Customer 360 & Real-Time Risk Intelligence Platform  ·  View 16 of 20

CI/CD and Environment Promotion

How a pipeline change reaches production, which gates stop a bad one, and how a failed release is rolled back.

Editable source SVG draw.io All views
Source
Source
Pipeline Repository
Spark, Flink, DAGs
Pipeline Repository...
Schema & DQ Rules
Versioned with code
Schema & DQ Rules...
Infrastructure as Code
Terraform modules
Infrastructure as Code...
Build & test
Build & test
Unit & Contract Tests
Sample fixtures
Unit & Contract Tests...
Schema Compatibility
Registry dry run
Schema Compatibility...
Security Scan
SAST, deps, secrets
Security Scan...
Non-production
Non-production
Development
Synthetic data only
Development...
Integration
Masked subset
Integration...
Performance
Peak volume replay
Performance...
Gates
Gates
Quality Regression
Golden dataset diff
Quality Regression...
Change Approval
Risk rated
Change Approval...
Production release
Production release
Blue / Green Jobs
Savepoint restore
Blue / Green Jobs...
DAG Sync
Paused then enabled
DAG Sync...
Feature Flags
Per source rollout
Feature Flags...
Verify
Verify
Smoke & Canary
Shadow consumer
Smoke & Canary...
SLO Watch
30 min burn window
SLO Watch...
Rollback
Previous savepoint
Rollback...
breaking, reject
breaking, reject
regression, return
regression, return
breach
breach
approved
approved
CI/CD Pipeline and Environment Promotion
CI/CD Pipeline and Environment Promotion
Application we own
Application we own
Decision point
Decision point
Risk / gap
Risk / gap
failure / alternate
failure / alternate
synchronous
synchronous
No production dataset is copied downward; lower environments use synthetic or masked data.
No production dataset is copied downward; lower environments use synthetic or masked data.
v 1.0 · owner Platform Engineering · date 2026-08
v 1.0 · owner Platform Engineering · date 2026-08Text is not SVG - cannot display

What is versioned together

  • Pipeline code, schema definitions, quality rules and infrastructure in one repository
  • A change to a rule is a code change, reviewed and released like any other
  • Every production artefact is traceable to a commit and an approval

Environment data policy

  • No production dataset is copied downward — ever
  • Development uses synthetic data; integration uses a masked, tokenised subset
  • Performance testing replays anonymised peak-volume traffic shapes

Safe release

  • Streaming jobs release blue/green from a savepoint, so state is not lost
  • Feature flags roll a change out per source rather than platform-wide
  • A 30-minute SLO burn window gates the release; breach triggers rollback
Solution Architecture v1.0 · Enterprise Data Architecture · 2026-0816-cicd-environments