[ 
https://issues.apache.org/jira/browse/FLINK-9060?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16412978#comment-16412978
 ] 

ASF GitHub Bot commented on FLINK-9060:
---------------------------------------

Github user sihuazhou commented on the issue:

    https://github.com/apache/flink/pull/5751
  
    Hi @kl0u , I have changed the code to meet the comments on JIRA, but 
unfortunately, I found some case that could make the situation a bit complex 
with that architecture, that is maybe the `State.clear()` is not the last 
invocation on the state, for example:
    ```java
    void process() {
         /*stuff before state.clear() */
         
        state.clear()
    
        /*stuff after state.clear()*/
         Integer value = state.get(); // this should return the default value 
of ValueState or null if default value if not setted
         state.update(1); // this will re-add the state
    }
    ```
    
    So, if we want to wrapper the state, we need do some works on the almost 
state's almost every api, not only the `clear()`, and the `list` that was used 
to store the cleared `keys` now has to be changed to `set` (because we need 
check whether a `key` has been cleared and also maybe remove it from the 
cleared keyset). I think this maybe a bit expensive, now I am a bit like to go 
back the simplest way to store the `keys` into a `list`, and loop the list to 
perform `process()`, what do you think?
    
    



> Deleting state using KeyedStateBackend.getKeys() throws Exception
> -----------------------------------------------------------------
>
>                 Key: FLINK-9060
>                 URL: https://issues.apache.org/jira/browse/FLINK-9060
>             Project: Flink
>          Issue Type: Bug
>          Components: State Backends, Checkpointing
>            Reporter: Aljoscha Krettek
>            Assignee: Sihua Zhou
>            Priority: Blocker
>             Fix For: 1.5.0
>
>
> Adding this test to {{StateBackendTestBase}} showcases the problem:
> {code}
> @Test
> public void testConcurrentModificationWithGetKeys() throws Exception {
>       AbstractKeyedStateBackend<Integer> backend = 
> createKeyedBackend(IntSerializer.INSTANCE);
>       try {
>               ListStateDescriptor<String> listStateDescriptor =
>                       new ListStateDescriptor<>("foo", 
> StringSerializer.INSTANCE);
>               backend.setCurrentKey(1);
>               backend
>                       .getPartitionedState(VoidNamespace.INSTANCE, 
> VoidNamespaceSerializer.INSTANCE, listStateDescriptor)
>                       .add("Hello");
>               backend.setCurrentKey(2);
>               backend
>                       .getPartitionedState(VoidNamespace.INSTANCE, 
> VoidNamespaceSerializer.INSTANCE, listStateDescriptor)
>                       .add("Ciao");
>               Stream<Integer> keys = backend
>                       .getKeys(listStateDescriptor.getName(), 
> VoidNamespace.INSTANCE);
>               keys.forEach((key) -> {
>                       backend.setCurrentKey(key);
>                       try {
>                               backend
>                                       .getPartitionedState(
>                                               VoidNamespace.INSTANCE,
>                                               
> VoidNamespaceSerializer.INSTANCE,
>                                               listStateDescriptor)
>                                       .clear();
>                       } catch (Exception e) {
>                               e.printStackTrace();
>                       }
>               });
>       }
>       finally {
>               IOUtils.closeQuietly(backend);
>               backend.dispose();
>       }
> }
> {code}
> This should work because one of the use cases of {{getKeys()}} and 
> {{applyToAllKeys()}} is to do stuff for every key, which includes deleting 
> them.



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

Reply via email to