Cadence Workflow

What it is

Cadence is an orchestration engine for running distributed application logic as durable, fault-tolerant workflows. You write business logic as ordinary code (loops, conditionals, function calls); Cadence persists execution state at every step so that if a worker crashes, a host dies, or a process gets OOM-killed, the workflow resumes exactly where it left off — no replay logic, no manual checkpointing. It's built for long-running, stateful processes (order fulfillment, provisioning pipelines, CI/CD orchestration, data pipelines) that need to survive failures measured in days or weeks, not milliseconds.

Under "Provisioning" here, think infrastructure automation and multi-step orchestration workflows rather than raw compute scheduling — Cadence sits above your infra, coordinating the steps.

Who builds it and why

Cadence originated at Uber to solve their own problem: coordinating long-lived, failure-prone business processes (trip lifecycle, driver onboarding, payments) without hand-rolling state machines and retry logic in every service. It's now maintained as an open governance project with contributions from Uber and external adopters running it in production at scale. 178 contributors and continuous commit activity (last commit 2026-07-01) indicate it's not a dead or single-vendor pet project — it has a real contributor base outside its origin company.

Worth noting: Temporal is a fork of Cadence, started by the original Cadence engineers after they left Uber. If you're evaluating Cadence, you're implicitly also evaluating that split and what it means for long-term direction.

Production readiness signal

• 9,361 GitHub stars, 178 contributors — healthy community size, not a niche experiment.
• CNCF Sandbox — early-stage in CNCF terms; this is a signal of governance intent, not maturity. Sandbox projects have no CNCF guarantees around API stability, security review, or long-term support.
• v1.4.1 — a mature version number for a project running in production at Uber-scale for years, but check the changelog for breaking changes between minors; workflow orchestration engines are unforgiving of subtle behavioral changes across versions.
• Apache-2.0, Go — standard, no licensing friction.
• Founded date: not publicly available.

Bottom line: this is real, battle-tested infrastructure at its origin company, but CNCF sandbox status means you're doing your own due diligence on governance and roadmap stability, not leaning on a foundation's maturity ladder.

Who should use this

• Teams running long-lived, multi-step business processes that must survive process/host failure without manual recovery scripts.
• Organizations already invested in Go-centric infrastructure who want workflow-as-code rather than YAML/DSL-based orchestration.
• Teams with the ops capacity to run and tune a stateful distributed system (Cadence requires a persistence layer — Cassandra, MySQL, or Postgres — plus its own service cluster).
• Anyone who evaluated Temporal and specifically wants Cadence's server architecture or existing Uber-scale operational patterns.

Who should NOT use this

• Teams wanting a managed, low-ops experience — Cadence self-hosting requires real operational investment (persistence tuning, cluster sizing, upgrade discipline). There's less commercial managed-hosting maturity around Cadence than around Temporal Cloud.
• Simple, short-lived task queues — if you just need retries and a job queue, Cadence is overkill; use something lighter (Sidekiq, Celery, cloud-native queue + Lambda).
• Teams wanting the most active ecosystem, docs, and SDK breadth for workflow orchestration today — Temporal has pulled ahead of Cadence in adoption momentum, SDK coverage, and community size since the fork.
• Anyone needing CNCF-graduated-level governance guarantees now — Cadence is sandbox-stage; don't assume foundation-backed long-term support.

Alternatives

• Temporal — direct fork of Cadence with a more active ecosystem, broader SDK support, and a managed cloud offering; the default comparison point.
• AWS Step Functions — fully managed, serverless workflow orchestration if you're all-in on AWS and want zero infrastructure ops.
• Apache Airflow — better fit for DAG-style data/ETL pipelines rather than long-running stateful business logic.

Pricing

Fully open source (Apache-2.0). No license cost. You pay in infrastructure (persistence store, Cadence server cluster) and operational engineering time to run it yourself. No official managed-hosting offering from the core project as of this writing — verify current status before committing, since this changes faster than docs get updated.