Cloud Run Complete Guide: Services, Jobs, Sidecars (2026)

Cloud Run is Google Cloud's serverless container platform. You can deploy any Docker image in any language and it auto-scales for you. Built on Knative, it supports both HTTP-request-driven services and batch jobs. Alongside GKE, it has become a first-choice option for microservices.

Cloud Run Key Features

Serverless: no cluster management — just bring a container
Auto scaling: 0 to N instances (default max 1000)
Pay-per-use: billed only for CPU and memory time during request processing
Any language: anything that builds into a Docker image (Go, Python, Java, Node.js, Rust, .NET, and more)
Custom domains: Google-managed SSL certificates supported
Long-running execution: up to 60 minutes per request (7 days per job)

Services and Jobs

Item	Service	Job
Trigger	HTTP request / Eventarc / Pub/Sub push	Manual / Cloud Scheduler / Workflows
Execution time	Up to 60 minutes per request	Up to 7 days per task
Parallelism	Horizontally scales with concurrent requests	Parallel tasks via the parallelism setting
Primary use case	Web APIs and microservices	Batch processing, ETL, report generation

Scaling Configuration

Min instances: 0 (default, cold starts apply) or 1+ (always-on)
Max instances: defaults to 100, configurable up to 1000
Concurrency: concurrent requests per instance (default 80)
CPU allocation: request-based (only during requests) or always allocated

Pricing

Item	Price (us-central1, 2026)
vCPU time	$0.00002400 per vCPU-second (during request processing)
Memory time	$0.00000250 per GiB-second
Request count	$0.40 per 1 million requests
Free tier	240,000 vCPU-seconds, 450,000 GiB-seconds, and 2 million requests per month

Cloud Run vs Cloud Functions vs GKE

Item	Cloud Run	Cloud Functions (2nd gen)	GKE Autopilot
Base	Container	Container (internally)	Kubernetes
Language	Any	Node / Python / Go / Java / .NET / Ruby / PHP	Any
Max execution time	60 minutes (service)	9 minutes (HTTP) / 60 minutes (event)	Unlimited (pod)
Scaling	0 → 1000	0 → 3000	Adjust pod count
Pricing	Pay-per-use	Pay-per-use	Pod time (vCPU/memory)
Primary use case	General microservices	Event-driven	Complex orchestration

Typical Architectures

Web API + Database

Cloud Run (backend) + Cloud SQL (PostgreSQL) + Cloud Storage (images) + IAP / Cloud Armor (security)

Event-Driven Microservices

Pub/Sub → Cloud Run (push) → writes to Firestore / BigQuery

Scheduled Batch Processing

Cloud Scheduler → Cloud Run job → processes BigQuery / GCS data

Multi-Region Deployment

Global external HTTPS load balancer + Cloud Run services in multiple regions + Cloud CDN

CI/CD Pipeline

Push code to GitHub or Cloud Source Repositories
Trigger an automated build with Cloud Build
Store the container image in Artifact Registry
Run canary or blue/green deployments via Cloud Deploy
Allow only signed images via Binary Authorization
Monitor with Cloud Monitoring and Cloud Logging

Security Best Practices

Create a dedicated service account (do not use the default SA)
Scope roles/run.invoker as narrowly as possible in IAM
Connect to internal resources via a VPC connector and Serverless VPC Access
Combine Cloud Armor and IAP for WAF and zero-trust access
Manage API keys and DB connection strings with Secret Manager
Verify image signatures with Binary Authorization

What is the difference between Cloud Run and Cloud Functions?

Cloud Run is container-based, supports any language, and handles long-running requests (up to 60 minutes). Cloud Functions is source-code based, supports specific languages, and targets short-running workloads. Cloud Functions 2nd gen actually runs on top of Cloud Run internally.

What is the difference between Cloud Run services and jobs?

Services are HTTP request-driven (and can stay always-on). Jobs are batch workloads that run until completion or timeout and then exit. Use Cloud Scheduler to trigger jobs on a schedule.

Should I choose GKE or Cloud Run?

If you want Google to handle container scaling for you, pick Cloud Run. If you need the Kubernetes API, custom workloads, or StatefulSets, pick GKE. Cloud Run is overwhelmingly easier for microservices.

How long is a cold start?

It depends on the base image and the application. Go, Rust, and Node.js typically come in under 100 ms; Java and Python take 1-3 seconds. Setting min instances to 1 eliminates cold starts but incurs always-on cost.

How does pricing work?

You pay for CPU time, memory time, and request count. Billing is request-based (CPU is allocated only during request processing) or always-on. The free tier includes 2 million requests per month.

Can Cloud Run hold state?

Cloud Run is stateless by design. Externalize state to Firestore, Cloud SQL, or Memorystore. Persistent volumes are not supported. Cloud Run Volumes (Cloud Storage mounts) are available in a limited form.

How do you set up authentication?

You can make services public (allUsers) or restrict them via IAM (roles/run.invoker). Putting Cloud Endpoints, API Gateway, or Apigee in front is also common. Service-to-service auth uses a service account plus an ID token.

How do you deploy Cloud Run across regions?

Yes, this is supported. Register Cloud Run services from multiple regions as backends behind a global external HTTPS load balancer, and traffic will route automatically to the closest region.

Check what you learned with practice questions

Practice with certification-focused question sets

View GCP Exam Prep

Author

NicheeLab Editorial Team

NicheeLab editorial team focused on data engineering and cloud certification learning. Content is structured around practical study needs and official exam domains.

Cloud Run Complete Guide: Serverless Container Platform Features, Pricing, and GKE Comparison