[ 
https://issues.apache.org/jira/browse/FLINK-8876?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16387729#comment-16387729
 ] 

ASF GitHub Bot commented on FLINK-8876:
---------------------------------------

Github user zentol commented on a diff in the pull request:

    https://github.com/apache/flink/pull/5645#discussion_r172509993
  
    --- Diff: 
flink-core/src/main/java/org/apache/flink/api/java/typeutils/runtime/kryo/KryoSerializer.java
 ---
    @@ -502,6 +572,22 @@ private void readObject(ObjectInputStream in) throws 
IOException, ClassNotFoundE
        // For testing
        // 
--------------------------------------------------------------------------------------------
     
    +   private void enterExclusiveThread() {
    +           // we use simple get, check, set here, rather than CAS
    +           // we don't need lock-style correctness, this is only a 
sanity-check and we thus
    +           // favor speed at the cost of some false negatives in this check
    +           Thread previous = currentThread;
    +           if (previous == null) {
    +                   currentThread = Thread.currentThread();
    +           } else {
    +                   throw new IllegalStateException("Serializer already 
accessed by thread " + previous.getName());
    --- End diff --
    
    you also have to check for equality between previous and 
Thread.currentThread, as copy() also calls serialize/deserialize().


> Improve concurrent access handling in stateful serializers
> ----------------------------------------------------------
>
>                 Key: FLINK-8876
>                 URL: https://issues.apache.org/jira/browse/FLINK-8876
>             Project: Flink
>          Issue Type: Improvement
>          Components: Core
>            Reporter: Stephan Ewen
>            Assignee: Stephan Ewen
>            Priority: Major
>             Fix For: 1.5.0, 1.6.0
>
>
> Some stateful serializers produce incorrect results when accidentally 
> accessed by multiple threads concurrently.
>  To better catch these cases, I suggest to add concurrency checks that are 
> active only when debug logging is enabled, and during test runs.
> This is inspired by Kryo's checks for concurrent access.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to