Skip to content

Introduction ​

Stackorder decides which Terraform or OpenTofu stacks a change affects and in what order to apply them. GitHub Actions does all of the running.

It is a GitHub App plus a small control-plane server. Execution, credentials, state and modules stay inside your GitHub organization or personal account and your AWS account. The server receives metadata and redacted, size-capped plan text, never cloud credentials. It stores no Terraform state, and reads none unless an operator turns on shared outputs: its AWS role can then read the state objects the operator lists, from which the server returns only the root-module outputs, to the stacks each producer lists as readers and to API keys. Who may read says what the server can prove about a reader.

About these pages

The guide, configuration, reference and operations pages describe the code as released. The design document explains why it works this way, with implementation notes where the code departs from it, and the architecture contract pins the names and shapes.

The two jobs of the server ​

By default the server does two things.

  1. Dependency resolution. It holds the graph of stacks, shared modules and the edges between them. For a change it computes the affected set, orders applies into waves, and serializes conflicting work with stack-level locks.
  2. Observability. It records every plan and apply per stack and per commit, surfaces drift, lists the resources in each stack's state with the applied run that last changed each one, and shows the dependency graph and which stacks consume each module at which version. All of it is available through a small web UI, a JSON API and Prometheus metrics.

An operator can add a third, off by default: shared outputs, where the server returns a stack's root-module outputs to the stacks that stack lists as readers, so their roles need no access to its state.

The Stackorder web UI overview: counts of repositories, stacks, drifted stacks and locks held, bars of stacks and runs by status, and a table of recent plan, apply and drift runs with their stacks, planned changes and durations.The Stackorder web UI overview: counts of repositories, stacks, drifted stacks and locks held, bars of stacks and runs by status, and a table of recent plan, apply and drift runs with their stacks, planned changes and durations.
The web UI's overview page, with sample data for an acme organization.

What it is not ​

Stackorder is deliberately not a state backend, a module registry, a secrets store, a policy engine or a runner.

Not aUse instead
State backend or state viewerThe S3 backend, with S3-native or DynamoDB locking. Stackorder stores where each stack's state lives, to link to it and, with shared outputs on, to serve the stack's root-module outputs to the stacks it lists as readers, and the list of resources in it, without their values. It never stores or shows the state itself.
Module registryGit or local paths. Stackorder indexes module references; it does not host code.
Policy engineOPA or conftest, Checkov, Infracost or any other tool as a step in the plan job. Stackorder records the verdict as a named check on the stack.
RunnerGitHub-hosted or self-hosted Actions runners, under your own OIDC-federated AWS role.
Multi-VCS toolNothing. Stackorder is GitHub only: the App, the OIDC claims and the check-run model are where its leverage comes from.
AWS role managerYour own IAM. You create one OIDC-trusted role per environment; Stackorder only tells the workflow which stack is running.

The pitch ​

The execution model of Stategraph (formerly Terrateam), where Terraform runs on your own CI runners, with one server container plus Postgres, no Docker on the runner side, and dependencies (stack to stack, stack to module, across repos) as the server's core data structure.

How it compares ​

HCP TerraformTerrakubeStategraphStackorder
Where Terraform runsHashiCorp-hosted VMs by default, or self-hosted agentsIts own executors: a pod pool, Kubernetes Jobs or self-hosted agentsYour GitHub Actions or GitLab CI runnersGitHub Actions
State backendBuilt inBuilt in, on its configured object storageBring your ownBring your own S3
Module registry and trackingBuilt-in private registry; the Explorer shows module usageBuilt-in private module and provider registryA module-aware indexer, off by default, plans the directories that use a changed local moduleNo registry; lists each module's consumers, and for git modules how many releases they are behind
Runtime footprintSaaS; self-hosted Terraform Enterprise runs containers with PostgreSQL, object storage and VaultAPI, executor, registry, UI, Dex (with OpenLDAP by default), a Redis-compatible store, object storage and PostgresServer + Postgres behind a public HTTPS URL; Docker container action on the runnerOne container + Postgres; non-Docker actions
Cross-stack dependenciesRun triggers between workspaces; linked StacksShared remote state in the stable 2.33 line; run triggers only in 2.34 pre-releasesLayered runs within one repositoryA graph of stacks, modules and cross-repo edges; applies in waves
Cloud credentials held by serverYes: stored as variables, or short-lived per-run credentials through OIDCYes: stored as variables; with dynamic credentials it holds an OIDC signing key and mints tokensNo; they stay on the runnerNo; with shared outputs on, read access to the state objects an operator lists

The comparison page goes through each row and links the source of every competitor claim, last reviewed 2026-09-30.

What a repository needs ​

Next steps ​

  • How it works: the trust zones, the execution model and the pull request lifecycle.
  • Concepts: stacks, modules, edges, waves, locks and drift.
  • Getting started: from an empty AWS account to a first stackorder apply.
  • Local demo: the server, the CLI and real plans on one machine, with no GitHub App or AWS account.