Improved Kubernetes orchestrator pod caching #3719

schustmi · 2025-06-02T13:33:25Z

Describe changes

This PR

enables orchestrator pod caching for scheduled runs.
delays the caching decision to run right before the actual step pod starts, not at orchestrator pod startup. This means the caching decision will be made with the latest state of all step runs available in the server, which might differ from the step runs available at orchestrator pod startup.
adds metadata to the Kubernetes orchestrator.
improves pod labels when using the Kubernetes orchestrator.

Changes to what we consider placeholder runs:
Previously, a run without an orchestrator_run_id in the database was considered a placeholder run. This PR changes that behaviour and instead uses the status of a run to decide whether the run is a placeholder run (runs with state initializing are placeholder runs). This change was necessary so that the orchestrator pod can create a pipeline run (with orchestrator run ID) for scheduled runs to enable the orchestrator pod caching.

Pre-requisites

Please ensure you have done the following:

I have read the CONTRIBUTING.md document.
I have added tests to cover my changes.
I have based my new branch on develop and the open PR is targeting develop. If your branch wasn't based on develop read Contribution guide on rebasing branch to develop.
IMPORTANT: I made sure that my changes are reflected properly in the following resources:
- ZenML Docs
- Dashboard: Needs to be communicated to the frontend team.
- Templates: Might need adjustments (that are not reflected in the template tests) in case of non-breaking changes and deprecations.
- Projects: Depending on the version dependencies, different projects might get affected.

Types of changes

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Breaking change (fix or feature that would cause existing functionality to change)
Other (add details above)

coderabbitai · 2025-06-02T13:33:31Z

Important

Review skipped

Auto reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

🪧 Tips

Chat

There are 3 ways to chat with CodeRabbit:

Review comments: Directly reply to a review comment made by CodeRabbit. Example:
- I pushed a fix in commit <commit_id>, please review it.
- Explain this complex logic.
- Open a follow-up GitHub issue for this discussion.
Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query. Examples:
- @coderabbitai explain this code block.
- @coderabbitai modularize this function.
PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
- @coderabbitai gather interesting stats about this repository and render them as a table. Additionally, render a pie chart showing the language distribution in the codebase.
- @coderabbitai read src/utils.ts and explain its main purpose.
- @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.
- @coderabbitai help me debug CodeRabbit configuration file.

Support

Need help? Create a ticket on our support page for assistance with any issues or questions.

Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments.

CodeRabbit Commands (Invoked using PR comments)

@coderabbitai pause to pause the reviews on a PR.
@coderabbitai resume to resume the paused reviews.
@coderabbitai review to trigger an incremental review. This is useful when automatic reviews are disabled for the repository.
@coderabbitai full review to do a full review from scratch and review all the files again.
@coderabbitai summary to regenerate the summary of the PR.
@coderabbitai generate docstrings to generate docstrings for this PR.
@coderabbitai generate sequence diagram to generate a sequence diagram of the changes in this PR.
@coderabbitai resolve resolve all the CodeRabbit review comments.
@coderabbitai configuration to show the current CodeRabbit configuration for the repository.
@coderabbitai help to get help.

Other keywords and placeholders

Add @coderabbitai ignore anywhere in the PR description to prevent this PR from being reviewed.
Add @coderabbitai summary to generate the high-level summary at a specific location in the PR description.
Add @coderabbitai anywhere in the PR title to generate the title automatically.

Documentation and Community

Visit our Documentation for detailed information on how to use CodeRabbit.
Join our Discord Community to get help, request features, and share feedback.
Follow us on X/Twitter for updates and announcements.

github-actions · 2025-06-03T14:51:02Z

Documentation Link Check Results

✅ Absolute links check passed
✅ Relative links check passed
_{Last checked: 2025-06-05 07:30:49 UTC}

…-pod-caching-v2

stefannica

Mostly questions, looks good overall.

stefannica · 2025-06-25T09:46:08Z

src/zenml/zen_stores/sql_zen_store.py

+            # In very rare cases, there can be multiple placeholder runs for
+            # the same deployment. By ordering by the orchestrator_run_id, we
+            # make sure that we use the placeholder run with the matching
+            # orchestrator_run_id if it exists, before falling back to the
+            # placeholder run without any orchestrator_run_id provided.
+            # Note: This works because both SQLite and MySQL consider NULLs
+            # to be lower than any other value. If we add support for other
+            # databases (e.g. PostgreSQL, which considers NULLs to be greater
+            # than any other value), we need to potentially adjust this.
+            .order_by(desc(PipelineRunSchema.orchestrator_run_id))


good to know !

src/zenml/zen_stores/schemas/pipeline_run_schemas.py

src/zenml/zen_stores/sql_zen_store.py

stefannica · 2025-06-25T10:04:36Z

src/zenml/orchestrators/step_run_utils.py

+                # As a last resort, we try to reuse the docstring/source code
+                # from the cached step run. This is part of the cache key
+                # computation, so it must be identical to the one we would have
+                # computed ourselves.
+                if request.source_code is None:
+                    request.source_code = cached_step_run.source_code
+                if request.docstring is None:
+                    request.docstring = cached_step_run.docstring
+


Can you explain this part a bit ? This wasn't necessary before and the _get_docstring_and_source_code already attempts to fetch these from various sources (the step instance and run template). Why is it absolutely necessary now that these be populated in the request, if it wasn't necessary before ?

This was never necessary (both of those fields are optional). It just now occurred to me that we can simply reuse the code/docstring of the cached step if we can't figure it out in any other way, so I added it. Otherwise, all steps that got cached from the orchestrator pod would not have any code/docstring associated with them, as

the code to import the step instance most likely doesn't exist in the orchestrator pod

there is no run template involved

src/zenml/integrations/kubernetes/step_operators/kubernetes_step_operator.py

stefannica · 2025-06-25T10:22:25Z

src/zenml/integrations/kubernetes/orchestrators/kubernetes_orchestrator_entrypoint.py

+                step_runs = Client().list_run_steps(
+                    size=1,
+                    pipeline_run_id=pipeline_run.id,
+                    name=step_name,
+                )


this will scale badly, I know it. For 1000 steps, it will make 1000 calls to the API. I think we might need a new endpoint that fetches the status of all steps in one go. Or maybe move this whole thing behind the REST API.

On second thought, this whole thing feels like something that should happen server-side when the pipeline run status is updated... this whole thing needs to be reconsidered.

My change here only improves upon the previous implementation, which fetched all steps of the pipeline run. My solution instead only fetches the steps for which the pod is actually in a failed state. However, I definitely agree that we could batch this into some requests, instead of fetching each step separately. I'll change it.

I'm now fetching only relevant steps, and doing so in batched chunks.

Co-authored-by: Stefan Nica <[email protected]>

schustmi added 2 commits June 2, 2025 15:26

Improved Kubernetes orchestrator pod caching

0877a4b

Enable for scheduled deployments

30e212f

github-actions bot added internal To filter out internal PRs and issues enhancement New feature or request labels Jun 2, 2025

schustmi added 6 commits June 2, 2025 15:42

Improve placeholder run detection

175e38e

Store orchestrator run ID, improve labeling

598b436

Index DB migration

b1d69b2

Docstring

a771454

Remove unnecessary index

b780b48

Merge branch 'develop' into feature/kubernetes-pod-caching-v2

53d14d4

schustmi requested a review from stefannica June 3, 2025 15:03

schustmi added 6 commits June 4, 2025 10:15

Order placeholder runs

54be72d

Merge branch 'develop' into feature/kubernetes-pod-caching-v2

7b3eba9

Fix tests

e81154d

Merge remote-tracking branch 'origin/develop' into feature/kubernetes…

685d26e

…-pod-caching-v2

Reuse docstring/source code from cache candidate

6f3f45c

Use more portable sorting

6c8791f

schustmi changed the base branch from main to develop June 5, 2025 08:24

schustmi and others added 4 commits June 16, 2025 17:58

Merge branch 'develop' into feature/kubernetes-pod-caching-v2

1609a9e

Merge branch 'develop' into feature/kubernetes-pod-caching-v2

237a95a

Formatting after merge

c167a69

Linting

233e33f

stefannica requested changes Jun 25, 2025

View reviewed changes

schustmi and others added 4 commits June 25, 2025 13:13

Apply suggestions from code review

7dcf476

Co-authored-by: Stefan Nica <[email protected]>

Add run name label

ec5e7a3

Fetch step runs for failed nodes in batches

50a1098

Merge branch 'develop' into feature/kubernetes-pod-caching-v2

a5c32c7

schustmi requested a review from stefannica June 26, 2025 07:42

schustmi added 3 commits June 26, 2025 15:35

Merge branch 'develop' into feature/kubernetes-pod-caching-v2

99073c5

Reduce wait time, compute cache after acquiring lock

e4b8190

Merge branch 'develop' into feature/kubernetes-pod-caching-v2

2b4b0ec

stefannica approved these changes Jun 27, 2025

View reviewed changes

schustmi merged commit 331eaef into develop Jun 27, 2025
48 of 49 checks passed

schustmi deleted the feature/kubernetes-pod-caching-v2 branch June 27, 2025 13:08

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Improved Kubernetes orchestrator pod caching #3719

Improved Kubernetes orchestrator pod caching #3719

Uh oh!

schustmi commented Jun 2, 2025 •

edited

Loading

Uh oh!

coderabbitai bot commented Jun 2, 2025

Review skipped

Chat

Support

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

Documentation and Community

Uh oh!

github-actions bot commented Jun 3, 2025 •

edited

Loading

Uh oh!

stefannica left a comment

Uh oh!

stefannica Jun 25, 2025

Uh oh!

Uh oh!

Uh oh!

stefannica Jun 25, 2025

Uh oh!

schustmi Jun 25, 2025

Uh oh!

Uh oh!

stefannica Jun 25, 2025

Uh oh!

stefannica Jun 25, 2025

Uh oh!

schustmi Jun 25, 2025

Uh oh!

schustmi Jun 25, 2025

Uh oh!

Uh oh!

Uh oh!

Improved Kubernetes orchestrator pod caching #3719

Improved Kubernetes orchestrator pod caching #3719

Uh oh!

Conversation

schustmi commented Jun 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Describe changes

Pre-requisites

Types of changes

Uh oh!

coderabbitai bot commented Jun 2, 2025

Review skipped

Chat

Support

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

Documentation and Community

Uh oh!

github-actions bot commented Jun 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Documentation Link Check Results

Uh oh!

stefannica left a comment

Choose a reason for hiding this comment

Uh oh!

stefannica Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

stefannica Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

schustmi Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

stefannica Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

stefannica Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

schustmi Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

schustmi Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

schustmi commented Jun 2, 2025 •

edited

Loading

github-actions bot commented Jun 3, 2025 •

edited

Loading