[jira] [Commented] (KAFKA-16277) CooperativeStickyAssignor does not spread topics evenly among consumer group
[ https://issues.apache.org/jira/browse/KAFKA-16277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17821435#comment-17821435 ] Cameron Redpath commented on KAFKA-16277: - Thanks for the guidance! :) > CooperativeStickyAssignor does not spread topics evenly among consumer group > > > Key: KAFKA-16277 > URL: https://issues.apache.org/jira/browse/KAFKA-16277 > Project: Kafka > Issue Type: Bug > Components: clients, consumer >Reporter: Cameron Redpath >Assignee: Cameron Redpath >Priority: Major > Fix For: 3.8.0 > > Attachments: image-2024-02-19-13-00-28-306.png > > > Consider the following scenario: > `topic-1`: 12 partitions > `topic-2`: 12 partitions > > Of note, `topic-1` gets approximately 10 times more messages through it than > `topic-2`. > > Both of these topics are consumed by a single application, single consumer > group, which scales under load. Each member of the consumer group subscribes > to both topics. The `partition.assignment.strategy` being used is > `org.apache.kafka.clients.consumer.CooperativeStickyAssignor`. The > application may start with one consumer. It consumes all partitions from both > topics. > > The problem begins when the application scales up to two consumers. What is > seen is that all partitions from `topic-1` go to one consumer, and all > partitions from `topic-2` go to the other consumer. In the case with one > topic receiving more messages than the other, this results in a very > imbalanced group where one consumer is receiving 10x the traffic of the other > due to partition assignment. > > This is the issue being seen in our cluster at the moment. See this graph of > the number of messages being processed by each consumer as the group scales > from one to four consumers: > !image-2024-02-19-13-00-28-306.png|width=537,height=612! > Things to note from this graphic: > * With two consumers, the partitions for a topic all go to a single consumer > each > * With three consumers, the partitions for a topic are split between two > consumers each > * With four consumers, the partitions for a topic are split between three > consumers each > * The total number of messages being processed by each consumer in the group > is very imbalanced throughout the entire period > > With regard to the number of _partitions_ being assigned to each consumer, > the group is balanced. However, the assignment appears to be biased so that > partitions from the same topic go to the same consumer. In our scenario, this > leads to very undesirable partition assignment. > > I question if the behaviour of the assignor should be revised, so that each > topic has its partitions maximally spread across all available members of the > consumer group. In the above scenario, this would result in much more even > distribution of load. The behaviour would then be: > * With two consumers, 6 partitions from each topic go to each consumer > * With three consumers, 4 partitions from each topic go to each consumer > * With four consumers, 3 partitions from each topic go to each consumer > > Of note, we only saw this behaviour after migrating to the > `CooperativeStickyAssignor`. It was not an issue with the default partition > assignment strategy. > > It is possible this may be intended behaviour. In which case, what is the > preferred workaround for our scenario? Our current workaround if we decide to > go ahead with the update to `CooperativeStickyAssignor` may be to limit our > consumers so they only subscribe to one topic, and have two consumer threads > per instance of the application. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (KAFKA-16277) CooperativeStickyAssignor does not spread topics evenly among consumer group
[ https://issues.apache.org/jira/browse/KAFKA-16277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17820880#comment-17820880 ] A. Sophie Blee-Goldman commented on KAFKA-16277: FYI I added you as a contributor so you should be able to self-assign tickets from now on. Thanks again for contributing a fix! > CooperativeStickyAssignor does not spread topics evenly among consumer group > > > Key: KAFKA-16277 > URL: https://issues.apache.org/jira/browse/KAFKA-16277 > Project: Kafka > Issue Type: Bug > Components: clients, consumer >Reporter: Cameron Redpath >Assignee: Cameron Redpath >Priority: Major > Fix For: 3.8.0 > > Attachments: image-2024-02-19-13-00-28-306.png > > > Consider the following scenario: > `topic-1`: 12 partitions > `topic-2`: 12 partitions > > Of note, `topic-1` gets approximately 10 times more messages through it than > `topic-2`. > > Both of these topics are consumed by a single application, single consumer > group, which scales under load. Each member of the consumer group subscribes > to both topics. The `partition.assignment.strategy` being used is > `org.apache.kafka.clients.consumer.CooperativeStickyAssignor`. The > application may start with one consumer. It consumes all partitions from both > topics. > > The problem begins when the application scales up to two consumers. What is > seen is that all partitions from `topic-1` go to one consumer, and all > partitions from `topic-2` go to the other consumer. In the case with one > topic receiving more messages than the other, this results in a very > imbalanced group where one consumer is receiving 10x the traffic of the other > due to partition assignment. > > This is the issue being seen in our cluster at the moment. See this graph of > the number of messages being processed by each consumer as the group scales > from one to four consumers: > !image-2024-02-19-13-00-28-306.png|width=537,height=612! > Things to note from this graphic: > * With two consumers, the partitions for a topic all go to a single consumer > each > * With three consumers, the partitions for a topic are split between two > consumers each > * With four consumers, the partitions for a topic are split between three > consumers each > * The total number of messages being processed by each consumer in the group > is very imbalanced throughout the entire period > > With regard to the number of _partitions_ being assigned to each consumer, > the group is balanced. However, the assignment appears to be biased so that > partitions from the same topic go to the same consumer. In our scenario, this > leads to very undesirable partition assignment. > > I question if the behaviour of the assignor should be revised, so that each > topic has its partitions maximally spread across all available members of the > consumer group. In the above scenario, this would result in much more even > distribution of load. The behaviour would then be: > * With two consumers, 6 partitions from each topic go to each consumer > * With three consumers, 4 partitions from each topic go to each consumer > * With four consumers, 3 partitions from each topic go to each consumer > > Of note, we only saw this behaviour after migrating to the > `CooperativeStickyAssignor`. It was not an issue with the default partition > assignment strategy. > > It is possible this may be intended behaviour. In which case, what is the > preferred workaround for our scenario? Our current workaround if we decide to > go ahead with the update to `CooperativeStickyAssignor` may be to limit our > consumers so they only subscribe to one topic, and have two consumer threads > per instance of the application. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (KAFKA-16277) CooperativeStickyAssignor does not spread topics evenly among consumer group
[ https://issues.apache.org/jira/browse/KAFKA-16277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17819877#comment-17819877 ] Cameron Redpath commented on KAFKA-16277: - We have a PR ready for review with a small change as you suggested, plus updated expected assignment in existing tests It's failing CI, but I don't think it's related to our changes > CooperativeStickyAssignor does not spread topics evenly among consumer group > > > Key: KAFKA-16277 > URL: https://issues.apache.org/jira/browse/KAFKA-16277 > Project: Kafka > Issue Type: Bug > Components: clients, consumer >Reporter: Cameron Redpath >Priority: Major > Attachments: image-2024-02-19-13-00-28-306.png > > > Consider the following scenario: > `topic-1`: 12 partitions > `topic-2`: 12 partitions > > Of note, `topic-1` gets approximately 10 times more messages through it than > `topic-2`. > > Both of these topics are consumed by a single application, single consumer > group, which scales under load. Each member of the consumer group subscribes > to both topics. The `partition.assignment.strategy` being used is > `org.apache.kafka.clients.consumer.CooperativeStickyAssignor`. The > application may start with one consumer. It consumes all partitions from both > topics. > > The problem begins when the application scales up to two consumers. What is > seen is that all partitions from `topic-1` go to one consumer, and all > partitions from `topic-2` go to the other consumer. In the case with one > topic receiving more messages than the other, this results in a very > imbalanced group where one consumer is receiving 10x the traffic of the other > due to partition assignment. > > This is the issue being seen in our cluster at the moment. See this graph of > the number of messages being processed by each consumer as the group scales > from one to four consumers: > !image-2024-02-19-13-00-28-306.png|width=537,height=612! > Things to note from this graphic: > * With two consumers, the partitions for a topic all go to a single consumer > each > * With three consumers, the partitions for a topic are split between two > consumers each > * With four consumers, the partitions for a topic are split between three > consumers each > * The total number of messages being processed by each consumer in the group > is very imbalanced throughout the entire period > > With regard to the number of _partitions_ being assigned to each consumer, > the group is balanced. However, the assignment appears to be biased so that > partitions from the same topic go to the same consumer. In our scenario, this > leads to very undesirable partition assignment. > > I question if the behaviour of the assignor should be revised, so that each > topic has its partitions maximally spread across all available members of the > consumer group. In the above scenario, this would result in much more even > distribution of load. The behaviour would then be: > * With two consumers, 6 partitions from each topic go to each consumer > * With three consumers, 4 partitions from each topic go to each consumer > * With four consumers, 3 partitions from each topic go to each consumer > > Of note, we only saw this behaviour after migrating to the > `CooperativeStickyAssignor`. It was not an issue with the default partition > assignment strategy. > > It is possible this may be intended behaviour. In which case, what is the > preferred workaround for our scenario? Our current workaround if we decide to > go ahead with the update to `CooperativeStickyAssignor` may be to limit our > consumers so they only subscribe to one topic, and have two consumer threads > per instance of the application. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (KAFKA-16277) CooperativeStickyAssignor does not spread topics evenly among consumer group
[ https://issues.apache.org/jira/browse/KAFKA-16277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17819013#comment-17819013 ] Cameron Redpath commented on KAFKA-16277: - Thanks [~ableegoldman] for the response - yes we will try to submit a patch soon. > CooperativeStickyAssignor does not spread topics evenly among consumer group > > > Key: KAFKA-16277 > URL: https://issues.apache.org/jira/browse/KAFKA-16277 > Project: Kafka > Issue Type: Bug > Components: clients, consumer >Reporter: Cameron Redpath >Priority: Major > Attachments: image-2024-02-19-13-00-28-306.png > > > Consider the following scenario: > `topic-1`: 12 partitions > `topic-2`: 12 partitions > > Of note, `topic-1` gets approximately 10 times more messages through it than > `topic-2`. > > Both of these topics are consumed by a single application, single consumer > group, which scales under load. Each member of the consumer group subscribes > to both topics. The `partition.assignment.strategy` being used is > `org.apache.kafka.clients.consumer.CooperativeStickyAssignor`. The > application may start with one consumer. It consumes all partitions from both > topics. > > The problem begins when the application scales up to two consumers. What is > seen is that all partitions from `topic-1` go to one consumer, and all > partitions from `topic-2` go to the other consumer. In the case with one > topic receiving more messages than the other, this results in a very > imbalanced group where one consumer is receiving 10x the traffic of the other > due to partition assignment. > > This is the issue being seen in our cluster at the moment. See this graph of > the number of messages being processed by each consumer as the group scales > from one to four consumers: > !image-2024-02-19-13-00-28-306.png|width=537,height=612! > Things to note from this graphic: > * With two consumers, the partitions for a topic all go to a single consumer > each > * With three consumers, the partitions for a topic are split between two > consumers each > * With four consumers, the partitions for a topic are split between three > consumers each > * The total number of messages being processed by each consumer in the group > is very imbalanced throughout the entire period > > With regard to the number of _partitions_ being assigned to each consumer, > the group is balanced. However, the assignment appears to be biased so that > partitions from the same topic go to the same consumer. In our scenario, this > leads to very undesirable partition assignment. > > I question if the behaviour of the assignor should be revised, so that each > topic has its partitions maximally spread across all available members of the > consumer group. In the above scenario, this would result in much more even > distribution of load. The behaviour would then be: > * With two consumers, 6 partitions from each topic go to each consumer > * With three consumers, 4 partitions from each topic go to each consumer > * With four consumers, 3 partitions from each topic go to each consumer > > Of note, we only saw this behaviour after migrating to the > `CooperativeStickyAssignor`. It was not an issue with the default partition > assignment strategy. > > It is possible this may be intended behaviour. In which case, what is the > preferred workaround for our scenario? Our current workaround if we decide to > go ahead with the update to `CooperativeStickyAssignor` may be to limit our > consumers so they only subscribe to one topic, and have two consumer threads > per instance of the application. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (KAFKA-16277) CooperativeStickyAssignor does not spread topics evenly among consumer group
[ https://issues.apache.org/jira/browse/KAFKA-16277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17818594#comment-17818594 ] A. Sophie Blee-Goldman commented on KAFKA-16277: Hey [~credpath-seek] – first up, while it's been a while since I've looked at the sticky assignor code, I'm not too surprised that this might be the case. The obvious emphasis (per the name) was put on "stickiness" and partition-number balance, with good data parallelism ie topic-level balance being best-effort at most. That said, I suspect the assignor could be making a better effort. Presumably what is happening is that during the phase where it attempts to re-assign previously-owned partitions back to their former owner, we make a pass over a sorted list of previously-owned partitions that is grouped by topic. The assignor will then assign partitions from this list one-by-one to its previous owner until it hits the expected total number of partitions. So in the scenario you describe, it's basically looping over (t1p0, t1p1, t1p2, t1p3...t1pN, t2p0, t2p1, t2p2...t2pN) and assigning the first N partitions to the first consumer, which would be everything from topic 1, then just dumping the remaining partitions – all of which belong to topic 2 – onto the new consumer. The fix should be fairly simple – we just need to group this sorted list by partition, rather than by topic (ie t1p0, t2p0, t1p1, t2p1...t1pN, t2pN). Would you be interested in submitting a patch for this? As for what you can do right now: technically even if a fix for this was merged, you'd have to wait for the next release. However, the assignment is technically completely customizable, so in theory you could just copy/paste all the code from the patched assignor into a custom ConsumerPartitionAssignor implementation and then plug that in instead of the "cooperative-sticky" assignment strategy. Otherwise, the workaround you suggest is a reasonable backup – with the obvious downside being that the two threads will have an unbalanced load between them, at least the overall node-level workload will be more even > CooperativeStickyAssignor does not spread topics evenly among consumer group > > > Key: KAFKA-16277 > URL: https://issues.apache.org/jira/browse/KAFKA-16277 > Project: Kafka > Issue Type: Bug > Components: clients, consumer >Reporter: Cameron Redpath >Priority: Major > Attachments: image-2024-02-19-13-00-28-306.png > > > Consider the following scenario: > `topic-1`: 12 partitions > `topic-2`: 12 partitions > > Of note, `topic-1` gets approximately 10 times more messages through it than > `topic-2`. > > Both of these topics are consumed by a single application, single consumer > group, which scales under load. Each member of the consumer group subscribes > to both topics. The `partition.assignment.strategy` being used is > `org.apache.kafka.clients.consumer.CooperativeStickyAssignor`. The > application may start with one consumer. It consumes all partitions from both > topics. > > The problem begins when the application scales up to two consumers. What is > seen is that all partitions from `topic-1` go to one consumer, and all > partitions from `topic-2` go to the other consumer. In the case with one > topic receiving more messages than the other, this results in a very > imbalanced group where one consumer is receiving 10x the traffic of the other > due to partition assignment. > > This is the issue being seen in our cluster at the moment. See this graph of > the number of messages being processed by each consumer as the group scales > from one to four consumers: > !image-2024-02-19-13-00-28-306.png|width=537,height=612! > Things to note from this graphic: > * With two consumers, the partitions for a topic all go to a single consumer > each > * With three consumers, the partitions for a topic are split between two > consumers each > * With four consumers, the partitions for a topic are split between three > consumers each > * The total number of messages being processed by each consumer in the group > is very imbalanced throughout the entire period > > With regard to the number of _partitions_ being assigned to each consumer, > the group is balanced. However, the assignment appears to be biased so that > partitions from the same topic go to the same consumer. In our scenario, this > leads to very undesirable partition assignment. > > I question if the behaviour of the assignor should be revised, so that each > topic has its partitions maximally spread across all available members of the > consumer group. In the above scenario, this would result in much more even > distribution of load. The behaviour would then be: > * With two consumers, 6
