Ellen Hui updated YARN-6413:
    Attachment: 0004-Registry-API-api-stubbed.patch

Hi [~jianhe], I've added some stubs, and a ServiceRecordKey implementation that 
just wraps the ZK path for the ZK implementation to use. The client can use it 
however they want, though I think it should be standardized one way or another.

My thought is this: the ZK implementation will implement RegistryOperations, 
RegistryStoreProtocol, and RegistryListenerProtocol. The RegistryOperations 
methods will be stubbed out, so anything that relies on them will still 
compile, and yarn-native-services will compile. After yarn-native-services 
moves over to to actually use the new APIs, RegistryOperations can then be 
fully deprecated.

I want to confirm with you that the methods in RegistryStoreProtocol and 
RegistryListenerProtocol, if I implement them and remove RegistryOperations 
entirely, will be ok for yarn-native-services. 

> Decouple Yarn Registry API from ZK
> ----------------------------------
>                 Key: YARN-6413
>                 URL: https://issues.apache.org/jira/browse/YARN-6413
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: amrmproxy, api, resourcemanager
>            Reporter: Ellen Hui
>            Assignee: Ellen Hui
>         Attachments: 0001-Registry-API-v2.patch, 0002-Registry-API-v2.patch, 
> 0003-Registry-API-api-only.patch, 0004-Registry-API-api-stubbed.patch
> Right now the Yarn Registry API (defined in the RegistryOperations interface) 
> is a very thin layer over Zookeeper. This jira proposes changing the 
> interface to abstract away the implementation details so that we can write a 
> FS-based implementation of the registry service, which will be used to 
> support AMRMProxy HA.
> The new interface will use register/delete/resolve APIs instead of 
> Zookeeper-specific operations like mknode. 

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org

Reply via email to