deployment-helm-docker
Reason about Apache Airflow deployments with the official Helm chart and Docker images, including values, image customization, Dag delivery, logging, secrets, scaling, and safe rendering checks.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- deployment-helm-docker
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Reason about Apache Airflow deployments with the official Helm chart and Docker images, including values, image customization, Dag delivery, logging, secrets, scaling, and safe rendering checks.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/deployment-workflows.md9.2 KB
- 参考文档references/docker-image-reference.md8.6 KB
- 参考文档references/helm-reference.md11.9 KB
- 参考文档references/troubleshooting.md10.8 KB
- 脚本scripts/render_helm_values_summary.py13.0 KB
03 SKILL.md 正文
<!-- SPDX-License-Identifier: Apache-2.0 -->
Airflow deployment with Helm and Docker
Use this sub-skill when the task is about deploying Airflow with the official Helm chart, extending or customizing official Docker images, reviewing chart values, planning Kubernetes operational choices, or debugging deployment failures.
Start here
- Read
references/deployment-workflows.mdto choose the deployment path: local container image iteration, Helm on Kubernetes, Dag distribution, logs, secrets/config, or scaling. - Read
references/helm-reference.mdwhen reviewing or drafting chart values, rendering manifests safely, or checking component ownership. - Read
references/docker-image-reference.mdbefore advising on packages, providers, image tags, entrypoints, or custom image builds. - Read
references/troubleshooting.mdwhen pods fail to start, rendered manifests are wrong, Dags do not appear, imports fail in the scheduler, logs disappear, migrations hang, or autoscaling behaves oddly. - Use
scripts/render_helm_values_summary.pyfor a no-cluster, no-Helm summary of a values file before suggesting deeper chart rendering.
Scope boundaries
This sub-skill covers:
- Official Airflow Docker image extension/customization, dependency installation timing, build arguments, entrypoint behavior, and image selection.
- Helm chart values for API server, scheduler, Dag processor, workers, triggerer, Redis, PgBouncer, metadata database, secrets, logging, Dag delivery, KEDA/HPA, and rendering checks.
- Safe validation commands such as local values summary,
helm template,helm lint,helm diff, and chart test selection when the user is already changing chart code. - Kubernetes deployment reasoning that does not require cluster access by default.
Do not use this sub-skill for:
- Day-to-day Airflow CLI, REST API,
airflowctl, or metadata operations; route those tooperations-cli-api. - Task SDK authoring, Dag syntax, assets, or provider APIs except when deployment packaging affects imports.
- Broad provider implementation deep dives or generated client work.
- General contribution workflow, release workflow, or non-chart CI maintenance except chart-specific rendered manifest tests and generated-doc cautions.
High-signal reminders
- Prefer building an immutable image for production dependencies. Runtime
pip installduring pod startup is fragile, slow, and repeated on every container start. - For Airflow 3 chart values, think in components:
apiServer,scheduler,dagProcessor,triggerer,workers, and optional services such as Redis, PgBouncer, StatsD, and OpenTelemetry. - Treat secrets as Secret references whenever possible. Avoid plain credentials in values files, and never echo secrets into handoffs or generated examples.
- Normal skill usage should not require a Kubernetes cluster. Start with values inspection and local rendering; use cluster commands only when the user explicitly asks for live troubleshooting.
- Use title-case Dag in prose. Preserve literal tokens such as
dags.gitSync,airflow dags list,DAG, anddag_idexactly when they are code/config/CLI tokens.
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).