[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17495304#comment-17495304 ] Yuanbo Liu commented on HDFS-15180: --- Since it's suggested to split sub-tasks for this proposal, I'll mark it as a related jira, not a sub task. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Sub-task > Components: datanode >Affects Versions: 3.2.0 >Reporter: Qi Zhu >Assignee: Mingxiang Li >Priority: Major > Labels: pull-request-available > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > Time Spent: 1h > Remaining Estimate: 0h > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.20.1#820001) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17479090#comment-17479090 ] Yuanbo Liu commented on HDFS-15180: --- [~sodonnell] Thanks for your comments. There's a background that needs to be clarified. Nowadays, the storage machine becomes bigger and bigger. We've seen 12TB x 36 disks (which means 436TB of single datanode) in production environment. Global lock will be the key impact of IO performance, we'd be glad if this Jira has further progress to discuss or even be merged. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Sub-task > Components: datanode >Affects Versions: 3.2.0 >Reporter: Qi Zhu >Assignee: Mingxiang Li >Priority: Major > Labels: pull-request-available > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > Time Spent: 40m > Remaining Estimate: 0h > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.20.1#820001) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17450379#comment-17450379 ] Stephen O'Donnell commented on HDFS-15180: -- In Cloudera, we have not been looking into this issue actively, but it is an interesting one. We have went ahead with HDFS-15160 in our latest release and so far have not seen any problems from it. Our hope is the relatively minor change in HDFS-15160 can have a large benefit and is easy to disable with a config switch if any problems are detected. This change probably has a bigger impact that HDFS-15160, but is more complicated and so carries more risk. It is good to know you have been running it with no issue for some time - that does help give us more confidence there are no issues with the approach. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: Qi Zhu >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.20.1#820001) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17449864#comment-17449864 ] tomscut commented on HDFS-15180: This is a great idea to introduce fine-grained locking for datanode. We introduced [HDFS-15380|https://issues.apache.org/jira/browse/HDFS-15180] and [HDFS-15382|https://issues.apache.org/jira/browse/HDFS-15382] based on branch-3.1. Ran on our cold data cluster for a few months and it worked well. [~Aiphag0] [~hexiaoqiao] [~sodonnell] Are we still promoting this work? > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: Qi Zhu >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.20.1#820001) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17290808#comment-17290808 ] Qi Zhu commented on HDFS-15180: --- Thanks [~hexiaoqiao] for your reply, great work as expected, i will help review when it rebased trunk and look forward the benchmark result. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: Qi Zhu >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17290654#comment-17290654 ] Xiaoqiao He commented on HDFS-15180: Thanks [~zhuqi] for your pings. This improvement has deployed in our production for over half a year, and the result is really good as expected. Please reference to v004 if you are interested to this feature. IMO we still need some more works if plan to push this feature to trunk (e.g. some benchmark utils and regression tests, etc.) considering this improvement is aggressive. I believe [~Aiphag0] would like to push it forward if other guys also are interested to this improvement. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: Qi Zhu >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17289853#comment-17289853 ] Qi Zhu commented on HDFS-15180: --- [~hexiaoqiao] [~sodonnell] [~Aiphag0] Is it going on, i think we can push this now. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: Qi Zhu >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17118797#comment-17118797 ] Xiaoqiao He commented on HDFS-15180: Thanks [~sodonnell] for your positive feedback. {quote} we were hoping to get HDFS-15160 committed first, as its a smaller and possibly simpler change {quote} It makes sense to me. my colleague [~Aiphag0] and me would like to follow up and push this feature forward after HDFS-15160 has been ready. Welcome any suggestions and discussion if anyone is interested in this changes. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17118537#comment-17118537 ] Stephen O'Donnell commented on HDFS-15180: -- [~hexiaoqiao] We are interested in this change, but we were hoping to get HDFS-15160 committed first, as its a smaller and possibly simpler change. HDFS-15160 is ready from a review point of view, but we are waiting for some tests on production clusters to see if it brings any problems. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17116541#comment-17116541 ] Xiaoqiao He commented on HDFS-15180: [~sodonnell],[~weichiu],[~junping_du] are you interested in the improvement? In our production env we apply this feature based on branch-2.7 and it works well from our side. Look forward to more feedback or suggestions and push this feature forward. Thanks. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17080192#comment-17080192 ] Aiphago commented on HDFS-15180: Hi [~zhuqi] , thanks for your feedback.I think GenerationStamp may be change before we split block pool lock.And in our version we use wrtie lock in DataNode.transferReplicaForPipelineRecovery,this is diff the patch and may related to this problem. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17078881#comment-17078881 ] Hadoop QA commented on HDFS-15180: -- | (x) *{color:red}-1 overall{color}* | \\ \\ || Vote || Subsystem || Runtime || Comment || | {color:blue}0{color} | {color:blue} reexec {color} | {color:blue} 0m 0s{color} | {color:blue} Docker mode activated. {color} | | {color:red}-1{color} | {color:red} patch {color} | {color:red} 0m 9s{color} | {color:red} HDFS-15180 does not apply to trunk. Rebase required? Wrong Branch? See https://wiki.apache.org/hadoop/HowToContribute for help. {color} | \\ \\ || Subsystem || Report/Notes || | JIRA Issue | HDFS-15180 | | Console output | https://builds.apache.org/job/PreCommit-HDFS-Build/29095/console | | Powered by | Apache Yetus 0.8.0 http://yetus.apache.org | This message was automatically generated. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17078878#comment-17078878 ] zhuqi commented on HDFS-15180: -- cc [~Aiphag0] The Block in org.apache.hadoop.hdfs.protocol GenerationStamp and bytes action should change to synchronized. There may some cases that when holding the read lock, but we update the Block GenerationStamp at the same time. !image-2020-04-09-11-20-36-459.png! > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png, image-2020-04-09-11-20-36-459.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17065536#comment-17065536 ] Aiphago commented on HDFS-15180: ping [~sodonnell] , [~linyiqun], [~weichiu] Any advice ?Thanks. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17064192#comment-17064192 ] Aiphago commented on HDFS-15180: Fix the problem in UT > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, HDFS-15180.004.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17063806#comment-17063806 ] Aiphago commented on HDFS-15180: ok,I'll fix later > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, image-2020-03-10-17-22-57-391.png, > image-2020-03-10-17-31-58-830.png, image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17063805#comment-17063805 ] Xiaoqiao He commented on HDFS-15180: Not sure why Jenkins report not attach here. Just forward the result generated by Jenkins here and it seems there are unit tests failed, maybe it is related with this changes. [~Aiphag0] please take another check. Also some checkstyle warning need to fix. -1 overall |Vote|Subsystem|Runtime|Comment| |0|reexec|25m 55s|Docker mode activated.| | | | |Prechecks| |+1|@author|0m 0s|The patch does not contain any @author| | | | |tags.| |+1|test4tests|0m 0s|The patch appears to include 9 new or| | | | |modified test files.| | | | |trunk Compile Tests| |+1|mvninstall|19m 47s|trunk passed| |+1|compile|1m 0s|trunk passed| |+1|checkstyle|0m 48s|trunk passed| |+1|mvnsite|1m 6s|trunk passed| |+1|shadedclient|16m 8s|branch has no errors when building and| | | | |testing our client artifacts.| |+1|findbugs|2m 47s|trunk passed| |+1|javadoc|0m 39s|trunk passed| | | | |Patch Compile Tests| |+1|mvninstall|1m 2s|the patch passed| |+1|compile|0m 56s|the patch passed| |-1|javac|0m 56s|hadoop-hdfs-project_hadoop-hdfs| | | | |generated 1 new + 585 unchanged - 0| | | | |fixed = 586 total (was 585)| |-0|checkstyle|0m 43s|hadoop-hdfs-project/hadoop-hdfs: The| | | | |patch generated 33 new + 460 unchanged| | | | | - 1 fixed = 493 total (was 461)| |+1|mvnsite|1m 3s|the patch passed| |+1|whitespace|0m 0s|The patch has no whitespace issues.| |+1|shadedclient|14m 6s|patch has no errors when building and| | | | |testing our client artifacts.| |-1|findbugs|2m 57s|hadoop-hdfs-project/hadoop-hdfs| | | | |generated 1 new + 0 unchanged - 0 fixed| | | | |= 1 total (was 0)| |+1|javadoc|0m 37s|the patch passed| | | | |Other Tests| |-1|unit|110m 35s|hadoop-hdfs in the patch passed.| |+1|asflicense|0m 38s|The patch does not generate ASF| | | | |License warnings.| | | |200m 51s| ||Reason||Tests|| |FindBugs|module:hadoop-hdfs-project/hadoop-hdfs Should org.apache.hadoop.hdfs.server.datanode.BlockPoolLockManager$TrackLog be a _static_ inner class? At BlockPoolLockManager.java:inner class? At BlockPoolLockManager.java:[lines 62-92]| |Failed junit tests|hadoop.hdfs.server.datanode.TestBlockPoolLockManager| | |hadoop.hdfs.server.namenode.ha.TestHAFsck| | |hadoop.hdfs.server.datanode.TestBPOfferService| | |hadoop.hdfs.TestDecommissionWithStriped| | |hadoop.hdfs.server.datanode.fsdataset.impl.TestLazyPersistReplicaRecovery| | |hadoop.hdfs.server.datanode.TestBlockRecovery| ||Subsystem||Report/Notes|| |Docker|Client=19.03.8 Server=19.03.8 Image:yetus/hadoop:367833cf417| |JIRA Issue|HDFS-15180| |JIRA Patch URL|[^HDFS-15180.002.patch]| |Optional Tests|dupname asflicense compile javac javadoc mvninstall mvnsite unit shadedclient findbugs checkstyle| |uname|Linux 590078cb7ea7 4.15.0-74-generic #84-Ubuntu SMP Thu Dec 19 08:06:28 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux| |Build tool|maven| |Personality|/testptch/patchprocess/precommit/personality/provided.sh| |git revision|trunk / 36123170381| |maven|version: Apache Maven 3.6.0| |Default Java|1.8.0_242| |findbugs|v3.1.0-RC1| |javac|[https://builds.apache.org/job/PreCommit-HDFS-Build/28991/artifact/out/diff-compile-javac-hadoop-hdfs-project_hadoop-hdfs.txt]| |checkstyle|[https://builds.apache.org/job/PreCommit-HDFS-Build/28991/artifact/out/diff-checkstyle-hadoop-hdfs-project_hadoop-hdfs.txt]| |findbugs|[https://builds.apache.org/job/PreCommit-HDFS-Build/28991/artifact/out/new-findbugs-hadoop-hdfs-project_hadoop-hdfs.html]| |unit|[https://builds.apache.org/job/PreCommit-HDFS-Build/28991/artifact/out/patch-unit-hadoop-hdfs-project_hadoop-hdfs.txt]| |Test Results|[https://builds.apache.org/job/PreCommit-HDFS-Build/28991/testReport/]| |Max. process+thread count|2998 (vs. ulimit of 5500)| |modules|C: hadoop-hdfs-project/hadoop-hdfs U: hadoop-hdfs-project/hadoop-hdfs| |Console output|[https://builds.apache.org/job/PreCommit-HDFS-Build/28991/console]| |Powered by|Apache Yetus 0.8.0 [http://yetus.apache.org|http://yetus.apache.org/]| > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > HDFS-15180.003.patch, image-2020-03-10-17-22-57-391.png, > image-2020-03-10-17-31-58-830.png, image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17063403#comment-17063403 ] Xiaoqiao He commented on HDFS-15180: Try to trigger Jenkins. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17063204#comment-17063204 ] Aiphago commented on HDFS-15180: Hi [~zhuqi], thanks for valuable suggestions. Change the lock style use try() without finally{}. Change transferReplicaForPipelineRecovery to read lock. Wait UT result.[^HDFS-15180.002.patch] > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, HDFS-15180.002.patch, > image-2020-03-10-17-22-57-391.png, image-2020-03-10-17-31-58-830.png, > image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17060700#comment-17060700 ] zhuqi commented on HDFS-15180: -- Thanks [~Aiphag0] for your works and the POC patch. CC [~hexiaoqiao] [~Aiphag0] LGTM the POC patch. There are some suggestions that : First , we'd better to use Lock to implement AutoCloseableLock, so that consistent with the new read write lock in DataNode and can use try() without finally{} consisely. Second, the get replica information in DataNode#transferReplicaForPipelineRecovery should change to read lock. And i am looking forward to the volume level lock, and remove remain IO lock. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, image-2020-03-10-17-22-57-391.png, > image-2020-03-10-17-31-58-830.png, image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17059203#comment-17059203 ] Xiaoqiao He commented on HDFS-15180: Thanks [~Aiphag0] for your works and the POC patch. Hi [~zhuqi], I just assign this JIRA to [~Aiphag0], please feel free to assign back if you are interested to work with [~Aiphag0] together. cc [~sodonnell],[~zhuqi] any suggestions are welcome here and look forward to hear your feedback and comments for the POC solution. Thanks. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: Aiphago >Priority: Major > Attachments: HDFS-15180.001.patch, image-2020-03-10-17-22-57-391.png, > image-2020-03-10-17-31-58-830.png, image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17059199#comment-17059199 ] Aiphago commented on HDFS-15180: Hi [~zhuqi] ,Thanks for your proposal.And we have split dataset lock in our early version about 2.7 ,and gray deploy in our produce cluster for weeks.It looks like a good improvement in our version.But the trunck version looks big different from our version and have many works to do.And our idea is to split lock to blockpool at first, second we try to split each blockpool lock to volume lock, third we try to remove remain IO in lock as HDFS-15000 say.If you are interesting with this we can do this together.And here is the demo patch,and may have some problem. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > Attachments: image-2020-03-10-17-22-57-391.png, > image-2020-03-10-17-31-58-830.png, image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17056011#comment-17056011 ] zhuqi commented on HDFS-15180: -- Hi [~sodonnell] Yeah, your comment is just my mean. If we need add the top lock time information to the datanode metrics , so that may help to make the future performance improvement decision. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > Attachments: image-2020-03-10-17-22-57-391.png, > image-2020-03-10-17-31-58-830.png, image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17055798#comment-17055798 ] Stephen O'Donnell commented on HDFS-15180: -- [~zhuqi] Thanks for sharing this, it looks promising. It is also good to see the patch used in a real world cluster without any issues. Looking at your chart, are the orange and blue lines using the old code until around 5th / 6th March, then you switched to the new RW Fair lock plus HDFS-15160 and the blocked thread count has reduced to almost zero? The green line has been using HDFS-15160 since at least 4th March? I am not sure what metrics we should track to prove this change it good, but blocked thread count seems like a good one for now. Your chart looks promising. It would also be good to see a line on your chart for 2 or 3 nodes where HDFS-15160 is NOT applied for a comparison over time, so we can clearly see the nodes with HDFS-15160 against nodes without it. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > Attachments: image-2020-03-10-17-22-57-391.png, > image-2020-03-10-17-31-58-830.png, image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17055747#comment-17055747 ] zhuqi commented on HDFS-15180: -- CC [~sodonnell] And my version is based cdh 5.16.1, and my lock is fair lock. What is the key performance factor about the RW in DataNode you want to know, i can try to confirm the RW improvement when i next to gray more nodes in our busy cluster. May be we need more metrics to help the performance improvement confirm, what do you think about it? Thanks a lot. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > Attachments: image-2020-03-10-17-22-57-391.png, > image-2020-03-10-17-31-58-830.png, image-2020-03-10-17-34-26-368.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17055739#comment-17055739 ] zhuqi commented on HDFS-15180: -- !image-2020-03-10-17-22-57-391.png! > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > Attachments: image-2020-03-10-17-22-57-391.png > > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17055694#comment-17055694 ] Stephen O'Donnell commented on HDFS-15180: -- Hi [~zhuqi] - I just wanted to confirm from your last message - Are you actually running HDFS-15160 on a busy cluster and seeing a big improvement from it? Have you seen any issues or has it worked OK? If so, that is great feedback and thanks for trying it and sharing the results. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17055671#comment-17055671 ] zhuqi commented on HDFS-15180: -- [~sodonnell] , [~hexiaoqiao] Thanks for your patient reply. [~sodonnell] have done some work, ref. HDFS-15150 introduce read write lock and HDFS-15160 is in progress currently. I have used HDFS-15160 in our product cluster to gray, and now the blocked thread in datanode has been reduced a lot.-- [~hexiaoqiao] i am looking forward to the {{BlockPoolLockManager}} to split {{dataLock}} more fine-grained, i can assign to [~Aiphag0] anytime if he wants to take it. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17051952#comment-17051952 ] Xiaoqiao He commented on HDFS-15180: [~zhuqi] Thanks for your proposal and involve me here. It is very valuable suggestion, actually I [~sodonnell] have done some work, ref. HDFS-15150 introduce read write lock and HDFS-15160 is in progress currently. Of course, HDFS-14997 (please with HDFS-15113 together if backport) is another way to avoid heavy IO to impact interactive with NN. Beside these works, I believe there are some other ways to split the global lock. My colleague [~Aiphag0] try touse {{BlockPoolLockManager}} to split {{dataLock}} more fine-grained. {{BlockPoolLockManager}} represents rwlock pool with many rwlocks and it is more convenient for different BlockPools and different Disks to acquire lock and improve parallel read and write. This work is nearly finished recently, and gray deploy in our produce cluster. HDFS-15000 will trace this work. Thanks [~zhuqi] again. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement > Components: datanode >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17039209#comment-17039209 ] Stephen O'Donnell commented on HDFS-15180: -- We have some work on going related to HDFS-15150 to move the DN lock to a ReentrantReadWrite lock. There is further work to do to move various code paths to use the read lock to reduce contention. HDFS-15160 is a start on this, but there will need to be more Jiras. These changes are relatively simple and safe. I think there could be further gains made with finer grained locking, but those changes could be difficult to prove there are no race conditions. I would suggest we try to complete moving various code paths to use the new read lock and remove any places where the lock is held during disk IO, and then see how things look. That may give good improvements, but its hard to say until we do some benchmarks and see some real world use. Then we can consider finer grained locking. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-15180) DataNode FsDatasetImpl Fine-Grained Locking via BlockPool.
[ https://issues.apache.org/jira/browse/HDFS-15180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17039167#comment-17039167 ] zhuqi commented on HDFS-15180: -- cc [~sodonnell] , [~linyiqun], [~weichiu] , [~hexiaoqiao] What do you think about it. Can you give some advice ? Thanks. > DataNode FsDatasetImpl Fine-Grained Locking via BlockPool. > --- > > Key: HDFS-15180 > URL: https://issues.apache.org/jira/browse/HDFS-15180 > Project: Hadoop HDFS > Issue Type: Improvement >Affects Versions: 3.2.0 >Reporter: zhuqi >Assignee: zhuqi >Priority: Major > > Now the FsDatasetImpl datasetLock is heavy, when their are many namespaces in > big cluster. If we can split the FsDatasetImpl datasetLock via blockpool. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org