Hi Piyush, Thanks a lot for sharing the information. It would be a great relief that you are good with Flink on Mesos as is.
As for the jira issues, I believe the most essential ones should have already been resolved. You may find some remaining open issues here [1], but not all of them are necessary if we decide to keep Flink on Mesos as is. At the moment and in the short future, I think helps are mostly needed on testing the upcoming release 1.12 with Mesos use cases. The community is currently actively preparing the new release, and hopefully we could come up with a release candidate early next month. It would be greatly appreciated if you fork as experienced Flink on Mesos users can help with verifying the release candidates. Thank you~ Xintong Song [1] https://issues.apache.org/jira/browse/FLINK-17402?jql=project%20%3D%20FLINK%20AND%20component%20%3D%20%22Deployment%20%2F%20Mesos%22%20AND%20status%20%3D%20Open On Tue, Oct 27, 2020 at 2:58 AM Piyush Narang <p.nar...@criteo.com> wrote: > Hi Xintong, > > > > Do you have any jiras that cover any of the items on 1 or 2? I can reach > out to folks internally and see if I can get some folks to commit to > helping out. > > > > To cover the other qs: > > - Yes, we’ve not got a plan at the moment to get off Mesos. We use > Yarn for some our Flink workloads when we can. Mesos is only used when we > need streaming capabilities in our WW dcs (as our Yarn is centralized in > one DC) > - We’re currently on Flink 1.9 (old planner). We have a plan to bump > to 1.11 / 1.12 this quarter. > - We typically upgrade once every 6 months to a year (not every > release). We’d like to speed up the cadence but we’re not there yet. > - We’d largely be good with keeping Flink on Mesos as-is and > functional while missing out on some of the newer features. We understand > the pain on the communities side and we can take on the work if we see some > fancy improvement in Flink on Yarn / K8s that we want in Mesos to put in > the request to port it over. > > > > Thanks, > > > > -- Piyush > > > > > > *From: *Xintong Song <tonysong...@gmail.com> > *Date: *Sunday, October 25, 2020 at 10:57 PM > *To: *dev <d...@flink.apache.org>, user <user@flink.apache.org> > *Cc: *Lasse Nedergaard <lassenedergaardfl...@gmail.com>, < > p.nar...@criteo.com> > *Subject: *Re: [SURVEY] Remove Mesos support > > > > Thanks for sharing the information with us, Piyush an Lasse. > > > > @Piyush > > > > Thanks for offering the help. IMO, there are currently several problems > that make supporting Flink on Mesos challenging for us. > > 1. *Lack of Mesos experts.* AFAIK, there are very few people (if not > none) among the active contributors in this community that are > familiar with Mesos and can help with development on this component. > 2. *Absence of tests.* Mesos does not provide a testing cluster, like > `MiniYARNCluster`, making it hard to test interactions between Flink and > Mesos. We have only a few very simple e2e tests running on Mesos deployed > in a docker, covering the most fundamental workflows. We are not sure how > well those tests work, especially against some potential corner cases. > 3. *Divergence from other deployment.* Because of 1 and 2, the new > efforts (features, maintenance, refactors) tend to exclude Mesos if > possible. When the new efforts have to touch the Mesos related components > (e.g., changes to the common resource manager interfaces), we have to be > very careful and make as few changes as possible, to avoid accidentally > breaking anything that we are not familiar with. As a result, the component > diverges a lot from other deployment components (K8s/Yarn), which makes it > harder to maintain. > > It would be greatly appreciated if you can help with either of the above > issues. > > > > Additionally, I have a few questions concerning your use cases at Criteo. > IIUC, you are going to stay on Mesos in the foreseeable future, while > keeping the Flink version up-to-date? What Flink version are you currently > using? How often do you upgrade (e.g., every release)? Would you be good > with keeping the Flink on Mesos component as it is (means that deployment > and resource management improvements may not be ported to Mesos), while > keeping other components up-to-date (e.g., improvements from programming > APIs, operators, state backens, etc.)? > > > > Thank you~ > > Xintong Song > > > > > > On Sat, Oct 24, 2020 at 2:48 AM Lasse Nedergaard < > lassenedergaardfl...@gmail.com> wrote: > > Hi > > > > At Trackunit We have been using Mesos for long time but have now moved to > k8s. > > Med venlig hilsen / Best regards > > Lasse Nedergaard > > > > > > Den 23. okt. 2020 kl. 17.01 skrev Robert Metzger <rmetz...@apache.org>: > > > > Hey Piyush, > > thanks a lot for raising this concern. I believe we should keep Mesos in > Flink then in the foreseeable future. > > Your offer to help is much appreciated. We'll let you know once there is > something. > > > > On Fri, Oct 23, 2020 at 4:28 PM Piyush Narang <p.nar...@criteo.com> wrote: > > Thanks Kostas. If there's items we can help with, I'm sure we'd be able to > find folks who would be excited to contribute / help in any way. > > -- Piyush > > > On 10/23/20, 10:25 AM, "Kostas Kloudas" <kklou...@gmail.com> wrote: > > Thanks Piyush for the message. > After this, I revoke my +1. I agree with the previous opinions that we > cannot drop code that is actively used by users, especially if it > something that deep in the stack as support for cluster management > framework. > > Cheers, > Kostas > > On Fri, Oct 23, 2020 at 4:15 PM Piyush Narang <p.nar...@criteo.com> > wrote: > > > > Hi folks, > > > > > > > > We at Criteo are active users of the Flink on Mesos resource > management component. We are pretty heavy users of Mesos for scheduling > workloads on our edge datacenters and we do want to continue to be able to > run some of our Flink topologies (to compute machine learning short term > features) on those DCs. If possible our vote would be not to drop Mesos > support as that will tie us to an old release / have to maintain a fork as > we’re not planning to migrate off Mesos anytime soon. Is the burden > something that can be helped with by the community? (Or are you referring > to having to ensure PRs handle the Mesos piece as well when they touch the > resource managers?) > > > > > > > > Thanks, > > > > > > > > -- Piyush > > > > > > > > > > > > From: Till Rohrmann <trohrm...@apache.org> > > Date: Friday, October 23, 2020 at 8:19 AM > > To: Xintong Song <tonysong...@gmail.com> > > Cc: dev <d...@flink.apache.org>, user <user@flink.apache.org> > > Subject: Re: [SURVEY] Remove Mesos support > > > > > > > > Thanks for starting this survey Robert! I second Konstantin and > Xintong in the sense that our Mesos user's opinions should matter most > here. If our community is no longer using the Mesos integration, then I > would be +1 for removing it in order to decrease the maintenance burden. > > > > > > > > Cheers, > > > > Till > > > > > > > > On Fri, Oct 23, 2020 at 2:03 PM Xintong Song <tonysong...@gmail.com> > wrote: > > > > +1 for adding a warning in 1.12 about planning to remove Mesos > support. > > > > > > > > With my developer hat on, removing the Mesos support would > definitely reduce the maintaining overhead for the deployment and resource > management related components. On the other hand, the Flink on Mesos users' > voices definitely matter a lot for this community. Either way, it would be > good to draw users attention to this discussion early. > > > > > > > > Thank you~ > > > > Xintong Song > > > > > > > > > > > > On Fri, Oct 23, 2020 at 7:53 PM Konstantin Knauf <kna...@apache.org> > wrote: > > > > Hi Robert, > > > > +1 to the plan you outlined. If we were to drop support in Flink > 1.13+, we > > would still support it in Flink 1.12- with bug fixes for some time > so that > > users have time to move on. > > > > It would certainly be very interesting to hear from current Flink on > Mesos > > users, on how they see the evolution of this part of the ecosystem. > > > > Best, > > > > Konstantin > >
