[
https://issues.apache.org/jira/browse/HADOOP-19450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17924629#comment-17924629
]
ASF GitHub Bot commented on HADOOP-19450:
-----------------------------------------
bhattmanish98 opened a new pull request, #7364:
URL: https://github.com/apache/hadoop/pull/7364
Description
-----------------------------------------------------------------------------------------------------------------------------------------------
CreatePath and RenamePath APIs are idempotent as subsequent retries on same
resource don’t change the server state[[1]](bookmark://rfcIdempotency).
However, when client experiences connection break on the CreatePath and the
RenamePath APIs, client cannot make sense if the request is accepted by the
server or not.
On connection failure, the client retries the request. The server might
return 404 (sourceNotFound) in case of RenamePath API and 409
(pathAlreadyExists) in case of CreatePath (overwrite=false) API. Now the client
doesn’t have a path forward. Reason being, in case of CreatePath, client
doesn’t know if the path was created on the original request or the path was
already there for some other request, in case of RenamePath, client doesn’t
know if the source was removed because of the original-try or it was not there
on the first place.
Proposed Solution
---------------------------------------------------------------------------------------------------------------------------------------------
Driver will send addition header "x-ms-client-transaction-id" which will
store by the server. In case first call fails because of time out and retry
happens and server throw source not found (in case of rename) and path already
exist (in case of create call). Driver will do list call on the path and check
whether the "x-ms-client-transaction-id" returned by server same as what driver
has at its end. In such case driver will return success to the caller.
> [ABFS] Rename/Create path idempotency client-level resolution
> -------------------------------------------------------------
>
> Key: HADOOP-19450
> URL: https://issues.apache.org/jira/browse/HADOOP-19450
> Project: Hadoop Common
> Issue Type: Task
> Components: fs/azure
> Affects Versions: 3.5.0
> Reporter: Manish Bhatt
> Assignee: Manish Bhatt
> Priority: Major
>
> CreatePath and RenamePath APIs are idempotent as subsequent retries on same
> resource don’t change the server state. However, when client experiences
> connection break on the CreatePath and the RenamePath APIs, client cannot
> make sense if the request is accepted by the server or not.
> On connection failure, the client retries the request. The server might
> return 404 (sourceNotFound) in case of RenamePath API and 409
> (pathAlreadyExists) in case of CreatePath (overwrite=false) API. Now the
> client doesn’t have a path forward. Reason being, in case of CreatePath,
> client doesn’t know if the path was created on the original request or the
> path was already there for some other request, in case of RenamePath, client
> doesn’t know if the source was removed because of the original-try or it was
> not there on the first place.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
