[jira] [Updated] (YARN-3545) Investigate the concurrency issue with the map of timeline collector
[ https://issues.apache.org/jira/browse/YARN-3545?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Vrushali C updated YARN-3545: - Parent Issue: YARN-7055 (was: YARN-5355) > Investigate the concurrency issue with the map of timeline collector > > > Key: YARN-3545 > URL: https://issues.apache.org/jira/browse/YARN-3545 > Project: Hadoop YARN > Issue Type: Sub-task >Reporter: Zhijie Shen >Assignee: Li Lu > Labels: YARN-5355, oct16-medium > Attachments: YARN-3545-YARN-2928.000.patch > > > See the discussion in YARN-3390 for details. Let's continue the discussion > here. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Updated] (YARN-3545) Investigate the concurrency issue with the map of timeline collector
[ https://issues.apache.org/jira/browse/YARN-3545?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Vrushali C updated YARN-3545: - Labels: YARN-5355 oct16-medium (was: YARN-5355) > Investigate the concurrency issue with the map of timeline collector > > > Key: YARN-3545 > URL: https://issues.apache.org/jira/browse/YARN-3545 > Project: Hadoop YARN > Issue Type: Sub-task >Reporter: Zhijie Shen >Assignee: Li Lu > Labels: YARN-5355, oct16-medium > Attachments: YARN-3545-YARN-2928.000.patch > > > See the discussion in YARN-3390 for details. Let's continue the discussion > here. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Updated] (YARN-3545) Investigate the concurrency issue with the map of timeline collector
[ https://issues.apache.org/jira/browse/YARN-3545?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Joep Rottinghuis updated YARN-3545: --- Parent Issue: YARN-5355 (was: YARN-2928) > Investigate the concurrency issue with the map of timeline collector > > > Key: YARN-3545 > URL: https://issues.apache.org/jira/browse/YARN-3545 > Project: Hadoop YARN > Issue Type: Sub-task >Reporter: Zhijie Shen >Assignee: Li Lu > Labels: YARN-5355 > Attachments: YARN-3545-YARN-2928.000.patch > > > See the discussion in YARN-3390 for details. Let's continue the discussion > here. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Updated] (YARN-3545) Investigate the concurrency issue with the map of timeline collector
[ https://issues.apache.org/jira/browse/YARN-3545?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Joep Rottinghuis updated YARN-3545: --- Labels: YARN-5355 (was: ) > Investigate the concurrency issue with the map of timeline collector > > > Key: YARN-3545 > URL: https://issues.apache.org/jira/browse/YARN-3545 > Project: Hadoop YARN > Issue Type: Sub-task >Reporter: Zhijie Shen >Assignee: Li Lu > Labels: YARN-5355 > Attachments: YARN-3545-YARN-2928.000.patch > > > See the discussion in YARN-3390 for details. Let's continue the discussion > here. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
[jira] [Updated] (YARN-3545) Investigate the concurrency issue with the map of timeline collector
[ https://issues.apache.org/jira/browse/YARN-3545?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Li Lu updated YARN-3545: Attachment: YARN-3545-YARN-2928.000.patch In this patch I'm using concurrent hash map to replace synchronized hash map. After removing the global lock, we need to consider two cases, concurrent putIfAbsent calls, and concurrent putIfAbsent call and get call. The case with concurrent putIfAbsent call and get call is addressed by a initialization barrier since the contention is low. With this solution on the best case each read will only have one volatile variable read, instead of getting the lock inside synchronized map. The case with multiple concurrent putIfAbsents is addressed by speculatively allocate a collector, and try to putIfAbsent it to the hash map. It then call postPut and publish this new collector to all readers if the putIfAbsent call succeed (returns null). If the putIfAbsent call failed, someone else has already allocated a collector and we need to use that collector. To speed up this case, I used a "fast path" such that the putIfAbsent call only tries to allocate collectors if there was no collector for it at the beginning of this method. I'd appreciate comments since I may miss something here... > Investigate the concurrency issue with the map of timeline collector > > > Key: YARN-3545 > URL: https://issues.apache.org/jira/browse/YARN-3545 > Project: Hadoop YARN > Issue Type: Sub-task >Reporter: Zhijie Shen >Assignee: Li Lu > Attachments: YARN-3545-YARN-2928.000.patch > > > See the discussion in YARN-3390 for details. Let's continue the discussion > here. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Updated] (YARN-3545) Investigate the concurrency issue with the map of timeline collector
[ https://issues.apache.org/jira/browse/YARN-3545?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Zhijie Shen updated YARN-3545: -- Issue Type: Sub-task (was: Bug) Parent: YARN-2928 > Investigate the concurrency issue with the map of timeline collector > > > Key: YARN-3545 > URL: https://issues.apache.org/jira/browse/YARN-3545 > Project: Hadoop YARN > Issue Type: Sub-task >Reporter: Zhijie Shen >Assignee: Li Lu > > See the discussion in YARN-3390 for details. Let's continue the discussion > here. -- This message was sent by Atlassian JIRA (v6.3.4#6332)