chore(docs): katib (optimizer) client support for kubeflow-mcp-server (kep#0001)#48
Open
Krishna-kg732 wants to merge 1 commit into
Open
chore(docs): katib (optimizer) client support for kubeflow-mcp-server (kep#0001)#48Krishna-kg732 wants to merge 1 commit into
Krishna-kg732 wants to merge 1 commit into
Conversation
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
Signed-off-by: Krishna Gupta <Krishnagupta.kg2k6@gmail.com>
Krishna-kg732
force-pushed
the
docs/KEP_katib_support
branch
from
6eef165 to
06df563
Compare
Krishna-kg732
changed the title
There was a problem hiding this comment.
Pull request overview
Adds a new KEP document proposing Katib (Optimizer) client support in kubeflow-mcp-server, describing the planned MCP tool surface, module structure, personas, and testing approach for integrating hyperparameter optimization workflows alongside the existing Trainer client.
Changes:
- Introduces KEP#0001 describing the Optimizer client scope (17 tools across Planning/Optimization/Discovery/Monitoring/Lifecycle).
- Documents proposed module layout, tool phase grouping, persona access, and cross-client workflow (
train -> tune -> retrain). - Outlines compatibility assumptions, risks/mitigations, and a unit/integration testing plan.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| @@ -0,0 +1,429 @@ | |||
| --- | |||
Comment on lines
+243
to
+244
| - When both clients active, trainer.pre_flight() covers cluster/GPU; | ||
| katib_pre_flight() covers Katib-specific readiness |
| - ALWAYS preview first (confirmed=False) | ||
| - maxTrialCount is required, no unbounded default | ||
| - Use early_stopping with medianstop for long-running trials | ||
| - Trial templates can reference TrainJobs — use trainer.list_runtimes() first |
Comment on lines
+312
to
+315
| trainer.pre_flight() # Validate cluster, GPU availability | ||
| katib_pre_flight() # Validate Katib readiness | ||
| trainer.list_runtimes() # Find training runtimes | ||
| create_hpo_experiment() # Create experiment with TrainJob trial template |
| create_hpo_experiment() # Create experiment with TrainJob trial template | ||
| wait_for_experiment() # Wait for completion | ||
| get_best_trial() # Get optimal hyperparameters | ||
| trainer.fine_tune() # Retrain with best config |
|
|
||
| | Tool | Next Hint | | ||
| |------|-----------| | ||
| | `trainer.wait_for_training` | "Use `create_hpo_experiment()` to optimize hyperparameters" | |
| | Persona | Optimizer Tools | | ||
| |---------|----------------| | ||
| | `readonly` | All read-only tools (13 tools: planning + discovery + monitoring) | | ||
| | `data-scientist` | readonly + `create_hpo_experiment`, `wait_for_experiment`, `delete_experiment` (MCP-owned only) | |
Comment on lines
+75
to
+76
| 1. Implement 17 MCP tools across 5 categories for Katib experiment, trial, | ||
| and suggestion lifecycle (see MCP Tools tables) |
Comment on lines
+282
to
+286
| "optimizer": { | ||
| "status": "implemented", | ||
| "sdk_client": "kubeflow.katib.KatibClient", | ||
| "sdk_version_min": "0.19.0", | ||
| "covered_methods": [ |
Krishna-kg732
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This KEP proposes implementing the Optimizer client module for
kubeflow-mcp-server, exposing Katib's Experiment, Trial, and Suggestionlifecycle as 17 MCP tools across 5 categories (Planning, Optimization,
Discovery, Monitoring, Lifecycle).
This implements the "Optimizer (Planned: Phase 2)" node already identified
in the architecture and stub module (
kubeflow_mcp.optimizer), making Katibthe natural second client after TrainerClient — completing the inner loop of
train -> evaluate -> tune -> retrainwithout leaving the MCP interface.Motivation
AI IDEs and orchestrator agents currently have no structured way to:
algorithm status
The existing stub declares 8 planned tools with
status: "stub"and noimplementations.
Goals
Monitoring, and Lifecycle categories
(
create_hpo_experiment/create_experiment_from_spec), following thetrainer's
fine_tune/run_custom_trainingpatternkubeflow.katib.KatibClientas the primary interface, withCustomObjectsApifallback where the SDK lacks coveragelimiting, circuit breaker, namespace enforcement)
install
Non-Goals
get_trial_metrics(), requires gRPC)tune()API (requires Python callables, not MCP-serializable)edit_experiment_budget()— deferredDetails
Full design : module structure, tool tables, persona coverage, SDK
compatibility, cross-client integration with TrainerClient, risks and
mitigations, and the testing plan , is in the KEP doc itself.
Status
open for review and discussion before implementation begins.
cc: @jaiakash , @abhijeet-dhumal , @andreyvelich