[jira] [Commented] (IGNITE-9449) Lazy unmarshalling of discovery events in TcpDiscovery
[ https://issues.apache.org/jira/browse/IGNITE-9449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16638168#comment-16638168 ] Nikolay Izhikov commented on IGNITE-9449: - Thanks, [~yzhdanov]. Waiting for your design. > Lazy unmarshalling of discovery events in TcpDiscovery > -- > > Key: IGNITE-9449 > URL: https://issues.apache.org/jira/browse/IGNITE-9449 > Project: Ignite > Issue Type: Improvement > Components: cache >Affects Versions: 2.4, 2.5, 2.6 >Reporter: Pavel Kovalenko >Priority: Major > Fix For: 2.8 > > > Currently disco-msg-worker thread spend major part of time on disocvery > message unmarshalling before send it to the next node. In most cases this is > unnecessary and message can be send immediately after receiving and notyfing > discovery-event-worker. > Responsibility of unmarshalling should moved to discovery-event-worker thread > and this improvement will significantly reduce latency of sending custom > messages across ring. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (IGNITE-9449) Lazy unmarshalling of discovery events in TcpDiscovery
[ https://issues.apache.org/jira/browse/IGNITE-9449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16638154#comment-16638154 ] Yakov Zhdanov commented on IGNITE-9449: --- [~NIzhikov] Nikolay, this is pretty complex issue and it seems this requires in-depth understanding of discovery protocol. I also think that this change should be accompanied by some more changes to discovery which may be a subject for new IEP. Can you please switch to another ticket for now and give me couple of weeks to summarize my ideas? Thanks! Yakov > Lazy unmarshalling of discovery events in TcpDiscovery > -- > > Key: IGNITE-9449 > URL: https://issues.apache.org/jira/browse/IGNITE-9449 > Project: Ignite > Issue Type: Improvement > Components: cache >Affects Versions: 2.4, 2.5, 2.6 >Reporter: Pavel Kovalenko >Assignee: Nikolay Izhikov >Priority: Major > Fix For: 2.8 > > > Currently disco-msg-worker thread spend major part of time on disocvery > message unmarshalling before send it to the next node. In most cases this is > unnecessary and message can be send immediately after receiving and notyfing > discovery-event-worker. > Responsibility of unmarshalling should moved to discovery-event-worker thread > and this improvement will significantly reduce latency of sending custom > messages across ring. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (IGNITE-9449) Lazy unmarshalling of discovery events in TcpDiscovery
[ https://issues.apache.org/jira/browse/IGNITE-9449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16638124#comment-16638124 ] Nikolay Izhikov commented on IGNITE-9449: - Hello, [~yzhdanov] Can you, please, answer, my questions? See it below ^ > Lazy unmarshalling of discovery events in TcpDiscovery > -- > > Key: IGNITE-9449 > URL: https://issues.apache.org/jira/browse/IGNITE-9449 > Project: Ignite > Issue Type: Improvement > Components: cache >Affects Versions: 2.4, 2.5, 2.6 >Reporter: Pavel Kovalenko >Assignee: Nikolay Izhikov >Priority: Major > Fix For: 2.8 > > > Currently disco-msg-worker thread spend major part of time on disocvery > message unmarshalling before send it to the next node. In most cases this is > unnecessary and message can be send immediately after receiving and notyfing > discovery-event-worker. > Responsibility of unmarshalling should moved to discovery-event-worker thread > and this improvement will significantly reduce latency of sending custom > messages across ring. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (IGNITE-9449) Lazy unmarshalling of discovery events in TcpDiscovery
[ https://issues.apache.org/jira/browse/IGNITE-9449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16638037#comment-16638037 ] Nikolay Izhikov commented on IGNITE-9449: - Hello, [~agoncharuk]. Can you, please, answer, my questions? > Lazy unmarshalling of discovery events in TcpDiscovery > -- > > Key: IGNITE-9449 > URL: https://issues.apache.org/jira/browse/IGNITE-9449 > Project: Ignite > Issue Type: Improvement > Components: cache >Affects Versions: 2.4, 2.5, 2.6 >Reporter: Pavel Kovalenko >Assignee: Nikolay Izhikov >Priority: Major > Fix For: 2.8 > > > Currently disco-msg-worker thread spend major part of time on disocvery > message unmarshalling before send it to the next node. In most cases this is > unnecessary and message can be send immediately after receiving and notyfing > discovery-event-worker. > Responsibility of unmarshalling should moved to discovery-event-worker thread > and this improvement will significantly reduce latency of sending custom > messages across ring. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (IGNITE-9449) Lazy unmarshalling of discovery events in TcpDiscovery
[ https://issues.apache.org/jira/browse/IGNITE-9449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16637331#comment-16637331 ] Nikolay Izhikov commented on IGNITE-9449: - Hello, [~Jokser]. I want to investigate this issue. Can you help me with it? If I understand you correctly, you described following a piece of code ServerImpl line 3234 {code:java} spi.writeToSocket(sock, out, pendingMsg, timeoutHelper.nextTimeoutChunk( spi.getSocketTimeout())); {code} which internally serialize `pendingMsg` with a configured marshaller. Do you have some numbers showing an impact of this marshalling? Do you have a benchmark or similar that we can use as an indicator of an issue? > Lazy unmarshalling of discovery events in TcpDiscovery > -- > > Key: IGNITE-9449 > URL: https://issues.apache.org/jira/browse/IGNITE-9449 > Project: Ignite > Issue Type: Improvement > Components: cache >Affects Versions: 2.4, 2.5, 2.6 >Reporter: Pavel Kovalenko >Assignee: Nikolay Izhikov >Priority: Major > Fix For: 2.8 > > > Currently disco-msg-worker thread spend major part of time on disocvery > message unmarshalling before send it to the next node. In most cases this is > unnecessary and message can be send immediately after receiving and notyfing > discovery-event-worker. > Responsibility of unmarshalling should moved to discovery-event-worker thread > and this improvement will significantly reduce latency of sending custom > messages across ring. -- This message was sent by Atlassian JIRA (v7.6.3#76005)