kubernetes-platform
Attach Kubernetes-specific kfp-kubernetes task configuration to Kubeflow Pipelines PipelineTask objects and verify the compiled platform YAML without requiring a cluster.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- kubernetes-platform
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Attach Kubernetes-specific kfp-kubernetes task configuration to Kubeflow Pipelines PipelineTask objects and verify the compiled platform YAML without requiring a cluster.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md7.4 KB
- 参考文档references/troubleshooting.md5.0 KB
- 参考文档references/workflows.md5.1 KB
- 脚本scripts/compile_kubernetes_config_pipeline.py5.9 KB
03 SKILL.md 正文
Kubernetes Platform
Use this sub-skill when the user already has a Kubeflow Pipelines DSL task and needs kfp-kubernetes platform configuration: Secrets, ConfigMaps, PVCs, ephemeral volumes, node placement, tolerations, pod metadata, image pull settings, field-path env vars, security context, timeout, or init containers.
Route first
Stay in this sub-skill for:
- installing/importing the
kfp-kubernetesaddon and checking its publickfp.kuberneteshelper surface; - applying Kubernetes helpers to
dsl.PipelineTaskobjects returned by component calls; - compile-only verification that the generated YAML contains Kubernetes platform markers.
Route away when the request is really about:
- general DSL components, pipeline structure, type hints, task dependencies, or artifacts ->
pipeline-authoring; - compiler flags, CLI wrapper details, package paths, or manifest-format options ->
compiler-and-cli; - submitting/running against a live KFP endpoint or proving runtime Secret/PVC/ConfigMap existence ->
client-and-registryorrepo-developmentdepending on context; - repository maintainer internals, backend manifests, deployment configuration, or generated Kubernetes resources ->
repo-development.
Operating checklist
- Verify imports and package alignment before advising on helper behavior:
Install withimport importlib.metadata as metadata import kfp from kfp import kubernetes assert metadata.version("kfp-kubernetes") == kfp.__version__ print(kubernetes.__all__)pip install kfp[kubernetes]orpip install kfp-kuberneteswhen the addon is absent. - Attach helpers only to a
PipelineTask, for exampletask = my_component(). Do not pass the component function, aContainerSpec, or the pipeline function itself. - Prefer public helpers exported from
kfp.kubernetes.__all__; seereferences/api-reference.mdfor signatures, return behavior, and YAML marker names. - Compile the pipeline and inspect the platform document/section. Kubernetes helper calls serialize under
platforms.kubernetes.deploymentSpec.executors.<executor>, not in ordinary component code. - State the verification boundary: compile success proves that the DSL emitted platform configuration, but it does not create cluster resources or prove that a Kubernetes Secret, ConfigMap, PVC, StorageClass, node label, toleration target, or admission policy exists.
Fast compile smoke
Run the bundled smoke helper in an environment with matching kfp and kfp-kubernetes installed. Resolve the script path relative to this sub-skill directory:
python scripts/compile_kubernetes_config_pipeline.py \
--output /tmp/kubernetes-platform-smoke.yaml
The script compiles a tiny pipeline, asserts that helper calls return the same PipelineTask, and checks for representative secretAsEnv, nodeSelector, and tolerations markers in the YAML. It never contacts a Kubernetes cluster.
Reference map
references/api-reference.md: installation, public helpers, signatures, return behavior, and platform YAML markers.references/workflows.md: practical compile-only workflows for Secrets/ConfigMaps, PVCs/volumes, scheduling, pod configuration, and validation boundaries.references/troubleshooting.md: import/version failures, wrong-object usage, compile-vs-runtime confusion, and invalid Kubernetes inputs.
Native evidence anchors
This skill is grounded in the kfp-kubernetes README/setup metadata, the exported kfp.kubernetes source modules, Sphinx API stubs, installed package facts, unit tests for Secret/ConfigMap/volume helpers, snapshot compile/read-write coverage, the core PVC sample, and SDK compilation tests that compare generated platform specs.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).