[
https://issues.apache.org/jira/browse/IGNITE-21079?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexander Lapin updated IGNITE-21079:
-------------------------------------
Description:
h3. Motivation
Currently, PartitionReplicaListener does not switch its operations to another
thread pool when handling ReplicaRequests, so, most of the time, they are
handled in the messaging thread (the thread that is used to handle incoming
messages). There is only one such thread per Ignite node, and its disruption
might harm node liveness a great deal (for instance, it may make the node drop
off the Physical Topology due to inability to ack a ping in a timely manner).
h3. Definition of Done
Calls to storages may cause I/O and block on locks, so they should be avoided
in the messaging thread.
h3. Implementation Notes
was:
Currently, PartitionReplicaListener does not switch its operations to another
thread pool when handling ReplicaRequests, so, most of the time, they are
handled in the messaging thread (the thread that is used to handle incoming
messages). There is only one such thread per Ignite node, and its disruption
might harm node liveness a great deal (for instance, it may make the node drop
off the Physical Topology due to inability to ack a ping in a timely manner).
Calls to storages may cause I/O and block on locks, so they should be avoided
in the messaging thread.
> PartitionReplicaListener should not interact with storages in the messaging
> thread
> ----------------------------------------------------------------------------------
>
> Key: IGNITE-21079
> URL: https://issues.apache.org/jira/browse/IGNITE-21079
> Project: Ignite
> Issue Type: Improvement
> Reporter: Roman Puchkovskiy
> Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> Currently, PartitionReplicaListener does not switch its operations to another
> thread pool when handling ReplicaRequests, so, most of the time, they are
> handled in the messaging thread (the thread that is used to handle incoming
> messages). There is only one such thread per Ignite node, and its disruption
> might harm node liveness a great deal (for instance, it may make the node
> drop off the Physical Topology due to inability to ack a ping in a timely
> manner).
> h3. Definition of Done
> Calls to storages may cause I/O and block on locks, so they should be avoided
> in the messaging thread.
> h3. Implementation Notes
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
