Platforms

Google Cloud Run

Run Rivet workers on Google Cloud Run.

Cloud Run is request-driven, so workers run in serverless mode: the control plane calls your /api/rivet endpoint to start actors, and the service URL has to be public and registered.

Requirements

A Google Cloud project with Cloud Run and Artifact Registry enabled
gcloud authenticated (gcloud auth login) with the project set
A control plane, either Rivet Cloud or your own

Steps

Package your app

FROM node:24-slim
WORKDIR /app
COPY package*.json ./
RUN npm ci --omit=dev
COPY . .
ENV RIVET_PORT=8080
CMD ["node", "server.js"]

RIVET_PORT is the port RivetKit’s listener binds. Cloud Run routes to 8080 by default, and RivetKit does not read Cloud Run’s PORT variable.

Build and push

gcloud builds submit --tag us-central1-docker.pkg.dev/YOUR_PROJECT/rivet-worker/rivet-worker:latest

Deploy

RIVETKIT_RUNTIME_MODE=serverless is required. Without it the app defaults to runner mode, which does not fit Cloud Run’s request-driven model.

gcloud run deploy rivet-worker \
  --image us-central1-docker.pkg.dev/YOUR_PROJECT/rivet-worker/rivet-worker:latest \
  --region us-central1 \
  --allow-unauthenticated \
  --min-instances 1 \
  --set-env-vars RIVET_ENDPOINT=<your-rivet-endpoint>,RIVET_PUBLIC_ENDPOINT=<your-rivet-public-endpoint>,RIVETKIT_RUNTIME_MODE=serverless

Both endpoint values are described in Workers.

Register the URL

Take the service URL from the deploy output, append /api/rivet, and set it as the serverless runner URL in the dashboard:

https://rivet-worker-xxxxx-uc.a.run.app/api/rivet

Verify

gcloud run services describe rivet-worker --region us-central1 \
  --format 'value(status.conditions[?type="Ready"].status)'

Expect True. The service then appears as connected in the dashboard.

Scaling and timeouts

Three Cloud Run settings interact with how many actors you can host:

--concurrency caps in-flight requests per instance, and each in-flight /api/rivet/start hosts one actor. It is therefore your actors-per-instance limit. Size it from per-actor memory and CPU; Cloud Run scales out more instances once instances hit the cap.
--max-instances caps how far Cloud Run scales out. Together with --concurrency it bounds the whole service at concurrency × max-instances actors. The deploy command above sets --min-instances 1, which is only the floor, so set the ceiling deliberately rather than inheriting the default.
--timeout caps request duration at 60 minutes. Set requestLifespan just under it, around 840 for a 15-minute timeout, so actors migrate on your schedule rather than being cut off. Lower drainGracePeriod to match: it defaults to 1800 and must be strictly less than requestLifespan. See the production checklist.

Actor start requests all originate from your control plane’s IPs, so any per-IP rate limit throttles the control plane long before real user traffic.