Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update Kubeflow Installation with Standalone Mode #3724

Merged
merged 26 commits into from
May 29, 2024

Conversation

andreyvelich
Copy link
Member

As we discussed multiple times before, we want to explain our users that Kubeflow components can be deployed:

  • in Standalone mode.
  • in Kubeflow Platform for end-to-end AI/ML experience.

In this PR I made changes to the installation section:
I removed What is Kubeflow? section from this page since we already have this section here: https://www.kubeflow.org/docs/started/introduction/#what-is-kubeflow, we can update it if you think it is out-of-date.

I made these sections:

  • Install Kubeflow Components Standalone
  • Install Kubeflow Platform from Raw Manifests
  • Install Kubeflow Platform from Packaged Distributions

From my point of view this is logically correct, for users to learn about Kubeflow ecosystem. Firstly, users will explore the value of individual Kubeflow components. Secondly, if users need end-to-end AI/ML experience with all Kubeflow components, they will explore the Raw Manifests and Package Distributions installation.

What do you think about it ?

I am happy to work on the language to make it less confusing for our users if you have other ideas on how we can improve it.

/hold for review

/assign @kubeflow/wg-notebooks-leads @kubeflow/wg-pipeline-leads @kubeflow/wg-training-leads @kubeflow/wg-data-leads @kubeflow/release-team @kubeflow/kubeflow-steering-committee @kubeflow/wg-deployment-leads

Copy link
Member

@tarilabs tarilabs left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I like this a lot, I believe it is important to make evident that you can opt-in to a subset of specific components and not "necessarily be force to the whole" (which was never the case).

Hence, I believe it's good the section "Install Kubeflow Components Standalone" comes first.

Maybe we can add some/additional brief paragraph when to chose either, for example:

  • for Components-Standalone: helpful when you want to try out locally or in your Kubernetes installation a specific component
  • for Raw-Manifests, in addition to being the basis for distribution (which is what is currently mentioned): helpful when you want to try out the end-to-end Kubeflow platform capabilities, or see how a specific component integrates in the Kubeflow ecosystem
  • for Packages Distributions: nothing to add here

for considerations, hope this helps

@@ -5,30 +5,65 @@ weight = 20

+++

## What is Kubeflow?
Kubeflow is the ecosystem of various applications created to address each stage in the AI/ML lifecycle,
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
Kubeflow is the ecosystem of various applications created to address each stage in the AI/ML lifecycle,
Kubeflow is an ecosystem of various applications created to address each stage in the AI/ML lifecycle,

@thesuperzapper
Copy link
Member

I think this update is much more confusing to new users and significantly confuses the message of "What is Kubeflow?".

To be clear, I am not against telling users that some components of Kubeflow can be installed on their own, but we need to be careful not to confuse people or make representations of support that are not true.


Alternate Proposal

The current page is actually pretty good at explaining the options users have to install Kubeflow.

I propose we simply add a 3rd option under "How to install Kubeflow?" so we have the following (in this order):

  1. Packaged Distributions
  2. Raw Manifests (advanced users)
  3. Standalone Components (advanced users)

We also must get firm commitments from the maintainers of components listed under "Standalone Components". Users will expect deployment support, and don't want to impose that on maintainers without their agreement.

Copy link
Contributor

@hbelmiro hbelmiro left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I agree with removing "What is Kubeflow?" from here.
Regarding the order of the types of installations, I don't see much difference in which one goes first. As long as we mention three types of installations in the beginning, the order doesn't seem to matter to me.

- Install [Kubeflow Pipelines](/docs/components/pipelines/v2/installation/quickstart/)
- Install [Kubeflow Training Operator](/docs/components/training/installation/#installing-training-operator)
- Install [Kubeflow MPI Operator](/docs/components/training/user-guides/mpi/#installation)
- Install [Kubeflow Katib] (TODO: Add link after #3723)
Copy link
Contributor

@hbelmiro hbelmiro Apr 29, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure about adding "TODO" here. Maybe "Not yet available", or something like that would be better. Also, adding to kubeflow/katib#2314 a task to update the link here.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I want to merge this PR after Katib and Model registry docs will be updated: #3723, #3698.
So I can add the appropriate links here.

- Install [Kubeflow MPI Operator](/docs/components/training/user-guides/mpi/#installation)
- Install [Kubeflow Katib] (TODO: Add link after #3723)
- Install [KServe](https://kserve.github.io/website/0.10/admin/serverless/serverless/)
- Install Kubeflow Model Registry (TODO: Add link after #3698)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same as above, but using the respective Model Registry issue.

@andreyvelich
Copy link
Member Author

helpful when you want to try out locally or in your Kubernetes installation a specific component

@tarilabs I added your suggestion to the manifests installation, thanks! For the standalone installation, do you think these messages are not sufficient to explain value of Standalone Components ?

Kubeflow components can be deployed as standalone applications.
You can integrate those components to your existing AI/ML platform.
For example, for Model Training you can install Training Operator or for Model Serving you can install KServe.

@andreyvelich
Copy link
Member Author

andreyvelich commented Apr 29, 2024

I think this update is much more confusing to new users and significantly confuses the message of "What is Kubeflow?".

@thesuperzapper Please can you explain why this update confuses users about Kubeflow Ecosystem value ?
From my perspective we make it very clear that Kubeflow project can be used in various ways.
The main motivation is to give new Kubeflow components (Model Registry, Spark Operator) visibility and grow the community around it (outside of Kubeflow Platform scope).

To be clear, I am not against telling users that some components of Kubeflow can be installed on their own, but we need to be careful not to confuse people or make representations of support that are not true.

Which message shows incorrect representation of support ? If users deploy Kubeflow components standalone and get some potential bugs/integrations issues, they should raise it to the components owners to get support.

Raw Manifests (advanced users)

What is the end-user value to tell that Raw Manifest and Standalone Installation is for advanced users ?
Some Kubeflow components are very light-weight and easy to install.

Users will expect deployment support, and don't want to impose that on maintainers without their agreement.

I am happy to remove some of the Kubeflow components from this page if WG leads don't want their components to be presented here.

@andreyvelich
Copy link
Member Author

cc @jeremyeder @bigsur0 Would love to get your feedback for this change as well.

@thesuperzapper
Copy link
Member

thesuperzapper commented Apr 29, 2024

@andreyvelich

I have made an alternative PR which tries to be less confusing:

It is a much smaller change to the structure, and uses tables to format the information about if each component can be deployed standalone.

If you agree that its better, lets move the discussion to that PR and close this.

@juliusvonkohout
Copy link
Member

@andreyvelich I have made an alternative PR which tries to be less confusing:

* [Add "Standalone Components" section to "Installing Kubeflow" #3726](https://github.com/kubeflow/website/pull/3726)

It is a much smaller change to the structure, and uses tables to format the information about if each component can be deployed standalone.

If you agree that its better, lets move the discussion to that PR and close this.

Doesnt it change this order here significantly?

- [**Install Kubeflow Components Standalone**](#install-kubeflow-components-standalone)
- [**Install Kubeflow Platform from Raw Manifests**](#install-kubeflow-platform-from-raw-manifests)
- [**Install Kubeflow Platform from Packaged Distributions**](#install-kubeflow-platform-from-packages-distributions)

seems to become

- [**Install Kubeflow Platform from Packaged Distributions**](#install-kubeflow-platform-from-packages-distributions)
- [**Install Kubeflow Components Standalone**](#install-kubeflow-components-standalone)
- [**Install Kubeflow Platform from Raw Manifests**](#install-kubeflow-platform-from-raw-manifests)

@tarilabs
Copy link
Member

@tarilabs I added your suggestion to the manifests installation, thanks! For the standalone installation, do you think these messages are not sufficient to explain value of Standalone Components ?

glad that helped, about the standalone very-imho is worthy to call out testing a component "locally" or in isolation, in addition to the existing AI/ML/K8s one might have.
for example, as a relatively newcomer to Kubeflow, I thought I needed it all if I wanted to try one single part :)
my2c, of course I can understand if this comment is not incorporated! 🚀 👍

@andreyvelich
Copy link
Member Author

andreyvelich commented Apr 30, 2024

@andreyvelich I have made an alternative PR which tries to be less confusing:

It is a much smaller change to the structure, and uses tables to format the information about if each component can be deployed standalone.

If you agree that its better, lets move the discussion to that PR and close this.

@thesuperzapper As we discussed, I am happy to incorporate your feedback in this PR based on our discussion during today's community call. A few thoughts from my side:


If you don't like the ordering of installation instructions we can make these 2 sections:

  • Install Kubeflow Components Standalone
  • Install Kubeflow Platform

In that case, we can move Distributions + Raw Manifests to the Install Kubeflow Platform section.


From my point of view, it doesn't give a lot of value for users to add Supports Standalone column to the table. We can always update the table with other Kubeflow Component, if it supports it. For example, we can add Kubeflow Notebooks in the future.


I am not sure if we should add the Note about limited or unavailable feature section to the Standalone Components. We can add the section to the Kubeflow Introduction:

  • What is Kubeflow ?
  • What is Kubeflow Standalone Components ?
  • What is Kubeflow Platform ?

And explain the differences there.

@andreyvelich
Copy link
Member Author

Thanks everyone who participated in our today's discussion to show the real power of Kubeflow ecosystem.
I strongly believe that this change will help us to grow the community especially with the new project adoption: Spark Operator and Model Registry.

I want to summarize our discussion today about this PR:

  1. We agree that it makes sense to explain that Kubeflow Components can be deployed standalone.
  2. We want to collect feedback from distribution owners for this change.
  3. We should re-use this ML Lifecycle diagram that @franciscojavierarceo and I created for Training Operator and re-use it in the ML Workflow guide
  4. We are discussing to add the following statement to the installation page:
    What is Kubeflow ?
    Example text:
Kubeflow is an ecosystem of various applications created to address each stage in the AI/ML lifecycle,
from exploration to training and serving. You can deploy Kubeflow components as standalone applications
or deploy Kubeflow as an end-to-end AI/ML platform. Anywhere you are running Kubernetes,
you should be able to run Kubeflow.

What is Kubeflow Standalone Components ?

TODO: Add text

What is Kubeflow Platform ?

TODO: Add text

If we agree with that, we can add these subsections to the installation page:
Installing Kubeflow Standalone Components
Installing Kubeflow Platform

In the future PRs we can split them between 2 separate pages if we want.

  1. If we agree, I can keep (Advanced User) and existing warnings for manifests installation in this PR. We can make followup discussion if we should change it.

  2. I am happy to take the table from PR: Add "Standalone Components" section to "Installing Kubeflow" #3726, with these columns: Kubeflow Component, Source Code.
    As I said before: Update Kubeflow Installation with Standalone Mode #3724 (comment) I am not sure if we should have "Supported Standalone" section.

Please let me know if I missed something.

@thesuperzapper Please can you share recordings from our today's call.

@andreyvelich
Copy link
Member Author

Distribution owners please give your feedback for this PR.
/cc @ca-scribner @gkcalat @zijianjoy @Linchin @Tomcli @yhwang @johnugeorge @nagar-ajay @rimolive @thesuperzapper @liuqi @xujinheng @alexeadem

Copy link

@andreyvelich: GitHub didn't allow me to request PR reviews from the following users: Linchin, alexeadem.

Note that only kubeflow members and repo collaborators can review this PR, and authors cannot review their own PRs.

In response to this:

Distribution owners please give your feedback for this PR.
/cc @ca-scribner @gkcalat @zijianjoy @Linchin @Tomcli @yhwang @johnugeorge @nagar-ajay @rimolive @thesuperzapper @liuqi @xujinheng @alexeadem

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
@andreyvelich
Copy link
Member Author

@kubeflow/wg-manifests-leads @jbottum @terrytangyuan I renamed Raw Manifests to Kubeflow Manifests.

Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
@terrytangyuan
Copy link
Member

/lgtm

Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
@google-oss-prow google-oss-prow bot removed the lgtm label May 25, 2024
Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
Copy link
Member

@thesuperzapper thesuperzapper left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@andreyvelich I am happy with the page now, and will LGTM once these small typos are fixed.

Comment on lines 155 to 156
You can use one of the following methods to install [Kubeflow Platform](/docs/started/introduction/#what-is-kubeflow-platform)
to get full suite of standalone Kubeflow components bundled together with additional tools.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The sentence had a few typos, and I dont think we need to say "standalone" as "kubeflow components" is more accurate in this case:

Suggested change
You can use one of the following methods to install [Kubeflow Platform](/docs/started/introduction/#what-is-kubeflow-platform)
to get full suite of standalone Kubeflow components bundled together with additional tools.
You can use one of the following methods to install the [Kubeflow Platform](/docs/started/introduction/#what-is-kubeflow-platform)
and get the full suite of Kubeflow components bundled together with additional tools.


Advanced users may choose to install the manifests for a specific Kubeflow version by following the
Kubeflow Manifests contain all Kubeflow Components, Kubeflow Central Dashboard, and other Kubeflow
applications which makes **Kubeflow Platform**. This installation is helpful when you want to try
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Its not valid to use "makes" in this context:

Suggested change
applications which makes **Kubeflow Platform**. This installation is helpful when you want to try
applications that comprise the **Kubeflow Platform**. This installation is helpful when you want to try


When using the raw manifests, the Kubeflow community is not able to provide support for environment-specific issues or custom configurations.
When using the Kubeflow manifests, the community is not able to provide support for environment-specific issues or custom configurations.
If you need support, please consider using a [packaged distribution](#packaged-distributions-of-kubeflow).
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We renamed this heading, so the link needs to be updated:

Suggested change
If you need support, please consider using a [packaged distribution](#packaged-distributions-of-kubeflow).
If you need support, please consider using a [packaged distribution](#packaged-distributions).

Copy link
Member

@thesuperzapper thesuperzapper left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@andreyvelich I am happy with the page now, and will LGTM once these small typos are fixed.

Signed-off-by: Andrey Velichkevich <andrey.velichkevich@gmail.com>
@andreyvelich
Copy link
Member Author

I believe, I addressed all of your comments.

@terrytangyuan
Copy link
Member

/lgtm

@google-oss-prow google-oss-prow bot added the lgtm label May 28, 2024
@james-jwu
Copy link

/lgtm

2 similar comments
@thesuperzapper
Copy link
Member

/lgtm

@johnugeorge
Copy link
Member

/lgtm

@juliusvonkohout
Copy link
Member

@andreyvelich If we can address #3724 (comment)

In a follow up PR

I can also
/lgtm

@thesuperzapper
Copy link
Member

thesuperzapper commented May 29, 2024

@andreyvelich If we can address #3724 (comment)

@juliusvonkohout that warning is important to make sure that people understand that using the manifests requires some advanced knowledge. It's very important that we keep it to prevent novice users trying to deploy the manifests (without knowing what they're getting into).

If you think the wording could be improved, let's discuss that in a follow up PR, because that warning already exists on the current page, so it's not really related to this PR and I really want to get this merged.

Copy link
Member Author

@andreyvelich andreyvelich left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks everyone for your review, I appreciate your time and effort for this!
/hold cancel

@google-oss-prow google-oss-prow bot merged commit 44fbaf9 into kubeflow:master May 29, 2024
6 checks passed
@andreyvelich andreyvelich deleted the update-kubeflow-install branch May 29, 2024 16:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.