Skip to content

deploy-manage-deployments-guide.md

Latest commit

 

History

History
126 lines (94 loc) · 4.04 KB

deploy-manage-deployments-guide.md

File metadata and controls

126 lines (94 loc) · 4.04 KB

Deploy: Manage Deployments with Amp

Use this prompt‑first workflow to plan strategy, promote safely, validate health, and keep a clean audit trail with reversible paths.

Note: Run Amp from your repo root (CLI) or open the workspace in your IDE; there’s no need to include repository paths in prompts.

Why this phase matters

  • Moves changes to users with minimal disruption and clear guardrails.
  • Balances velocity and safety via progressive exposure and gates.
  • Produces an auditable trail of intent, actions, and outcomes.
  • Captures learning for faster, safer future deploys.

Pre‑flight checklist

  • Strategy: rolling, blue‑green, or canary; rationale and fallback.
  • Backout: clear rollback steps, ownership, and time budget.
  • Observability: health checks, error/latency/SLO dashboards, alert thresholds.
  • Data safety: backward‑compatible migrations, flag‑gated behavior, recovery path.
  • Change control: approvals recorded, window confirmed, comms plan ready.

1) Plan the change (strategy and gates)

Prompt (single message):

Select a deployment strategy (rolling, blue‑green, canary) for <service>. Define traffic ramps,
entry/exit gates, kill‑switches, and database compatibility approach. List required dashboards and alerts.

What you’ll get:

  • Strategy with ramps and stop conditions
  • DB compatibility and flag defaults
  • Observability plan for promotion

2) Implement (deploy & promote)

Prompt (single message):

Prepare steps to deploy to pre‑prod, verify, then canary, ramp by cohorts/percent, and promote to full.
Keep feature flags default‑safe and configuration decoupled. Provide a step‑by‑step checklist.

What you’ll get:

  • Promotion checklist with ramp schedule
  • Flag and config guidance
  • Owners per step

3) Validate (smoke & health)

Prompt (single message):

Run smoke tests; verify health, error rate, latency, and saturation against baselines and SLOs.
Confirm key user journeys and downstream/backward compatibility. Capture artifacts for audit.

What you’ll get:

  • Health and journey verification results
  • SLO burn‑rate view and regression flags
  • Artifacts for audit trail

4) Read the history (incidents & rollback triggers)

Prompt (single message):

Review prior incidents and known risky areas for <service>. Define rollback triggers (e.g., error rate,
p95 latency, SLO burn) and decision points (stop/pause/rollback) with owners.

What you’ll get:

  • Context from past incidents
  • Concrete rollback triggers and owners
  • Decision points synchronized to the ramp

5) Risk & compliance (controls)

Prompt (single message):

Ensure approvals, audit notes, change ticket linkage, and freeze exceptions. Capture evidence of checks,
sign‑offs, and timing within change windows. Provide a post‑deploy summary template.

What you’ll get:

  • Change management evidence
  • Post‑deploy summary template
  • Clear audit trail coverage

Exit criteria and handoffs

  • Metrics stable vs baseline; SLO burn within budget; no critical errors
  • Smoke tests pass; critical journeys verified; flags in intended state
  • Dashboards/alerts updated; runbook deltas captured
  • Release notes and change ticket finalized; audit trail complete
  • Ops/on‑call briefed; support informed of known issues

Quick reference

Prompt templates

Strategy: plan rolling/blue‑green/canary; ramps, gates, DB compatibility, required dashboards.
Promote: pre‑prod → canary → cohorts → full; step‑by‑step checklist with owners.
Validate: smoke, health, journeys, SLOs; capture artifacts for audit.
History: prior incidents and rollback triggers with owners and thresholds.
Compliance: approvals, tickets, evidence, and post‑deploy summary.

Developer‑run verification (optional)

# Create a release tag and branch
git tag -a vX.Y.Z -m "Release vX.Y.Z" && git push origin vX.Y.Z

Closing note

Use Amp to plan, gate, and document every deployment. Amp chooses tools automatically and returns synthesized results—not raw logs.