Windmill Production Deployment Checklist

Use this Windmill production deployment checklist to move a tested automation live, configure OAuth, verify reruns, and control seats without surprises.

Nov 18, 2026
Windmill Production Deployment Checklist
A Windmill production deployment checklist should cover more than clicking deploy. Before moving an automation from dry run to live, confirm the production workspace, re-enter production credentials, configure every OAuth redirect URI, test the live path, verify reruns and tenant behavior, and review who needs a paid seat. This sequence reduces the chance that a workflow works in development but fails at the handoff.
Photo by Chris Ried on Unsplash
Photo by Chris Ried on Unsplash

1. Confirm the production deployment path

Identify the target production workspace and the person responsible for approving the release.
Confirm whether the project uses Windmill's single-workspace draft-and-deploy flow, a dev/prod workspace pair, Git promotion, or UI deployment to a paired target.
Record the version, flow, scripts, resources, variables, and dependencies included in the release.
Keep the tested dry-run version available until the production run has passed.
Windmill documents several production patterns, including draft-and-deploy, workspace forks, Git promotion, and UI deployment to a paired target. It also supports attaching test scripts to scripts or flows so regressions can be caught during deployment. 1
A useful release note can be as short as: “This flow is moving from dry run to production, these dependencies are included, and this is the rollback version.” That gives the next operator a clear handoff instead of forcing them to reconstruct the change from memory. For a broader view of how staged rollout fits into an automation program, see Crawl, Walk, Run: the smart way to start automation projects.

2. Re-enter production credentials in Windmill

Confirm which credentials are required by the live flow.
Ask the credential owner to provide or reconnect production credentials in the production Windmill workspace.
Do not assume credentials from a dry-run or development workspace are available in production.
Check that each resource points to the production account, endpoint, folder, project, or tenant.
Verify that secrets are stored in Windmill's resource or variable configuration rather than pasted into scripts, logs, or handoff documents.
Run a permission check with the least access needed for the workflow.
Treat credential setup as a release dependency, not a last-minute troubleshooting task. A flow can be logically correct and still fail because one resource is connected to a test account, one secret was never added to production, or a token belongs to a user who is not allowed to run the live operation.

3. Add and test OAuth redirect URLs

List every production account that needs an OAuth connection.
In Google Cloud Console, open the OAuth client used by the production integration.
Add the exact authorized redirect URI required by that Windmill instance and provider.
Repeat the configuration for each production account or OAuth client where the redirect URI is not shared.
Save the change, then complete a fresh authorization flow in a private browser window.
Confirm that the callback returns to the intended production workspace and creates the expected resource.
Remove obsolete redirect URIs only after confirming that no active connection still depends on them.
The URI is provider- and instance-specific. Windmill's documentation shows provider-specific callback patterns, such as https://<YOUR_INSTANCE>/oauth/callback/gcloud for Google Cloud resources and https://<YOUR_INSTANCE>/user/login_callback/google for Google SSO. Copy the exact value required for the connection you are configuring instead of reusing a URI from another account. 2
Do not mark OAuth complete just because the URI is visible in Google Cloud. The acceptance test is a new authorization attempt that reaches the correct production callback and leaves a usable connection in Windmill.

4. Promote the tested automation

Confirm the dry-run inputs, expected outputs, and any records that must not be modified during testing.
Run one controlled production test with a known-safe input.
Compare the output with the dry-run result, including file names, mappings, notifications, and downstream records.
Deploy the approved flow and all required dependencies together.
Capture the deployed version or commit identifier.
Keep a rollback path: the prior deployed version, a workspace fork, or a reversible configuration change.
If the flow was built as code or scripts, add a visual or written walkthrough for operators who will support it later. A production handoff should make the main path, failure points, and retry behavior inspectable without requiring the original builder to explain every step.
For teams that maintain several automations, Documentation best practices in Zapier offers a related principle: document the operational decisions that make a workflow maintainable, not just the steps that make it run once.

5. Verify reruns, tenants, and idempotency

Run the production flow once with the approved test input.
Rerun it deliberately and confirm whether the result should be skipped, updated, or safely duplicated.
Check tenant or account creation behavior on a rerun.
Distinguish an expected “already exists” response from a genuine failure.
Confirm that partial completion does not leave records in a state that blocks the next run.
Record the correct retry action for each known error.
A rerun that reports “already a tenant” may be harmless if the tenant already exists and the following steps complete normally. It is not harmless if the flow stops before the intended work is finished. Test reruns as a separate scenario because first-run success does not prove that the automation is idempotent.

6. Check production access and seat costs

List the people who need to administer, troubleshoot, or approve the production workflow.
Remove or disable user accounts that no longer need access, according to your access policy.
Decide who should receive production access and who only needs an output or notification.
Confirm each person's Windmill role (builder or non-builder) and the current plan's seat pricing before inviting users.
Review the seat count by role after every access change.
Windmill prices seats by role (as of November 2026). A builder, meaning a workspace admin or anyone whose role can write code or build flows or apps, counts as a full seat. A user who only runs, views, or manages what others built counts as half a seat, and so does each enabled service account. App guests who sign in through your identity provider take no seat up to a monthly limit. Check the current plan before making access decisions, because allowances differ between editions. 3

7. Complete the handoff

Name the production owner and backup owner.
Store the flow name, deployed version, credential owner, OAuth client, and rollback method in the team runbook.
Add the next review date for permissions, redirect URIs, and seat count.
Share the expected success signal and the first place to look when a run fails.
Mark the release complete only after the production test and rerun checks pass.
A production deployment is complete when another operator can verify the live result, understand what is safe to rerun, and restore the previous version without contacting the builder for basic context.

Windmill production deployment checklist: final check

Before you close the release, confirm all of the following:
The production workspace and deployment path are documented.
Production credentials are connected and permissions are tested.
OAuth redirect URLs are configured for every required account.
A controlled live run passed.
A rerun produced the expected behavior.
Tenant or duplicate handling is understood.
The rollback version is known.
Production access and seat counts are reviewed.
The owner, backup owner, and runbook are in place.

Get help shipping your Windmill release

Windmill releases usually stall on the handoff, not the flow logic: a resource still pointed at a test account, a redirect URI missing for one OAuth client, or a rerun nobody tested. If that's where your release is stuck, book a discovery call and one of our consultants will map the production rollout with you.