Guidewire CI CD Pipeline

Guidewire CI/CD Pipeline

Overview

Ship Gosu changes from a developer's branch to production without breaking running policies. Guidewire Cloud Console (GCC) deployment slots are the carrier-side promotion mechanism — dev → uat → prod — with config packages as the unit of promotion. The CI side validates compile, runs GUnit, and packages; the CD side promotes packages through slots with deploy-time gates.

Five production failures this skill prevents:

1. Bypassed compile/test gates — a gradle build failure on a developer's machine reaches CI as "works on my machine"; CI runs the same gates as a clean clone, fails fast, never lands on a slot.
2. Promotion without UAT regression — UW logic change goes dev → prod skipping UAT; the change interacts badly with a product line not exercised in dev; bound policies start showing wrong premium.
3. Schema change without rollback plan — a Gosu interface adds a required field; rolling back the deploy is fine, but rolling back the database column requires a separate migration that may have been forgotten.
4. Canary that isn't a canary — "canary" deploy serves 100% of traffic immediately because the slot router has no per-slot routing; a real canary needs traffic split, not just a separate slot.
5. Rollback after policies bound on the new code — a defective rate-plan deploy bound 200 policies in the 30 minutes before detection; reverting the code does not unbind the policies; they need targeted endorsement-correction or rate adjustment.

Prerequisites

• A working Guidewire Studio + runServer setup per guidewire-local-dev-loop
• GCC tenant with at least three slots (dev, uat, prod) configured for the integration's product line
• A CI runner (GitHub Actions, GitLab CI, Jenkins) with JDK 17 and access to a private artifact registry for config packages
• An age private key (per guidewire-security-and-rbac) provisioned to CI for decrypting environment-specific secrets

Instructions

Build the pipeline in this order. Each step targets one of the five production failures listed in Overview.

1. PR-time gates (run on every push)

Three checks, all blocking. The PR cannot merge to main if any fails.

# .github/workflows/ci.yml
jobs:
  validate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-java@v4
        with: { java-version: 17, distribution: temurin }
      - run: ./gradlew compileGosu                       # gate 1: code compiles
      - run: ./gradlew test                              # gate 2: GUnit passes
      - run: ./gradlew check                             # gate 3: lint + arch + style

compileGosu catches type errors that Studio's incremental compile sometimes misses. test runs the GUnit suite — every UW rule, validation rule, and entity-level Gosu has tests, and CI runs them all in <5 minutes for a normal config zone. check includes static analysis (ban on direct DB writes from Gosu, ban on certain plugin patterns, etc.).

2. Build the config package on merge to main

package:
  needs: validate
  if: github.ref == 'refs/heads/main'
  steps:
    - uses: actions/checkout@v4
    - run: ./gradlew packageConfig
    - run: |
        VERSION=$(git describe --tags --always)
        aws s3 cp build/libs/configuration.zip \
          s3://gw-config-artifacts/${{ github.event.repository.name }}/${VERSION}.zip

Config packages are immutable once published. Promotion deploys an existing package to a slot; never modifies it in place. The package name should encode the git SHA so traceability is exact.

3. Deploy to dev slot (continuous)

Every merge to main deploys to dev automatically. Dev is the first place real Cloud API runtime exercises the change.

# Promotion API (illustrative — exact GCC API surface varies by tenant)
curl -X POST "$GCC_API/slots/dev/deploy" \
  -H "Authorization: Bearer $GCC_TOKEN" \
  -d '{"package": "configuration-abc1234.zip"}'

Wait for deploy to complete (poll status), then run smoke tests against the dev tenant — at minimum, the verification probe from guidewire-install-auth and a synthetic FNOL from guidewire-observability-and-incident-response. If smoke fails, page the on-call author.

4. Promote to UAT with regression gate

UAT is where the product team's regression suite runs. Promotion to UAT is manual (a button in the CD tool) or scheduled (nightly). UAT regression covers:

• A representative submission per active product line (Personal Auto, BOP, Workers Comp, etc.)
• Quote-and-bind for each, asserting expected premium ranges
• A FNOL per loss-cause, asserting the claim opens with the expected exposure shape
• Renewal of a sample in-force policy, asserting the renewal premium falls in expected band

Failed regression blocks promotion to prod automatically. Manual override requires an approver outside the deploying team.

5. Canary to prod with traffic split

For high-risk changes (rate plan changes, UW rule logic, payment authorization changes), use a true canary: a separate prod slot serving a percentage of real traffic. GCC supports slot-level traffic splitting on tenants where it's enabled.

prod slot:    serves 95% of traffic, runs the previous package
prod-canary:  serves 5%, runs the new package

Watch the canary's bind-success-rate, FNOL p99 latency, and 4xx error rate against the main prod slot. After 24h with no SLO breach, ramp to 100% and retire the canary slot.

For low-risk changes (UI text, localization, non-rule Gosu utility classes), skip the canary — single-step promotion to prod is acceptable.

6. Rollback decision tree

A bad change is detected. The rollback decision is not "revert the package":

Bad change detected in prod
├─ Has any policy been bound or claim opened against the bad code? → check the integration_audit table
│   ├─ NO → revert the package; deploy the previous version; done
│   └─ YES → revert the package AND determine if the bound resources are recoverable
│           ├─ Premium calculation wrong → script-driven rate adjustment endorsements (per affected policy)
│           ├─ Reserve / payment wrong → finance-team-led correction; do not silent-fix
│           └─ FNOL routing wrong → rerun routing logic against affected claims
└─ No policy/claim impact, just operational error (logs noisy, dashboard wrong) → revert and PIR

The bound-state recovery path is what makes Guidewire rollbacks different from typical SaaS rollbacks. Code reverts, regulatory state does not.