Hi Lalala, I think this approach can work as long as the generated query plan contains the same sub plan for the previous queries as before. Otherwise Flink won't be able to match the state to the operators of the plan. I think Timo and Dawid should know definitely whether this is possible or not.
Cheers, Till On Mon, Nov 23, 2020 at 10:33 AM lalala <lal...@activist.com> wrote: > Hi Kostas, > > Yes, that would satisfy my use case as the platform is always > future-oriented. Any arbitrary query is executed on the latest data. > > From your comment, I understand that even the session mode does not > optimize > our readers. I wish Flink could support arbitrary job submission and graph > generation in runtime, so we could submit jobs dynamically from main() as > we > do in Spark. > > If we want to group similar jobs, what would you recommend us for arbitrary > long-running jobs? Can we somehow take a snapshot of the queries running > under a job graph then resubmit them with the new query? > > I assume if we do the following under a single job(main method); > > ’’’ > Source: create table A... > Query1: select * from A > Query 2: select * from A > ’’’ > > Both queries will share the same reader as they are part of a single job > graph. Can we somehow take a snapshot of this and submit another query with > them again under the same job graph? > > I really appreciate your time for answering my questions, > > Best. > > > > > -- > Sent from: > http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/ >
