Hi All Our use case is that we need to process elements for the same key sequentially, and this processing involves async operations.
If any part of the processing fails, we store the offending and all subsequent incoming messages for that key in the state and not process any further messages for that key, until a retry succeeds or a human sends a 'skip' command message. diagram: https://mermaid.live/edit#pako:eNplkL1uwzAMhF-F0JQADrp76FR06tSOcQfWom3V-nFFqoUR591L20mXaqAOxHd3AC-mTZZMbfqM0wAvr00EfS62KbjYn-8C6Jui8DucTo_wmT4Oz97FEVhQqCtxXR13q_IBo-XzXWyehUc3LSu2Uyq2qIXpq2iyQ-9nmCjDSPMCmUISOuwfaEErLsVbw2272VOEDp0vmSqw5HEmC4GYsSeQpKjkv7j_buQ5tjAV4YeehOHHyQDsLAF1HbXCCyQZKB-2CTyzUOCjqUygHNBZPdxljW2MAoEaU6u0mMfGNPGqHBZJb1piasmFKlMmqxd7cqj3Dqbu0DNdfwHTGoek mermaid (in case mermaid.live goes down in the future): graph LR incoming[incoming events] --> job(Flink statefun job) commands[commands] -->|skip| job job --> |sequentially per key| remote(remote function) remote --> |on failure, delayed message to retry| remote remote --> |async puts/gets with side effects| other(other systems) Having the processing happen outside of Flink is nice-to-have from an independent scalability point of view, but is not strictly required. So long story short - no cyclic messaging, but also no way I can think of to use existing native Flink operators like async i/o (which when I last checked a few years back didn't have access to keyed state) P.S. Please note that there is already a pull request that has something to do wtih Flink 1.15, albeit without a description or a jira: https://github.com/apache/flink-statefun/pull/314 On Wed, 26 Oct 2022 at 19:54, Galen Warren <ga...@cvillewarrens.com> wrote: > Hi Gordon (and others), > > I'm also using this project for stateful messaging, including messaging > among functions. > > I've contributed a small amount of code in the past and have also enabled > Flink 1.15 compatibility in a local fork, so I might be able to help out > here. > > Thanks, > Galen > > On Wed, Oct 26, 2022 at 1:34 PM Ken Krugler <kkrugler_li...@transpac.com> > wrote: > > > Hi Gordon, > > > > We’re using it for stateful messaging, and also calling remote > > Python-based functions. > > > > So yes, also very interested in what is going to happen with the this > > subproject in the future. > > > > — Ken > > > > > > > > > Begin forwarded message: > > > > > > From: "Tzu-Li (Gordon) Tai" <tzuli...@apache.org> > > > Subject: Re: Stateful Functions with Flink 1.15 and onwards > > > Date: October 26, 2022 at 10:25:26 AM PDT > > > To: dev@flink.apache.org > > > Reply-To: dev@flink.apache.org > > > > > > Hi Filip, > > > > > > Thanks for bringing this up. > > > > > > The hard truth is that committers who were previously active on the > > > StateFun subproject, including myself, all currently have other > focuses. > > > Indeed, we may need to discuss with the community on how to proceed if > > > there seems to be no continued committer coverage. > > > > > > If it's just a matter of upgrading the supported Flink version, I'm > still > > > familiar enough with the subproject to probably be able to drive this > (or > > > if your team is up to it, I can assist you on that). > > > > > > For the long-term, as a data point I'm curious to see how many users > are > > > using StateFun in production today, and how you're using it? > > > > > > - Do your applications have arbitrary / cyclic / bi-directional > > > messaging between individual functions? > > > - Or are you utilizing StateFun simply to allow your stateful > functions > > > to run remotely as separate processes? > > > > > > If the majority is only the latter category, there might be a case to > > > support remote functions natively in Flink (which has been a discussion > > in > > > the past). > > > > > > Thanks, > > > Gordon > > > > > > On Wed, Oct 26, 2022 at 3:30 AM Filip Karnicki < > filip.karni...@gmail.com > > > > > > wrote: > > > > > >> Hi, I noticed that the development on stateful functions has come to a > > bit > > >> of a halt, with a pull request to update statefun to use Flink 1.15 > > being > > >> in the `open` state since May 2022. > > >> > > >> What do we think is the future of this sub-project? > > >> > > >> The background to this question is that my team is on a shared Flink > > >> cluster which will soon be upgrading to Flink 1.15. If I need to > > re-write > > >> all our code as a native Flink job (rather than a remote stateful > > function) > > >> then I need to get started right away. > > >> > > >> Many thanks, > > >> Fil > > >> > > > > -------------------------- > > Ken Krugler > > http://www.scaleunlimited.com > > Custom big data solutions > > Flink, Pinot, Solr, Elasticsearch > > > > > > > > >
