[
https://issues.apache.org/jira/browse/IGNITE-18691?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Pavel Pereslegin updated IGNITE-18691:
--------------------------------------
Description:
When we mapping a query, we need to know who is the primary replica for each
partition.
Current implementation based on storing such leaders into an "ordered" list.
The element number in which corresponds to the partition number.
InternalTable has several methods to get current assignments (see
{{InternalTable#assignments}} and {{InternalTable#primaryReplicas}}).
Currently each of them explicitly sort result using partition numbers.
This doesn't seem to be optimal, since for each query we can call sort on the
same "set of replicas" multiple times (for each table).
One possible solution is to simply change the type of paritionMap inside
InternalTableImpl to some kind of sorted implementation, but the disadvantages
of this change should be carefully investigated.
was:
When we mapping a query, we need to know who is the primary replica for each
partition.
Current implementation based on storing such leaders into an "ordered" list.
The element number in which corresponds to the partition number.
InternalTable has several methods to get current assignments (see
{{InternalTable#assignments}} and {{InternalTable#primaryReplicas}}).
Currently each of them explicitly sort result using partition numbers.
This doesn't seem to be optimal since we're calling "assignments" for each
table while preparing a single query.
One possible solution is to simply change the type of paritionMap inside
InternalTableImpl to some kind of sorted implementation, but the disadvantages
of this change should be carefully investigated.
> Sql. Reduce overhead of sorting assignments.
> --------------------------------------------
>
> Key: IGNITE-18691
> URL: https://issues.apache.org/jira/browse/IGNITE-18691
> Project: Ignite
> Issue Type: Task
> Components: sql
> Reporter: Pavel Pereslegin
> Priority: Major
> Labels: ignite-3
>
> When we mapping a query, we need to know who is the primary replica for each
> partition.
> Current implementation based on storing such leaders into an "ordered" list.
> The element number in which corresponds to the partition number.
> InternalTable has several methods to get current assignments (see
> {{InternalTable#assignments}} and {{InternalTable#primaryReplicas}}).
> Currently each of them explicitly sort result using partition numbers.
> This doesn't seem to be optimal, since for each query we can call sort on the
> same "set of replicas" multiple times (for each table).
> One possible solution is to simply change the type of paritionMap inside
> InternalTableImpl to some kind of sorted implementation, but the
> disadvantages of this change should be carefully investigated.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
