GitHub Actions hit by degraded availability as hosted runner delays stall CI pipelines

GitHub Actions degraded as runner assignment stalls

According to GitHub's status page, Actions — the company's workflow automation and continuous integration service — moved into a state of degraded availability on October 5, 2026, after roughly ninety minutes of escalating investigation into slow and failing job starts. The symptom described across the incident updates is a delay in assigning GitHub-hosted runners to queued jobs, which stretches out workflow start times and, in the later updates, was tied to outright job failures.

The disruption drew enough attention to reach the Hacker News front page shortly after 21:00 UTC, according to Hacker News — an early signal of how many developers were affected.

How the incident unfolded

The status page's update trail shows the picture sharpening over about an hour and a half:

• 19:11 UTC — GitHub opens an investigation into reports of degraded performance for Actions.
• 19:15 UTC — The company identifies delays in assigning GitHub-hosted runners to Actions jobs, warning that some workflows may take longer to start across runner configurations.
• 19:50 UTC — An update reports the runner-assignment delays persist and affect workflow start times across multiple runner configurations, with mitigation work under way.
• 20:39 UTC — GitHub says it is investigating job failures in addition to delays in runner assignment and workflow start times.
• 20:47 UTC — The incident is formally described as degraded availability, with the investigation continuing.

GitHub lists Actions as the only service affected by the incident.

What slow runner assignment does to pipelines

GitHub-hosted runners are the ephemeral compute that Actions provisions on demand for each job. When the assignment layer slows down, jobs do not necessarily fail straight away; they sit in a queue, and every downstream step in a pipeline waits with them. For repositories using matrix builds — the same workflow fanned out across many operating systems, language versions or configurations — a per-job assignment delay multiplies across the entire matrix, so a single scheduling bottleneck can hold up a release for a long time.

The job failures acknowledged at 20:39 UTC suggest that for at least some users the problem went beyond queuing slowness. The status updates do not quantify how many jobs, repositories or users were affected, and beyond a general statement that teams are working to mitigate the impact, no root cause or technical detail has been shared.

Coping while the incident is live

The status updates name GitHub-hosted runner assignment specifically, which is the layer of the service GitHub provisions itself. Teams that run self-hosted runners bypass that provisioning path, although the incident page does not break out whether self-hosted setups saw any impact. For repositories that depend on hosted runners, the options during a provider-side incident are limited: wait out the queue, retry failed jobs once availability recovers, and watch the GitHub status page for the next scheduled update.

Why it matters

GitHub Actions sits underneath a large share of modern software delivery, from open-source project CI to enterprise release pipelines. A degradation in one scheduling component — hosted runner assignment — is enough to slow or block builds across a huge number of repositories simultaneously, with nothing users can fix on their own side.

The incident is also a reminder of the concentration risk in CI. When one provider's runner fleet is the default for so many projects, its availability becomes a dependency for the wider software ecosystem. The pace of the escalation also stands out: it took roughly 96 minutes from the first report of degraded performance to the incident being classified as degraded availability with confirmed job failures. The question after recovery will be what broke in runner assignment at this scale, and whether GitHub can prevent a single scheduling bottleneck from stalling builds globally again.