SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsdevops-guidance

devops-guidance

Deployment & operations

View repository

Updated 10-22 19:52

About this skill

Snowflake DevOps Guidance

Overview

This skill gives opinionated guidance on how to ship changes to Snowflake — schema migrations, stored procedures, Streamlit apps, Notebooks, Snowpark code — using CI/CD and Database Change Management (DCM) tools. It helps you pick a tool, structure a pipeline, and avoid common anti-patterns.

Load references/raw-guidance.md for the full reference (terminology, tool comparisons, native features, curated links).

When to Use

  • Picking a DCM tool (schemachange, Flyway, Liquibase, Terraform, DCM Projects)
  • Designing a CI/CD pipeline for Snowflake
  • Deciding between imperative and declarative deployment
  • Deploying "complex" objects (Streamlit, Notebooks, Snowpark procs, dbt projects)
  • Choosing where the pipeline runs (external CI vs inside Snowflake)
  • Handling rollback / failed deployments
  • Evaluating Snowflake-native features: CREATE OR ALTER, EXECUTE IMMEDIATE FROM, DCM Projects, Git Repositories

Workflow

Step 1: Classify the question

Category Example
Terminology "What is DCM?" "DevOps vs CI/CD?"
Tool selection "schemachange vs Flyway?" "Should I use Terraform?"
Deployment approach "Imperative or declarative?"
Object types "How do I deploy a Streamlit app?"
Where to run "Can the pipeline run from Snowflake itself?"
Native features "Should I use DCM Projects or EXECUTE IMMEDIATE FROM?"
Rollback "How do I undo a failed deploy?"

Step 2: Apply the core positions

  1. Imperative beats declarative for pipelines. Versioned migration scripts are predictable and reviewable.
  2. Declarative tools still have value for source-of-truth definitions, scaffolding, and drift detection — just not as the pipeline driver.
  3. Snowflake CLI (snow) is the right tool for complex objects (Streamlit, Notebooks, Snowpark stored procs, dbt projects).
  4. schemachange is the recommended starting point for imperative DCM.
  5. Run the pipeline from an external CI/CD system (GitHub Actions, GitLab CI, Jenkins, Azure DevOps).
  6. Roll forward, not back. Treat undo features as a last resort; write a forward fix.
  7. Combining tools is normal — DCM tool for SQL migrations + Snowflake CLI for complex objects.
  8. DCM Projects is Snowflake's native declarative option. Weigh it against the imperative approach based on object coverage and your team's needs; use the bundled dcm skill to build one and the ci-cd skill to automate it.
  9. EXECUTE IMMEDIATE FROM is not a deployment pipeline.
  10. Git Repository in Snowflake is a dev-loop convenience, not a CI/CD substitute.

Step 3: Recommend a tool

Scenario Pick
New to DCM schemachange
Already using Flyway/Liquibase Stay
Only simple objects Any imperative DCM tool
Only complex objects Snowflake CLI
Mixed simple + complex schemachange + Snowflake CLI
Cloud infra adjacent to Snowflake Terraform is fine for that slice
Account-level environment setup Terraform is fine

For Terraform, be specific about tradeoffs: state file management, HCL format, incomplete provider, learning curve.

Step 4: Point to resources

Pull links from references/raw-guidance.md. For production pipelines, prefer an imperative DCM tool run from an external CI/CD system over EXECUTE IMMEDIATE FROM scripts run inside Snowflake, because the external pipeline adds build steps, Git triggers, approvals, testing, and staged environment promotion.

Step 5: Summarize and ask for follow-ups

Close with the headline recommendation and ask what to dig into next.

Common Mistakes

  • Calling DCM "IaC". They overlap but aren't the same; keep the terms separate.
  • Treating CREATE OR ALTER as a DCM tool. It's a useful SQL primitive, not a migration system.
  • Using EXECUTE IMMEDIATE FROM as a deploy pipeline. It runs a script; it doesn't track state, order, or environments.
  • Wiring Terraform directly into the deploy pipeline for app objects. State drift and partial provider coverage will bite.
  • Confusing Snowflake CLI (snow) with SnowSQL (snowsql). SnowSQL is legacy; use snow.
  • Believing a Snowflake Git Repository connection equals DevOps. It doesn't.
  • Designing for rollback first. Optimize for fast forward-fixes and observability.

Stopping Points

This skill is advisory and runs no destructive operations, so no execution stopping points are required. Two soft checkpoints:

  • ✋ If the question is ambiguous, ask before recommending a tool.
  • ✋ If the user wants an approach this guidance argues against (e.g., Terraform-driven pipeline, EXECUTE IMMEDIATE FROM as CI), surface the tradeoffs and confirm before proceeding.

Output

A direct recommendation plus the reasoning, links to the right quickstarts and docs, and a flag on any anti-patterns in the user's current setup.