[ https://issues.apache.org/jira/browse/HDDS-3459?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17088754#comment-17088754 ]
runzhiwang edited comment on HDDS-3459 at 4/23/20, 12:49 AM: ------------------------------------------------------------- I will share the opinion after I do a deep work. was (Author: yjxxtd): [~nanda] Hi, thanks for your explanation. My basic opinion is: 1. Distinguish the command according to their types. Such as close pipeline command and delete block command can be processed by different threads. 2. Monitor the time of thread process command, if some thread cost too long to process one command and the command is idempotent, we can interrupt and retry to process the command. Please correct me if I am wrong. And I will enrich the opinion after I do a deep work. > Datanode use a single thread to process the command of scm > ---------------------------------------------------------- > > Key: HDDS-3459 > URL: https://issues.apache.org/jira/browse/HDDS-3459 > Project: Hadoop Distributed Data Store > Issue Type: Bug > Reporter: runzhiwang > Assignee: runzhiwang > Priority: Major > Attachments: screenshot-1.png, screenshot-2.png, screenshot-3.png, > screenshot-4.png, screenshot-5.png, screenshot-6.png, screenshot-7.png > > > *What's the problem ?* > Datanode use a single [command-processor-thread > |https://github.com/apache/hadoop-ozone/blob/master/hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/statemachine/DatanodeStateMachine.java#L482] > to process the command of scm. Sometimes the thread maybe blocked a very > long time, then all the command from scm can not be processed, it maybe cause > some problem. > For example, a group has leader, follower1, follower2, Steps to reproduce the > problem are as following: > 1. Some datanodes crash, and follower2 begin streaming container data to > other datanode, then the command-processor-thread was blocked at > [cont.writeLock()|https://github.com/apache/hadoop-ozone/blob/master/hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/statemachine/commandhandler/DeleteBlocksCommandHandler.java#L130] > when it try to delete block, because streaming container data need to hold > the RwLock of container. > 2. follower2 report close pipeline > 3. scm send close pipeline command > 4. leader and follower1 remove group, but follower2 can not remove group > because the command-processor-thread was blocked. > 5. follower2 then begin LeaderElection about 12 hours until streaming > container data finish and release the RwLock, leader and follower1 response > group not found. > You can see find it in following screenshot. > 1. follower2 begin streaming container data from 2020-04-17 23:38:39 > !screenshot-1.png! > 2. follower2 report close pipeline 2020-01-48 01:14:39 > !screenshot-2.png! > 3. scm send close pipeline command > !screenshot-3.png! > !screenshot-4.png! > 4. leader remove group > !screenshot-5.png! > follower1 remove group > !screenshot-6.png! > 5. follower2 then begin LeaderElection about 12 hours until 2020-04-18 > 13:06:20. > !screenshot-7.png! -- This message was sent by Atlassian Jira (v8.3.4#803005) --------------------------------------------------------------------- To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org