Chapter Five

Messaging & Async

When components stop calling each other directly, they exchange messages, schedule work, orchestrate steps, and react to events. Five services for the decoupling work that makes a real distributed system tractable.

5 services

The decisions in this chapter shape how a system holds together at the seams. Choosing Pub/Sub where Cloud Tasks is the right tool, or hand-rolling orchestration with chained Pub/Sub messages instead of Workflows, leads to brittle architectures that resist debugging. Each of these five services has a narrow niche; learning the niches matters as much as learning the APIs.

Pick by the Pattern, Not the API

One message fans out to many independent subscriptions (topics → subscriptions)→

Pub/Sub

Deliver tasks to one target (HTTP or App Engine), rate-limited with retries→

Cloud Tasks

Trigger work on a schedule (cron)→

Cloud Scheduler

Coordinate a multi-step sequence with state and error handling→

Cloud Workflows

Route events from Google Cloud, SaaS, or custom (CloudEvents) sources to handlers→

Eventarc

Services in This Chapter

Global messaging service. Topics and subscriptions, push and pull delivery, at-least-once with optional exactly-once, ordering keys, dead letter topics, schemas.

MessagingPub-Sub

Managed task queue for rate-limited HTTP work with retries. The right tool when Pub/Sub fan-out is not what you need.

Cloud Scheduler

Managed cron. HTTP, Pub/Sub, and App Engine targets with retry policies and timezone-aware schedules.

Cloud Workflows

Serverless orchestration. YAML-defined state machines with native GCP connectors, error handling, parallel steps, and long-running executions.

OrchestrationServerless

Event router. Triggers Cloud Run, Cloud Functions, GKE, and Workflows from 90+ GCP event sources in the CloudEvents format.