· via dev.to (home feed)
GitHub outage that paused Actions and Copilot renews case for break-glass deploys
A roughly 33-minute GitHub incident on November 11, 2026 disrupted Actions, Pages and Copilot, briefly halting deploys. A dev.to essay argues CI pipelines are production dependencies that need tested manual fallbacks.

A short outage with a sharp edge
According to a dev.to post by platform engineer Nazmul Ahasan, GitHub opened a status-page incident at 20:48 UTC on November 11, 2026, described as affecting multiple GitHub products including Actions, Pages and Copilot. Actions, Pages and Copilot were flagged as a major outage, with webhooks marked degraded. GitHub reported mitigation at 21:07 UTC and full resolution at 21:21 UTC, once the backlog of queued Actions jobs had drained.
That is roughly 33 minutes — but for any team whose only route to production ran through a GitHub Actions workflow, hotfixes, rollbacks and configuration changes were all off the table until service returned.
One hosted service in the critical path
Ahasan's core argument is that teams invest heavily in redundant databases and load balancers, then funnel every production change through a single hosted service, a single webhook system, and credentials issued by that same service. He lists the links in a typical hosted pipeline's chain:
- the source-control host being reachable
- webhooks delivering the trigger
- hosted runners accepting jobs
- CI-issued identity, such as OIDC tokens, used to reach the cloud
- artifacts and container images fetched from registries mid-job
Any one of these failing halts deploys. Most days you can simply wait; the dangerous day is the one with an active incident, a bad release needing rollback, or a certificate about to expire.
Anatomy of a break-glass deploy
Ahasan defines the fallback as a documented, tested and audited way to push a known-good artifact to production when the normal pipeline is unavailable — not a back door, but a deliberately slower and more controlled path where every use is reviewed afterwards. He calls for four components.
A runner under your own control: a self-hosted runner or secondary CI system able to execute the same deploy step. His caveat is important — self-hosted runners registered with GitHub still depend on GitHub dispatching the job, so they cover hosted-runner capacity problems but not a full Actions outage. Genuine independence requires a deploy that can be started from a hardened bastion or a second CI system with no dependency on the first.
Artifacts that already exist: the fallback ships something already built and scanned, not something compiled on a laptop. Release artifacts live in storage you own, and deployed images are mirrored into your own registry, so a deploy never pulls from a public registry or the CI provider's cache.
Credentials that survive a CI outage: OIDC federation between Actions and a cloud account is a good default because it eliminates long-lived keys, but it only works while Actions is minting tokens. Ahasan recommends a separate, narrowly scoped break-glass role limited to deploying, requiring phishing-resistant MFA and approval from a second person, with short sessions and an alert fired on every assumption.
A runbook people have actually executed: an untested runbook, he writes, is little more than a guess.
He also sketches a sample procedure: open an incident and record who approved the fallback; assume the break-glass role with MFA and a 900-second session; point the deployment at an already-scanned image pinned by digest rather than tag; verify the rollout and log what changed in the incident ticket. Pinning by digest removes any ambiguity about which "latest" was shipped under pressure.
Drill it, then review it
The accompanying checklist adds operational hygiene. Store the manual runbook somewhere independent of your source host, with named owners and an approval rule. Keep cached artifacts and mirrored images referenced by digest. Run scheduled drills — quarterly, for example, first in staging and then in production with a no-op change — and time them, fix whatever breaks and fold the fixes back into the runbook. Finally, review every real use of the path and rotate the credentials afterwards where policy requires it.
Why it matters
This particular incident ended in well under an hour, and GitHub cleared the queued jobs afterwards. The next one may last longer, or sit at a different provider entirely. The broader lesson is one of dependency analysis: as teams consolidate on hosted CI and OIDC-based identity, the delivery pipeline becomes a single point of failure for change even while production itself is healthy. Applying the same failover thinking normally reserved for databases — redundancy, independent credentials, rehearsed recovery — turns a CI/CD outage from a shipping blocker into a manageable inconvenience.
- #github-actions
- #ci-cd
- #devops
- #incident-response
- #reliability