[jira] [Comment Edited] (FLINK-24208) Allow idempotent savepoint triggering

["Austin Cawley-Edwards (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-24208?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17417718#comment-17417718 ] 

Austin Cawley-Edwards edited comment on FLINK-24208 at 9/20/21, 4:24 PM:
-------------------------------------------------------------------------

[~chesnay], definitely on the same page that the user provides a specific ID.

Maybe I did not understand the sketch, but from the current description I would assume sending the same `triggerId` if one is already exists would result in an error (like 409 Conflict), so it can be detected as a duplicate on the user side.

I think what we rather want to say (and what I thought the sketch implements) is that duplicate requests result in the same success response (i.e., 200 w/ a `TriggerResponse`). It does not matter to the user that another request with the same `triggerId` has already been sent.

 

I would suggest updating the description to:

 
 As a user of Flink, I want to be able to trigger a savepoint from an external system in a way that is safe to retry so I can guard against ephemeral errors, like network failures, and simplify my upgrade routines.

By passing a custom ID to the savepoint request, I can let Flink deduplicate the requests and not worry about storing the result or sending the request again.


was (Author: austince):
[~chesnay], definitely on the same page that the user provides a specific ID.

Maybe I did not understand the sketch, but from the current description I would assume sending the same `triggerId` if one is already in progress would result in an error (like 409 Conflict) so it can be detected as a duplicate.

I think what we rather want to say (and what I thought the sketch implements) is that duplicate requests result in the same success response (i.e., 200 w/ a `TriggerResponse`). It does not matter to the user that another request with the same `triggerId` has already been sent. 

 

I would suggest updating the description to:

 
As a user of Flink, I want to be able to trigger a savepoint from an external system in a way that is safe to retry so I can guard against ephemeral errors, like network failures, and simplify my upgrade routines.

By passing a custom ID to the savepoint request, I can let Flink deduplicate the requests and not worry about storing the result or sending the request again.

> Allow idempotent savepoint triggering
> -------------------------------------
>
>                 Key: FLINK-24208
>                 URL: https://issues.apache.org/jira/browse/FLINK-24208
>             Project: Flink
>          Issue Type: Improvement
>          Components: Runtime / Checkpointing, Runtime / REST
>            Reporter: Robert Metzger
>            Priority: Major
>
> As a user of Flink, I want to be able to trigger a savepoint from an external system in a way that I can detect if I have requested this savepoint already.
> By passing a custom ID to the savepoint request, I can check (in case of an error of the original request, or the external system) if the request has been made already.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)