A year after Cloudflare’s November 18, 2025 outage, one detail from its incident report still deserves a place in recovery exercises. Turnstile became unavailable, and users without an active dashboard session could not log in to the Cloudflare dashboard. A service used to protect the login page had become a dependency for accessing operational controls. Cloudflare incident report
For infrastructure teams reviewing their own recovery plans, that raises a practical question. If a provider or authentication service fails, can the people responsible for recovery still reach the tools they need?
What happened
Cloudflare reported that a database permissions change caused duplicate entries to appear in a feature file used by Bot Management. The file grew beyond an expected limit and was distributed across the network, triggering failures in traffic processing. The incident affected several services, including the dashboard login flow. Cloudflare incident report
Cloudflare later announced its “Code Orange: Fail Small” resilience programme following the November incident and another outage in December 2025. In a subsequent update, the company reported completing that programme, including work on controlling configuration rollouts. Cloudflare resilience programme · Programme completion update
The checks below are suggestions for reviewing your own environment, based on the recovery dependencies illustrated by the incident.
1. Test access with a fresh session
An engineer who is already logged in may have a very different experience from the next person joining the incident.
During an approved exercise, open a fresh browser session and follow the access path your on-call engineer would use. Include the identity provider, MFA, VPN or access gateway, and the cloud or infrastructure console.
Record where each step depends on an external service. Then work through a specific failure scenario. If the identity provider is unavailable, which consoles remain accessible? If a gateway fails, can the team still reach the systems needed to restore it?
Where you maintain emergency access accounts, test them under your access policy. Check that the credentials are retrievable, the permissions are sufficient, and the activity is logged. Give someone responsibility for reviewing access after the exercise.
2. Check whether your alerts can see a failed login
A basic health check might confirm that an application returns a successful HTTP response while users are unable to sign in.
Choose one important user journey and inspect what your monitoring actually tests. For a customer portal, that might include loading the login page, completing authentication, and opening an authenticated page.
Where synthetic login checks are appropriate, use a dedicated test account with limited permissions and protect its credentials. Also check how alerts reach the on-call team. An alert sent only through a tool that relies on the same unavailable identity service may be difficult to retrieve.
Write down what the team would observe in each scenario. “The homepage is healthy, but fresh logins fail” gives responders a more specific starting point than a generic availability alert.
3. Make the recovery instructions reachable
Runbooks often live in the same collaboration environment people use every day. Check how responders would retrieve them if that environment, its login service, or the corporate network became unavailable.
Keep an approved, access-controlled copy of the essential recovery information somewhere responders can reach during the scenario you are testing. Include the escalation contacts, the emergency access procedure, and the steps needed to locate the current runbook.
Review permissions and update ownership as part of the exercise. A reachable document still needs current instructions and a named person responsible for maintaining them.
One exercise for this week
Pick one production service and assume the team’s primary login path is unavailable.
Ask the on-call engineer to demonstrate how they would receive the alert, reach the recovery instructions, and access the controls needed to investigate. Use an approved test environment or a tabletop exercise where a live test would disrupt service.
Record the first step that depends on an unavailable component. Assign an owner and a follow-up date to that gap.
Have you tested fresh logins during a recovery exercise? Which dependency was easiest to overlook?
Author’s note: I work in marketing at Advange and cover enterprise IT news for infrastructure and operations teams. I used AI assistance to research and draft this post.