[
https://issues.apache.org/jira/browse/KAFKA-18788?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Abhinav Dixit updated KAFKA-18788:
----------------------------------
Description:
The current simple assignor doesn't offer perfect sharing between share
consumers of a share group. Hence, I came up with the following steps -
# depending on the no. of members and subscribed topic partitions (whether
they are greater, equal or lesser), we need to come up with a hard limit on
maximum and minimum no. of assignments for each members. There can also be a
range that we can come up with regarding no. of assignment of topic partitions.
There should also be a range of no. of members assigned to a topic partition.
# have a rebalancing as the fourth step in the assignment where we make sure
that point 1 is followed. This can be a random assignment for the first few
iterations but depending on those iteration results, we can have a concrete
assignment by moving partitions from one member to another.
> Improve simple assignor by having a max no. of partitions limit and min no.
> of partitions threshold assignment to members
> -------------------------------------------------------------------------------------------------------------------------
>
> Key: KAFKA-18788
> URL: https://issues.apache.org/jira/browse/KAFKA-18788
> Project: Kafka
> Issue Type: Sub-task
> Reporter: Abhinav Dixit
> Assignee: Abhinav Dixit
> Priority: Major
>
> The current simple assignor doesn't offer perfect sharing between share
> consumers of a share group. Hence, I came up with the following steps -
> # depending on the no. of members and subscribed topic partitions (whether
> they are greater, equal or lesser), we need to come up with a hard limit on
> maximum and minimum no. of assignments for each members. There can also be a
> range that we can come up with regarding no. of assignment of topic
> partitions. There should also be a range of no. of members assigned to a
> topic partition.
> # have a rebalancing as the fourth step in the assignment where we make sure
> that point 1 is followed. This can be a random assignment for the first few
> iterations but depending on those iteration results, we can have a concrete
> assignment by moving partitions from one member to another.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
