[jira] [Commented] (IGNITE-21541) Avoid partition-operations pool when it not lead to starvation
[
https://issues.apache.org/jira/browse/IGNITE-21541?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17821989#comment-17821989
]
Vladislav Pyatkov commented on IGNITE-21541:
The issue with thread switching in the replica manager is also fixed in
IGNITE-20373.
> Avoid partition-operations pool when it not lead to starvation
> --
>
> Key: IGNITE-21541
> URL: https://issues.apache.org/jira/browse/IGNITE-21541
> Project: Ignite
> Issue Type: Improvement
>Reporter: Vladislav Pyatkov
>Assignee: Vladislav Pyatkov
>Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> Chnaging pools and related parking/unparking lead to an increase in latency.
> Sometimes we can avoid extra pool changing, for example, by doing it for
> embedded operations.
> {code:title=ReplicaManager#onReplicaMessageReceived}
> ExecutorService stripeExecutor =
> ReplicationGroupStripes.stripeFor(request.groupId(), requestsExecutor);
> stripeExecutor.execute(() -> handleReplicaRequest(request,
> senderConsistentId, correlationId));
> {code}
> This code changes a thread, even if it is not necessary.
> h3. Definition of done
> ReplicaManager should not switch threads if it does not lead to starvation
> (in my opinion, the splitting is needed only in the case of the network
> thread).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (IGNITE-21541) Avoid partition-operations pool when it not lead to starvation
[
https://issues.apache.org/jira/browse/IGNITE-21541?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17820021#comment-17820021
]
Vladislav Pyatkov commented on IGNITE-21541:
This issue may not be needed after the thread model is corrected.
> Avoid partition-operations pool when it not lead to starvation
> --
>
> Key: IGNITE-21541
> URL: https://issues.apache.org/jira/browse/IGNITE-21541
> Project: Ignite
> Issue Type: Improvement
>Reporter: Vladislav Pyatkov
>Assignee: Vladislav Pyatkov
>Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> Chnaging pools and related parking/unparking lead to an increase in latency.
> Sometimes we can avoid extra pool changing, for example, by doing it for
> embedded operations.
> {code:title=ReplicaManager#onReplicaMessageReceived}
> ExecutorService stripeExecutor =
> ReplicationGroupStripes.stripeFor(request.groupId(), requestsExecutor);
> stripeExecutor.execute(() -> handleReplicaRequest(request,
> senderConsistentId, correlationId));
> {code}
> This code changes a thread, even if it is not necessary.
> h3. Definition of done
> ReplicaManager should not switch threads if it does not lead to starvation
> (in my opinion, the splitting is needed only in the case of the network
> thread).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
