Skip to content

k8s

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

k8s

README.md

Kubernetes deployment

Minimal, opinionated manifests for running c3x-pricing-api on Kubernetes.

Contents

File Purpose
namespace.yaml c3x-pricing namespace.
secret.example.yaml Template for DATABASE_URL, API_KEY, GCP_API_KEY. Do not commit a filled copy.
deployment.yaml API server Deployment + Service with /healthz liveness and /readyz readiness probes.
cronjob-aws.yaml Weekly AWS scrape (Sunday 03:00 UTC).
cronjob-azure.yaml Daily Azure scrape (03:15 UTC).
cronjob-gcp.yaml Daily GCP scrape (03:30 UTC).

Quick start

The manifests reference ghcr.io/c3xdev/c3x-pricing-api:IMAGE_TAG as a placeholder. Substitute it for a concrete release version (recommended) or for latest (not recommended for production — no rollback safety, surprise upgrades on pod restart).

# 1. Pick a version. Browse https://github.com/c3xdev/c3x-pricing-api/releases
#    or use the latest tag programmatically:
VERSION=$(curl -fsSL https://api.github.com/repos/c3xdev/c3x-pricing-api/releases/latest | grep -oE '"tag_name": *"[^"]+' | cut -d'"' -f4)
echo "Deploying $VERSION"

# 2. Apply with substitution
kubectl apply -f namespace.yaml
kubectl apply -f secret.example.yaml  # edit first!
for f in deployment.yaml cronjob-aws.yaml cronjob-azure.yaml cronjob-gcp.yaml; do
  sed "s|:IMAGE_TAG|:$VERSION|g" "$f" | kubectl apply -f -
done

To upgrade, repeat with the new $VERSION; kubectl rollout undo then correctly rolls back to the previously-deployed image.

Design notes

  • No leader election needed. Scrapes use a Postgres advisory lock. Two concurrent runs of the same vendor will just have one skip.
  • Separate CronJobs per vendor so a slow AWS run does not delay the Azure schedule, and so kubectl get cronjob shows you exactly which vendor is stale.
  • concurrencyPolicy: Forbid means a still-running scrape will not be interrupted by the next tick. The advisory lock is a belt-and-braces safety.
  • Probes hit /healthz (liveness, no DB) and /readyz (readiness, DB ping) so a transient DB blip does not restart your pod.
  • You are expected to bring your own Postgres (managed is recommended). Adjust the Secret accordingly.