[
https://issues.apache.org/jira/browse/FLINK-10578?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16653581#comment-16653581
]
Jiayi Liao commented on FLINK-10578:
------------------------------------
[~kkl0u] Thanks for pointing out the risks and problems.
I've read the proposal before and it's exactly what I need. So at first, I
tried to implement the operations on savepoints of FsStateBackend in bravo, but
the effort is too big because of the FsStateBackend's internal code structure,
and as the mails said, bravo should live as part of the Flink project to make
deeper intergration possible in the long run. That's why I give up using bravo
and try to solve it in another way.
What you said about breaking the fault-tolerace guarantees is really a problem
that I didn't think about too much before, I'll close this and maybe reopen it
if anyone really needs this in the future.
> Support writable state corresponding to queryable state
> -------------------------------------------------------
>
> Key: FLINK-10578
> URL: https://issues.apache.org/jira/browse/FLINK-10578
> Project: Flink
> Issue Type: New Feature
> Components: Queryable State, State Backends, Checkpointing
> Reporter: Jiayi Liao
> Assignee: Jiayi Liao
> Priority: Major
>
> I can see that now we support queryable state, why don't we also support the
> writable state? I've already done this on our own business to initialize the
> state in window during the program's first run. I think other companies may
> have other needs to modify the state from outside the job.
> The synchronization is a problem but from my perspective, very few objects
> will be affected because we can directly operate on the stateTable(for the
> FsStateBackend), and about the performance, I think it'll be okay if users
> know about it.
> I'm not very aware of what will be affected except what I said above, please
> let me know more if anyone has doubt about this.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
