Deployment Workflows

CI/CD patterns for automated model deployment and testing.

Git-triggered deployments

Connect your repository and Upbox deploys automatically on every push to main. Configure branch rules in `upbox.yaml` to control which branches trigger deployments.

Blue-green deployments

Deploy new versions alongside existing ones, validate with test traffic, then switch over atomically. Rollback is instant if issues are detected.

Canary releases

Route a percentage of traffic to new versions with `upbox deploy --canary 10%`. Gradually increase traffic as confidence grows, with automatic rollback on error spikes.

Shadow deployments

Mirror production traffic to new models without affecting responses. Compare outputs and latency before promoting. Enable with `upbox deploy --shadow`.

Staging environments

Create isolated staging environments with `upbox env create staging`. Deploy and test before promoting to production with `upbox promote staging production`.

Model validation gates

Define validation tests in `upbox.yaml` that run before deployment completes. Tests can check accuracy thresholds, latency requirements, or custom metrics.

GitHub Actions integration

Use the official `upbox/deploy-action` in your workflows. Supports model testing, deployment, and automatic PR comments with deployment URLs.

GitLab CI integration

Include the Upbox template in your `.gitlab-ci.yml` for seamless deployments. Supports review apps that deploy models for each merge request.

Scheduled deployments

Configure deployment windows with cron expressions. Useful for deploying retrained models on a schedule or avoiding deployments during peak hours.

Multi-region deployments

Deploy to specific regions with `upbox deploy --regions us-east,eu-west,ap-southeast`. Traffic automatically routes to the nearest region.