[
https://issues.apache.org/jira/browse/HDFS-17552?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17867635#comment-17867635
]
ASF GitHub Bot commented on HDFS-17552:
---------------------------------------
KeeProMise commented on code in PR #6888:
URL: https://github.com/apache/hadoop/pull/6888#discussion_r1685887046
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/ipc/Client.java:
##########
@@ -96,6 +96,8 @@ public class Client implements AutoCloseable {
private static final ThreadLocal<Integer> retryCount = new
ThreadLocal<Integer>();
private static final ThreadLocal<Object> EXTERNAL_CALL_HANDLER
= new ThreadLocal<>();
+ public static final ThreadLocal<CompletableFuture<Object>>
CALL_FUTURE_THREAD_LOCAL
+ = new ThreadLocal<>();
private static final ThreadLocal<AsyncGet<? extends Writable, IOException>>
ASYNC_RPC_RESPONSE = new ThreadLocal<>();
Review Comment:
@szetszwo Hi, thanks for you review! Do you mean that I need to delete this
Client.ASYNC_RPC_RESPONSE, and then use CompletableFuture in all places where
ASYNC_RPC_RESPONSE is used? Such modification may require repairing many unit
tests, because ASYNC_RPC_RESPONSE is used by many unit tests, and
ProtobufRpcEngine2 and ProtobufRpcEngine also use ASYNC_RPC_RESPONSE. If I
delete ASYNC_RPC_RESPONSE, then how should the ASYNC_RETURN_MESSAGE attribute
of ProtobufRpcEngine2 and ProtobufRpcEngine be processed? Is it also needed to
removed and use CompletableFuture, which may affect more unit tests and code.
> [ARR] IPC client uses CompletableFuture to support asynchronous operations.
> ---------------------------------------------------------------------------
>
> Key: HDFS-17552
> URL: https://issues.apache.org/jira/browse/HDFS-17552
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Jian Zhang
> Assignee: Jian Zhang
> Priority: Major
> Labels: pull-request-available
> Attachments: HDFS-17552.patch
>
>
> h3. Description
> In the implementation of asynchronous Ipc.client, the main methods used
> include HADOOP-13226, HDFS-10224, etc.
> However, the existing implementation does not support `CompletableFuture`;
> instead, it relies on setting up callbacks, which can lead to the "callback
> hell" problem. Using `CompletableFuture` can better organize asynchronous
> callbacks. Therefore, on the basis of the existing implementation, by using
> `CompletableFuture`, once the `client.call` is completed, the asynchronous
> thread handles the response of this call without blocking the main thread.
>
> *Test*
> new UT TestAsyncIPC#testAsyncCallWithCompletableFuture()
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
