Skip to main content
Drift is the gap between what an environment should run and what it does run. It has two directions: the platform runs something other than the release you promoted, or your deploy/kcl/ declares something nobody has deployed yet. Forge answers each with one read-only command, and a third for workloads that something other than forge deploys. Forge can answer these without logging in to a cluster because every release records the digests it pins and every deploy records the shape it applied (What gets deployed).

Before you begin

1

Check the environment runs its release

env status compares the images the platform serves with the digests of the release the environment is promoted to:
terminal
DRIFTED means a different image is running than the release pins, and MISSING means a pinned image isn’t running at all. UNREACHABLE with UserSuspended is an environment someone stopped.
2

Find what your checkout declares but nobody deployed

env diff renders this checkout and compares it, object by object, with the bundle the last deploy recorded. Here someone added a worker workload to deploy/kcl/prod/main.k and never deployed it:
terminal
It also reports removed objects, image changes, runtime and cluster moves, secrets newly needed and domain changes. Nothing is built, pushed or contacted. prod: ok is the render’s status; error, impure or stale means the comparison didn’t happen, and is never reported as “no changes”. Use --all to diff every environment.
A deploy records each workload’s release version in its telemetry attributes, and a render of your checkout doesn’t carry one. So right after a release deploy, every workload reads config changed. Read the added, removed and image lines, and confirm what runs with env status.
3

Fail a CI job on undeployed changes

env diff exits 0 whatever it finds, so read its JSON:
terminal
A gate can be jq -e over the same document, failing when status isn’t ok or diff.added or diff.removed is non-empty.
4

Verify workloads forge doesn't run

For a workload on your own cluster or under your own deploy tool, env verify checks the running image and config per workload:
terminal
It exits 0 when every workload matches, 1 on drift, and 2 when something couldn’t be checked. Hosted workloads are read by env status, so on a hosted environment env verify reports them UNVERIFIED and names env status:

What this does not do

  • It doesn’t fix drift. All three commands only read. To converge, deploy: reliant forge env deploy prod ships the checkout, and naming the version the environment already runs redeploys it. See Releases and promotion.
  • It doesn’t watch. Nothing runs these checks on a schedule or alerts on drift. Run them by hand or from CI.
  • Helm charts aren’t compared by default. Templating them needs helm and the network; pass --charts to include them.

Next steps

Keep Helm, Fly.io or Cloud Run

Verify workloads your own tooling deploys.

Recover from a bad deploy

Put an environment back on an earlier release.

Environments

How each environment binds its workloads.

env diff reference

Every flag of env diff, env verify and env status.