Deploy & Rollback

Deploy & Rollback

Agent Stack Deploy

# Build + deploy in one step
cd pulsar && uv run pulsar images build agent --env <ENV> --deploy

# Or build then deploy separately
cd pulsar && uv run pulsar images build agent --env <ENV>
cd pulsar && uv run pulsar deploy agent --env <ENV> --version <TAG>

Production requires explicit confirmation — the CLI will prompt “BUILD PROD” / “DEPLOY TO PROD”.

Core Stack Deploy

# Requires AWS credentials
export AWS_PROFILE=pulsestream

# Build + deploy
cd pulsar && uv run pulsar images build core --env <ENV> --deploy

# Single component only (tag must match running deployment)
cd pulsar && uv run pulsar images build core --env <ENV> --component frontend

Partial builds

When building a single component, the image tag must match what’s currently deployed. Check with:

kubectl get deployment -n <namespace> -o jsonpath='{.items[*].spec.template.spec.containers[*].image}'

Emergency Rollback

Option 1: Helm Rollback (fastest)

# Check revision history
helm history pulsestream-agent -n pulsestream-agent-<ENV>

# Roll back to previous revision
helm rollback pulsestream-agent <REVISION> -n pulsestream-agent-<ENV>

Option 2: Deploy Previous Version

cd pulsar && uv run pulsar deploy agent --env <ENV> --version <PREVIOUS_TAG>

Option 3: Scale Down (circuit breaker)

cd pulsar && uv run pulsar scale down --env <ENV>
# Investigate, then:
cd pulsar && uv run pulsar scale up --env <ENV>

Post-Deploy Verification

1. Check pod rollout: kubectl rollout status deployment/<service> -n <namespace>
2. Verify health endpoints respond
3. Check agent connectivity via MCP: get_system_health
4. Monitor error rates in Grafana for 10 minutes

Dry Run

Always dry-run first in non-emergency situations:

cd pulsar && uv run pulsar images build agent --env <ENV> --dry-run
cd pulsar && uv run pulsar deploy agent --env <ENV> --version <TAG> --dry-run