GitHub's runner controller 0.15.0 is an October 1 release. InfoQ wrote it up on October 20.
GitHub dates runner controller 0.15.0 to October 1. The notes change how scale sets upgrade. InfoQ repeated them on October 20. The tag was not cut today.
The short version
GitHub's changelog, dated October 1, 2026, says release 0.15.0 updates runner scale sets in place, switches controller writes to patches, and makes listener rate limits and reconcile concurrency configurable. InfoQ's October 20 item restates that changelog.
The tag the changelog dates to October 1
GitHub's changelog entry for Actions Runner Controller 0.15.0 is dated October 1, 2026. It describes reliability, scalability, and observability changes for runner scale sets. The release page for gha-runner-scale-set-0.15.0 carries the same day. InfoQ's item, which the software wire carried on October 20, is a later account of that tag. The controller did not ship on the morning the wire repeated it. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.the software desk.
Upgrades that patch instead of replacing
The changelog says patch version upgrades now update resources in place, which reduces disruption between autoscaling runner sets and ephemeral runner sets. Controller updates use patch requests instead of full update requests, which the note says cuts the Kubernetes API payload. Runner status for EphemeralRunnerSet and AutoscalingRunnerSet moves to metrics, which the note says reduces status patch requests. Those are descriptions of requests the controller sends. They are not a count of requests a particular cluster saved. The document named in the sources is the record this page will keep.when one region is the control plane.
Knobs the note names, and one it does not
Listener rate limits are configurable through QPS and burst settings. Reconcile concurrency is configurable globally and per controller with max-concurrent-reconciles flags. terminationGracePeriodSeconds is configurable and, the note says, now lines up with the controller manager's graceful shutdown timeout. A scale set can be reregistered when the recorded set no longer exists in the Actions service. The changelog does not publish a default QPS, and this page will not invent one. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.what a license obligates.
What the October 20 retelling is for
InfoQ's October 20 headline says 0.15.0 targets reliability at scale. That sentence tracks the changelog. It does not add a customer count, a failure rate, or a region. A reader who wants the behavior reads the October 1 note and the release, then treats October 20 as the day a second outlet repeated it. Asked directly, Which knobs does the changelog say an operator can set? Listener Kubernetes client rate limits, through QPS and burst settings, and controller concurrency, globally and per controller, through max-concurrent-reconciles flags. Shutdown grace is also configurable.Linux 7.3 without the rc suffix.
Ephemeral runners leave faster, on a condition
The changelog says ephemeral runners are deleted faster because the server-side check for runner removal is skipped when the runner pod exits successfully. The condition is the clean exit. A pod that does not exit cleanly is outside that sentence. This page will not turn the sentence into a promise that every finished job disappears on a timer. Asked directly, Does 0.15.0 move runners off Kubernetes? No. The changelog is about a controller that already runs on Kubernetes. Smaller API payloads are a traffic change, not a new place for the jobs to live.
What this page will not import
It will not call 0.15.0 a GitHub Actions price change. It will not say self-hosted runners became GitHub-hosted. It will not treat an experimental chart, which the release page separates from the supported charts, as the version an operator should install. The supported artifacts the release names are the controller image and the two helm charts. Asked directly, Does InfoQ add a measurement? No. The October 20 item is an account of the October 1 notes. It is not a second cluster's benchmark.
What a later note would change
A changelog correction that moves the date, or a release that yanks 0.15.0, would change the first sentence. A measured API-call reduction from a cluster that publishes its method would be a different brief. Until then the record is the October 1 notes and an October 20 retelling. Asked directly, When did GitHub publish 0.15.0? The changelog entry is dated October 1, 2026. The GitHub release page shows the same day for gha-runner-scale-set-0.15.0. InfoQ's writeup is October 20. The document named in the sources is the record this page will keep.Futureweb.
The cluster is still the operator's
Scale sets, listeners, and ephemeral pods are objects in a Kubernetes cluster someone else runs. A smaller patch payload does not move that cluster into GitHub's network. The Actions service is still the place a job is registered. The controller is still the process that asks the cluster for pods. Asked directly, What does an in-place patch upgrade change for a running scale set? The changelog says patch version upgrades now update resources in place, so the mapping between an autoscaling runner set and an ephemeral runner set is less disruptive. It does not say a running job is paused.
What stays on the record
October 1 for the tag. In-place patch upgrades, patch requests instead of full updates, status by metrics, configurable QPS and burst, configurable reconcile concurrency, configurable shutdown grace, reregistration of a missing scale set, and faster deletion after a clean pod exit. October 20 for InfoQ's account. The short answer this page will stand on is this: GitHub's changelog, dated October 1, 2026, says release 0.15.0 updates runner scale sets in place, switches controller writes to patches, and makes listener rate limits and reconcile concurrency configurable. InfoQ's October 20 item restates that changelog.
Sources
The reports this brief is filing. Futureweb did not republish them.
Questions
When did GitHub publish 0.15.0?
The changelog entry is dated October 1, 2026. The GitHub release page shows the same day for gha-runner-scale-set-0.15.0. InfoQ's writeup is October 20.
What does an in-place patch upgrade change for a running scale set?
The changelog says patch version upgrades now update resources in place, so the mapping between an autoscaling runner set and an ephemeral runner set is less disruptive. It does not say a running job is paused.
Which knobs does the changelog say an operator can set?
Listener Kubernetes client rate limits, through QPS and burst settings, and controller concurrency, globally and per controller, through max-concurrent-reconciles flags. Shutdown grace is also configurable.
Does InfoQ add a measurement?
No. The October 20 item is an account of the October 1 notes. It is not a second cluster's benchmark.
Does 0.15.0 move runners off Kubernetes?
No. The changelog is about a controller that already runs on Kubernetes. Smaller API payloads are a traffic change, not a new place for the jobs to live.