Saltar a contenido

Production: sizing, HA, backup & upgrades

Run Vex Raptor reliably for a team or a client book. This page covers capacity, availability, data durability, and safe upgrades. For the base install see Docker Compose.

Architecture

A production deployment runs five roles:

Service Role Scale by
API HTTP + SSE, auth, orchestration Replicas behind your proxy
Worker Runs scans (ARQ jobs) Add workers for scan throughput
PostgreSQL System of record Vertical + managed HA
Redis Queue, quota, shared scan state Managed/replicated
Web Static console Replicas / CDN

Redis is what lets multiple API/worker replicas agree on quota and scan state — it is required for horizontal scaling.

Sizing

Start here and adjust from real metrics:

Deployment API Workers PostgreSQL Redis
Lab / eval 1 vCPU / 1 GB shared SQLite/1 GB optional
Small team 2 vCPU / 2 GB 1–2 2 vCPU / 4 GB 1 GB
MSSP / heavy 2+ replicas N per concurrency target managed, 4+ vCPU managed HA

Scan concurrency is bounded by worker count and the per-target rate cap. Add workers to raise throughput; raising the rate cap risks disrupting targets.

High availability

  • Run 2+ API replicas and 2+ workers across availability zones.
  • Use managed PostgreSQL with automated failover and PITR.
  • Use managed/replicated Redis.
  • Health probe: GET /health (unversioned, no auth) for load-balancer checks.

Backup & restore

Back up PostgreSQL (system of record) and your report storage. Redis is ephemeral (queue/quota/state) and does not require backup.

# Backup
pg_dump "$DATABASE_URL" | gzip > vex-$(date +%F).sql.gz

# Restore into a fresh database
gunzip -c vex-2026-07-09.sql.gz | psql "$DATABASE_URL"

Never destroy the data volume

docker compose down -v deletes the PostgreSQL volume. Use down (without -v) to stop the stack. Test restores regularly — an untested backup is a hope, not a backup.

Upgrades & migrations

  1. Read the release notes for breaking changes.
  2. Back up PostgreSQL first.

  3. Pull the new image and apply database migrations:

    docker compose pull
docker compose run --rm api alembic upgrade head
docker compose up -d

  4. Verify GET /health and run a smoke scan against a lab target.

Roll back by redeploying the previous image tag; restore the DB backup only if a migration must be reverted.

Pin image tags

Deploy specific version tags (not latest) so replicas are identical and rollbacks are deterministic.