Hi Zanini! Welcome back, hope you had a great time during your PTO. If I understand correctly, the way you see the integration between repos relies heavily on timestamped SNAPSHOTs. This way: 1. `kogito-images` would stop building `kogito-apps`, and would rely on timestamped SNAPSHOTs being published from `drools/optaplanner/kogito-runtimes/kogito-apps`. 2. `kie-tools` would start publishing timestamped SNAPSHOTs for its UI components, so they would be consumed downstream by `kogito-serverless-operator`.
If that's what you mean, let me elaborate on some of the consequences of this integration model, where we stop using the mutable 999-SNAPSHOT and start relying more on timestamped SNAPSHOTs for cross-repo synchronization: a. Cross-repo PRs would not be possible anymore, depending on which repos are involved in the ensemble. This would create boundaries between certain groups of our repos, namely: drools/optaplanner/kogito-runtimes/kogito-apps --> kogito-images --> kie-tools, --> kogito-serverless-operator. So 4 "independent clusters". b. kogito-images, kie-tools, and kogito-serverless-operator, using timestamped SNAPSHOTs between themselves, would make the most upstream of the three (in this case, `kogito-images`), to be the one defining what timestamped SNAPSHOT version we should use for its upstream dependencies (drools/optaplanner/kogito-runtimes/kogito-apps). In more details: If `kogito-images` depends on `drools/optaplanner/kogito-runtimes/kogito-apps` using version 999-20240501, that's the version `kie-tools` and `kogito-serverless-operator` need to use as well, as the drools/optaplanner/kogito-runtimes/kogito-apps would come transitively from the `kogito-images` dependency these two repos would have. This, IMHO, causes a lot of confusion, since the dates on the timestamped SNAPSHOT versions have implicit transitive dependencies to other upstream projects that also have their own timestamped SNAPSHOTs published. I.e., 999-20240501 of `kie-tools` won't be aligned with 999-2024051 of `drools/optaplanner/kogito-runtimes/kogito-apps`. c. Updating dependencies across the board would be much more difficult, as we would need to rely on a chain reaction starting from the `drools/optaplanner/kogito-runtimes/kogito-apps` cluster all the way down to `kogito-serverless-operator`, releasing timestamped SNAPSHOTs and sending PRs for each "cluster" along the way. I'm not sure this is the way we want to move forward, to be honest, as these consequences need to be thoroughly examined by people in the community, so that all the side-effects of this choice are clear to everyone. To conclude, having `kie-tools` building a "template image" to verify that it will not break `kogito-serverless-operator` downstream, and having `kogito-serverless-operator` building parts of `kie-tools` to incorporate the latest changes, IMHO, is: 1. Duplication of building logic. IMHO, we can't have individual repos creating its own up- and downstream CI systems. 2. Conflicting with the timestamped SNAPSHOTs strategy for integrating repos, where boundaries are clearly defined. 3. Wasteful use of resources (same that is happening with `kogito-images` building `kogito-apps` right now). I'm sorry for bringing all this to this seemingly simple proposal, but I'm afraid taking steps without knowing where we're going isn't the way to move forward. Please let me know if I misunderstood something. On Wed, May 8, 2024 at 3:15 PM ricardo zanini fernandes <[email protected]> wrote: > > Hi! > > Sorry for the late reply, I was on PTO. > > Replies inline. > > > The `kie-tools` CI is not prepared to build > `kogito-serverless-operator`, so the way `kogito-serverless-operator` > references the SonataFlow Quarkus Dev UI will be important to > establish the boundaries between both repos. To further develop the > SonataFlow Quarkus Dev UI and have its changes reflect on the > `kogito-swf-devmode` image, we need to come up with a strategy that is > both safe, consistent, and enforces correctness. There's also the fact > that currently `kie-tools` depends on timestamped SNAPSHOTs of > Kogito/Drools, while `kogito-serverless-operator` uses 999-SNAPSHOT, > if I'm not mistaken. Can you elaborate a little bit more on how you > see this reference being done? Please consider cross-repo PRs for big > migrations like the Quarkus 3.8.4 that is currently happening. > > This will make us discuss yet again the need of snapshots for the UI > components used by the images. The images must be the end of the build > pipeline, where we aggregate every component we ship in a single instance > to release. Yet, they are a core part of the cloud platform, hence part of > the Operator BDD, testing, integration, and delivering. Intermediate repos > can't depend on them unless they are also responsible for maintenance and > release like what we're doing with the consoles. > > Also, the images can have a respin anytime we need a new component bump. We > do this all the time to fix CVEs or bug fixes in a specific component that > is part of the image. > > The devui can be developed and maintained within the kie-tools repo and > have tests that verify that component. The integration will be made on the > Images/Operator side once we grab a latest snapshot. Alternatively, we can > do: > 1. The tools CI can fetch a template image, inject the component, and run > the required tests > 2. The image/operator CI can fetch and build the UI components and run the > integration tests > > > Also, regarding point "4. Review the PR GHA checks to run CLI tests > once there's a change in the cmd module" of the proposed EPIC, I think > we might run into problems, since the `cmd` module also depends on the > `api` and `workflowproj` modules of `kogito-serverless-operator.` I'm > afraid changes made to these two modules would also need to trigger a > build of the `cmd` module, and they can potentially break it. > > Sorry to not get into details there, but the new cmd module will be part of > the Operator's workspace, any changes in any cross dependencies there will > trigger the CI checks, as we are doing today in the builder and > workflowproj modules. It will be way easier to maintain the CLI. > > > These considerations alone, IMHO, expose one of the biggest challenges > we have in our community right now, which is that the definition and > implementation of the dependency graph between repos/modules/packages > is currently spread across many different "build systems", like the > new proposed GHA jobs exclusive to the `kogito-serverless-operator` > repo, the Build Chain system we have for the Drools/Kogito repos, the > `kie-tools` CI, and the many Jenkins jobs we have on Apache's Jenkins. > There's also the fact that we have `kogito-images` selectively > building parts of `kogito-apps` during its own build to include them > in the images it produces. > > I agree. We would need to sit together and solve this as a team, having a > nice integration across every repo. The kogito-images building parts of the > apps is something we currently working on, IIRC. The apps will deploy a > snapshot as part of the nightlies and the images will be using it. > > > I just think it is important to highlight that this proposal would > only address a LOCAL problem related exclusively to the SonataFlow > section of the Apache KIE community, while, at the same time, not > contributing to reducing the segmentation of the Apache KIE community > as a whole. > > Yes, the proposal is to starting addressing the local Apache KIE SonataFlow > Cloud platform first. But I agree that we need to refactor our CI as a > whole, which is something we should start in a new thread. > > Cheers! > > On Mon, Apr 22, 2024 at 10:02 PM Tiago Bento <[email protected]> wrote: > > > Zanini and Alex, > > > > The task we agreed on for after releasing Apache KIE 10 is > > https://github.com/apache/incubator-kie-issues/issues/1040. It only > > describes deleting the temporary copies we'll have on KIE Tools and > > reverting things back to where they were, using the fixed 10.0.0 > > version. > > > > Moving `kn-plugin-workflow` and the `kogito-swf-{devmode,builder}` > > images to `kogito-serverless-operator` would be a new move, which I > > understand is the scope of this proposal thread. > > > > Doing so, however, will make `kogito-serverless-operator` depend on > > `kie-tools`, since the SonataFlow Quarkus Dev UI is hosted there now, > > and it is a dependency of the `kogito-swf-devmode` image. > > > > I'm saying this because I think we need to further discuss the > > consequences of this change... > > > > The `kie-tools` CI is not prepared to build > > `kogito-serverless-operator`, so the way `kogito-serverless-operator` > > references the SonataFlow Quarkus Dev UI will be important to > > establish the boundaries between both repos. To further develop the > > SonataFlow Quarkus Dev UI and have its changes reflect on the > > `kogito-swf-devmode` image, we need to come up with a strategy that is > > both safe, consistent, and enforces correctness. There's also the fact > > that currently `kie-tools` depends on timestamped SNAPSHOTs of > > Kogito/Drools, while `kogito-serverless-operator` uses 999-SNAPSHOT, > > if I'm not mistaken. Can you elaborate a little bit more on how you > > see this reference being done? Please consider cross-repo PRs for big > > migrations like the Quarkus 3.8.4 that is currently happening. > > > > Also, regarding point "4. Review the PR GHA checks to run CLI tests > > once there's a change in the cmd module" of the proposed EPIC, I think > > we might run into problems, since the `cmd` module also depends on the > > `api` and `workflowproj` modules of `kogito-serverless-operator.` I'm > > afraid changes made to these two modules would also need to trigger a > > build of the `cmd` module, and they can potentially break it. > > > > These considerations alone, IMHO, expose one of the biggest challenges > > we have in our community right now, which is that the definition and > > implementation of the dependency graph between repos/modules/packages > > is currently spread across many different "build systems", like the > > new proposed GHA jobs exclusive to the `kogito-serverless-operator` > > repo, the Build Chain system we have for the Drools/Kogito repos, the > > `kie-tools` CI, and the many Jenkins jobs we have on Apache's Jenkins. > > There's also the fact that we have `kogito-images` selectively > > building parts of `kogito-apps` during its own build to include them > > in the images it produces. > > > > With all that said, I'm not opposed to moving the `kn-workflow-plugin` > > package from `kie-tools` to the `cmd` module of > > `kogito-serverless-operator`. In fact, like I said in the past, I > > think it makes a lot of sense that they're part of the same > > compilation unit, as they're very closely related, and should > > therefore be in sync at all times. > > > > I just think it is important to highlight that this proposal would > > only address a LOCAL problem related exclusively to the SonataFlow > > section of the Apache KIE community, while, at the same time, not > > contributing to reducing the segmentation of the Apache KIE community > > as a whole. > > > > Regards, > > > > Tiago Bento > > > > On Fri, Apr 19, 2024 at 4:08 PM ricardo zanini fernandes > > <[email protected]> wrote: > > > > > > Alex, > > > > > > Yes, in the proposal we just barely outlined. I create the EPIC to have > > > more details and start working on them as soon as we agree. > > > > > > On Fri, Apr 19, 2024 at 4:24 PM Alex Porcelli <[email protected]> wrote: > > > > > > > Thank you for outlining the tasks post the 10.x release. It's > > > > important to note that these are already included in the amended > > > > proposal [1], specifically in steps 9 and 10, which the community has > > > > voted on. If there are concerns about the execution of these steps, > > > > I'd like to reassure you that they will proceed as planned, in > > > > compliance with Apache guidelines. > > > > > > > > Looking ahead, after the release of version 10, we already agreed that > > > > we'll need to have a thorough discussion regarding the codebase > > > > structure. This will allow us to refine our understanding of the > > > > sub-brands, their interrelationships, and their strategic positioning. > > > > I agree that this is crucial for our next steps and look forward to > > > > our collaborative efforts in shaping this. > > > > > > > > [1] - https://lists.apache.org/thread/pw327lkpmy9gxklq4t5zbwzxjo2y3c0w > > > > > > > > On Fri, Apr 19, 2024 at 2:50 PM ricardo zanini fernandes > > > > <[email protected]> wrote: > > > > > > > > > > Folks, > > > > > > > > > > I've outlined the tasks we need once we release 10.x from kie-tools: > > > > > https://github.com/apache/incubator-kie-issues/issues/1102 > > > > > > > > > > Once we release, we can detail this planning and start working on it > > to > > > > > have a streamlined process for the next release. > > > > > > > > > > Please let me know if it makes sense to you. We can break down and > > detail > > > > > the tasks once we agree on this initial plan. > > > > > > > > > > Cheers! > > > > > -- > > > > > Ricardo Zanini Fernandes > > > > > Vida longa e próspera. > > > > > > > > --------------------------------------------------------------------- > > > > To unsubscribe, e-mail: [email protected] > > > > For additional commands, e-mail: [email protected] > > > > > > > > > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: [email protected] > > For additional commands, e-mail: [email protected] > > > > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
