reporter: Jose Armando Garcia Sancio >Assignee: Jose Armando Garcia Sancio >Priority: Critical > Fix For: 3.5.0 > > > Typically, the HWM is increase after one round of Fetch requests from the > majority of the replicas. The HWM is propagated after another round of Fetch > requests. If the LEO doesn't change the propagation of the HWM can be delay > by one Fetch wait timeout (500ms). > Looking at the KafkaRaftClient implementation we would have to have an index > for both the fetch offset and the last sent high-watermark for that replica. > Another issue here is that we changed the KafkaRaftManager so that it doesn't > set the replica id when it is an observer/broker. Since the HWM is not part > of the Fetch request the leader would have to keep track of this in the > LeaderState. > {code:java} > val nodeId = if (config.processRoles.contains(ControllerRole)) { > OptionalInt.of(config.nodeId) > } else { > OptionalInt.empty() > } {code} > We would need to find a better solution for > https://issues.apache.org/jira/browse/KAFKA-13168 or improve the FETCH > request so that it includes the HWM. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (KAFKA-14145) Faster propagation of high-watermark in KRaft topic partitions
[ https://issues.apache.org/jira/browse/KAFKA-14145?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17647815#comment-17647815 ] A. Sophie Blee-Goldman commented on KAFKA-14145: [~jagsancio] moving this to 3.5.0 since we are past code freeze for 3.4 > Faster propagation of high-watermark in KRaft topic partitions > -- > > Key: KAFKA-14145 > URL: https://issues.apache.org/jira/browse/KAFKA-14145 > Project: Kafka > Issue Type: Task > Components: kraft >Reporter: Jose Armando Garcia Sancio >Assignee: Jose Armando Garcia Sancio >Priority: Critical > Fix For: 3.5.0 > > > Typically, the HWM is increase after one round of Fetch requests from the > majority of the replicas. The HWM is propagated after another round of Fetch > requests. If the LEO doesn't change the propagation of the HWM can be delay > by one Fetch wait timeout (500ms). > Looking at the KafkaRaftClient implementation we would have to have an index > for both the fetch offset and the last sent high-watermark for that replica. > Another issue here is that we changed the KafkaRaftManager so that it doesn't > set the replica id when it is an observer/broker. Since the HWM is not part > of the Fetch request the leader would have to keep track of this in the > LeaderState. > {code:java} > val nodeId = if (config.processRoles.contains(ControllerRole)) { > OptionalInt.of(config.nodeId) > } else { > OptionalInt.empty() > } {code} > We would need to find a better solution for > https://issues.apache.org/jira/browse/KAFKA-13168 or improve the FETCH > request so that it includes the HWM. -- This message was sent by Atlassian Jira (v8.20.10#820010)