You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ratis.apache.org by GitBox <gi...@apache.org> on 2022/01/17 22:23:57 UTC

[GitHub] [ratis] adoroszlai opened a new pull request #581: RATIS-1490. Follower should not change to candidate if already stopped

adoroszlai opened a new pull request #581:
URL: https://github.com/apache/ratis/pull/581


   ## What changes were proposed in this pull request?
   
   `RaftServerImpl` creates a new `FollowerState` for each vote request received.  The old `FollowerState` may trigger change to candidate role even after it was shutdown by the server.  Both `RaftServerImpl#requestVote` and the change to candidate by `FollowerState` requires lock on `server` (`RaftServerImpl` object).  The order of events which triggers the problem:
   
    * `RaftServerImpl` executing `requestVote` [with lock on itself](https://github.com/apache/ratis/blob/b57be797fe97c00abc0e3d7353d30e1b2961714d/ratis-server/src/main/java/org/apache/ratis/server/impl/RaftServerImpl.java#L1099)
    * `FollowerState` [checks conditions, then waits for lock on `server`](https://github.com/apache/ratis/blob/b57be797fe97c00abc0e3d7353d30e1b2961714d/ratis-server/src/main/java/org/apache/ratis/server/impl/FollowerState.java#L126-L130)
    * `RaftServerImpl` [shuts down old `FollowerState`, starts new one](https://github.com/apache/ratis/blob/b57be797fe97c00abc0e3d7353d30e1b2961714d/ratis-server/src/main/java/org/apache/ratis/server/impl/RaftServerImpl.java#L475-L477)
    * `RaftServerImpl` [sets last RPC time on new `FollowerState`](https://github.com/apache/ratis/blob/b57be797fe97c00abc0e3d7353d30e1b2961714d/ratis-server/src/main/java/org/apache/ratis/server/impl/RaftServerImpl.java#L1116-L1117)
    * `RaftServerImpl` [releases lock](https://github.com/apache/ratis/blob/b57be797fe97c00abc0e3d7353d30e1b2961714d/ratis-server/src/main/java/org/apache/ratis/server/impl/RaftServerImpl.java#L1127)
    * *old* `FollowerState` [enters block with lock on `server`, triggers change to candidate](https://github.com/apache/ratis/blob/b57be797fe97c00abc0e3d7353d30e1b2961714d/ratis-server/src/main/java/org/apache/ratis/server/impl/FollowerState.java#L130-L138)
   
   Let `FollowerState` check if it is still running after entering the `synchronized` block.
   
   https://issues.apache.org/jira/browse/RATIS-1490
   
   ## How was this patch tested?
   
   Tweaked `FollowerState`, `RaftServerImpl` and `LeaderElectionTests` with code injection.  It required too much test-specific code, even in `synchronized` blocks, so I think it should not be part of this patch.  Ideas for unit tests welcome.
   
   Regular CI:
   https://github.com/adoroszlai/incubator-ratis/runs/4845475132


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: issues-unsubscribe@ratis.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

[GitHub] [ratis] codings-dan commented on pull request #581: RATIS-1490. Follower should not change to candidate if already stopped

Posted by GitBox <gi...@apache.org>.
codings-dan commented on pull request #581:
URL: https://github.com/apache/ratis/pull/581#issuecomment-1015119458


   I think this change is good, but it doesn't seem to be very easy to test it. Is it possible to consider adding a way to start Application in software like Intellij IDEA and start several ratis servers locally, which can easily add breakpoints and reproduce the problem, which is helpful for debugging and testing.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: issues-unsubscribe@ratis.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

[GitHub] [ratis] adoroszlai commented on pull request #581: RATIS-1490. Follower should not change to candidate if already stopped

Posted by GitBox <gi...@apache.org>.
adoroszlai commented on pull request #581:
URL: https://github.com/apache/ratis/pull/581#issuecomment-1015492855


   Thanks @codings-dan, @szetszwo for the review, @szetszwo for merging this.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: issues-unsubscribe@ratis.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

[GitHub] [ratis] szetszwo merged pull request #581: RATIS-1490. Follower should not change to candidate if already stopped

Posted by GitBox <gi...@apache.org>.
szetszwo merged pull request #581:
URL: https://github.com/apache/ratis/pull/581


   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: issues-unsubscribe@ratis.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org