Jackie-Jiang commented on PR #9199:
URL: https://github.com/apache/pinot/pull/9199#issuecomment-1215987719

   > > This has the same behavior as setting `numInstancesPerPartition` to 1, 
and `numPartitions` to the partitions for the table. Currently we don't support 
explicitly setting partitions for `RealtimeSegmentAssignment`, so we should fix 
that instead of introducing this new strategy
   > 
   > I feel this is not exactly the same as `numInstancesPerPartition=1`, 
instead `numInstancesPerPartition` shall be the same as the number of replicas. 
In fact, the strategy is somewhat deterministic as the added stragy name 
suggests. And the main purpose is to make the assignment declarative and reduce 
the operational cost, e.g. we dont need to update `numInstancesPerPartition` 
when we increase `numPartitions` to scale out the table etc
   
   This new strategy is trying to put all segments from the same segment 
partition to the same server, which is exactly the same as what 
`numInstancesPerPartition = 1` is doing (and it is also deterministic).
   In order to achieve this, you should use the partition based replica-group 
assignment, and always put `numInstancesPerPartition = 1`. You may increase 
`numPartitions` if needed (should be very rare because it should be the same as 
table partitions, and changing it requires re-partitioning all the segments). 
Noted that `numPartitions` can be larger than the total instances within a 
replica-group. E.g. with 9 servers, you can have 3 replica-groups (3 servers 
per replica), then 18 partitions, and each server will host 6 partitions. You 
may scale it up to 18 servers, and run a rebalance without changing 
`numInstancesPerPartition` or `numPartitions`. After rebalancing, each server 
will host 3 partitions.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscr...@pinot.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@pinot.apache.org
For additional commands, e-mail: commits-h...@pinot.apache.org

Reply via email to