← All posts

FinOpsAid blog

GitHub Actions in October 2026 — retention now deletes your run history

In September, three Actions changes touched cost visibility without saying so. October shipped more, and the pattern held. None of these entries is labelled "billing", but two of them change what you can measure and one changes where your jobs run.

Here's the month, in the order we'd read it if spend tracking is your job.

Retention now covers workflow runs, checks and statuses

As of October 1, 2026, checks, workflow runs and statuses are governed by the same GitHub Actions retention setting that controls artifacts and logs (GitHub Changelog, October 2026). Records older than the retention period configured for your enterprise, organization or repository are cleaned up automatically.

Three details in the announcement matter more than the headline:

  • It covers third-party checks too. It applies to checks and statuses created by GitHub Actions and by third-party applications. Your CI isn't the only thing losing history.
  • Public repositories cap at 90 days. Repository retention stays subject to organization and enterprise caps on top of that.
  • It doesn't come back. "Changing the setting won't restore data that was previously removed." Lengthening retention after the fact recovers nothing.

The setting is now labelled "Check, workflow run, status, artifact and log retention" in the UI. That label is longer for a reason. One dial now controls two things that used to be separate decisions.

Here's why that's a cost question. Until now, retention was mostly a storage lever. You shortened it to stop paying for old artifacts and logs, and we've written about why storage shows up on the bill at all. Shortening retention now also deletes the run records themselves. Any report, script or tool that reconstructs Actions usage from GitHub's run history can only see back as far as that setting.

That includes spreadsheets built from the runs API, "minutes by workflow" dashboards, and year-over-year comparisons. Do you know what your org's retention is set to? Whatever it is, that's now the length of your CI history on GitHub.

Our finding: FinOpsAid's workflow-runs connector reads each run and its billable time from GitHub's /timing endpoint during the nightly sync, and stores the result in our own database. So run history we've already synced isn't affected when GitHub's retention cleanup later deletes the original records. A newly connected org is different. Its first backfill can only reach as far back as its retention setting allows, because that's all GitHub still holds. The trade-off we'd flag to any team: cutting retention to save on storage now also cuts the run history any tool, ours included, can rebuild from scratch. Decide the two together. If you want long CI history and short retention, something has to copy the records out before the cleanup runs.

Linux jobs may now run on AWS

On October 28, GitHub said Actions now schedules eligible jobs across a multi-cloud compute pool, so eligible Linux jobs may run on Amazon Web Services as well as Microsoft Azure (GitHub Changelog, October 2026). The stated goal is capacity: faster starts and shorter queue times during peak demand.

Scope, from the announcement:

Detail
Eligible Standard 2-core and 4-core Linux runners, and compatible 2-core and 4-core larger runners without specialized configurations
Not eligible Windows runners, and runners with private networking or other specialized configurations
Free, Pro, Team Automatic
Enterprise Cloud Opt-in: Enterprise settings → Actions → Hosted compute cloud providers → enable AWS compute
Workflow changes None; labels and syntax are unchanged

We didn't read any pricing change in the announcement, and it doesn't mention billing at all. We're not claiming there isn't one. We're saying this changelog doesn't describe one, so treat the per-minute rate as unchanged until GitHub says otherwise.

The real action item is network. Runners in the expanded pool use additional outbound IP addresses. GitHub warns that policies allowing only Azure or Microsoft-owned ranges can block connections or authentication from these runners, and it names Microsoft Entra Conditional Access policies specifically. Its recommendation is to sync allowlists with the actions array from GET https://api.github.com/meta rather than cloud-provider ranges.

There's a hidden cost here. A job that can't reach your artifact registry doesn't fail for free. It burns minutes on retries and timeouts first, and on a Free or Team plan nothing told you it moved.

macOS 14 is gone, and Xcode 27 is GA

The macOS 14 runner image was retired on November 2, 2026, after eight brownout windows through October (GitHub Changelog, October 2026). The retired labels are macos-14, macos-14-large and macos-14-xlarge. GitHub's suggested replacements are all arm64: macos-latest (now macos-26), macos-15, macos-latest-xlarge (macos-26-xlarge) and macos-15-xlarge.

Separately, the Xcode 27 image went GA on October 15 under xcode-27 and xcode-27-xlarge (GitHub Changelog, October 2026). It runs on Apple Silicon only, not Intel. GitHub's framing is that each image now tracks one major Xcode version rather than an OS version.

The September lesson applies directly. macOS minutes are the expensive ones on the Actions bill. A forced image change is exactly when a build's duration drifts, because toolchains and pre-installed software change underneath it. If you moved off macos-14 in a hurry during a brownout, benchmark the replacement now. Don't wait for next month's invoice to tell you.

What each October 2026 GitHub Actions change touches A grid with seven October 2026 GitHub Actions changes as rows and four columns: the bill, run history, network allowlists, and workflow files. Retention covering runs, checks and statuses touches run history directly and the bill indirectly through storage. Multi-cloud compute touches network allowlists directly and the bill only indirectly through failed or retried jobs; the changelog describes no price change. The macOS 14 retirement touches workflow files directly and the bill indirectly through duration drift on a new image. Xcode 27 touches workflow files. Workflow lockfiles touch workflow files. Actions Runner Controller 0.15.0 touches the bill indirectly through faster ephemeral runner cleanup on self-hosted clusters. New Meta API blob storage hosts touch network allowlists. The pattern: only one change touches run history, and it is the only one that cannot be undone. WHAT EACH OCTOBER CHANGE TOUCHES

BILL RUN HISTORY ALLOWLIST WORKFLOW FILES

Retention covers runs + checks

Multi-cloud compute (AWS)

macOS 14 retired

Xcode 27 image GA

Workflow lockfiles GA

ARC 0.15.0

Meta API blob storage hosts

direct indirect direct and irreversible: deleted history isn't restored
Our reading of seven GitHub Changelog entries, October 1–28, 2026, retrieved 2026-11-03. "Indirect" bill effects are our analysis (storage, retried jobs, duration drift, runner cleanup); none of these entries announces a price change.

Lockfiles, ARC 0.15.0 and new blob storage hosts

Three smaller changes complete the month.

Workflow lockfiles are GA on Enterprise Cloud (including data residency), Team and Free as of October 28 (GitHub Changelog, October 2026). The gh actions-lock CLI extension writes .github/workflows/actions.lock, covering actions and their transitive dependencies. Onboarded workflows run the exact commits recorded there, even if a tag moves. Dependabot regenerates the lockfile entry in the same pull request as a version bump. Admins can enforce adoption with a Require lockfile policy. GitHub is careful to say lockfiles protect against unexpected changes but "don't determine whether a dependency is safe."

Actions Runner Controller 0.15.0 is a self-hosted release (GitHub Changelog, October 2026). Patch upgrades now update resources in place, controller updates use patch requests, and concurrency and client rate limits are configurable. Ephemeral runners are also deleted faster when the pod exits successfully. If you run ARC, that last one is where cluster cost lives: capacity that lingers after a job is capacity you pay for. Our piece on whether self-hosted runners are cheaper covers the rest of that trade.

The Meta API lists new blob storage hosts for Actions on github.com (GitHub Changelog, October 2026). They are productionresultssa30 through productionresultssa49 and productionresultssa100 through productionresultssa199 on .blob.core.windows.net. If you already allow productionresultssa*.blob.core.windows.net, nothing changes. If you allowlist hosts one by one, job summaries and logs can break.

One more date worth knowing: pull_request_target is blocked by default in public repositories from December 2, with brownouts on November 9 and 16. We cover it separately in pull_request_target blocked by default.

What to do this month

  1. Read your retention setting before you change it. Check it at enterprise, organization and repository level. It now decides how much run history GitHub keeps, not just how much storage you pay for.
  2. Export what you need before shortening retention. Deleted runs and checks aren't restored when you lengthen the setting again.
  3. Sync allowlists to the Meta API. Use the actions array for runner IPs and the new blob hosts for logs. Review Entra Conditional Access policies that assume Azure-only ranges.
  4. On Enterprise Cloud, decide on AWS compute deliberately. It's opt-in for you. Flip it after your allowlists are ready, not before.
  5. Benchmark your macOS replacement jobs. Compare durations before and after the macos-14 move. The expensive minutes are where drift costs most.
  6. Pilot lockfiles on one high-traffic workflow. Pair it with Dependabot so lockfile updates arrive in the same pull request.
  7. If you run ARC, upgrade and watch runner idle time. Faster ephemeral cleanup only helps if you're measuring how long runners linger.

Where FinOpsAid fits

FinOpsAid is a read-only GitHub App with a nightly sync. Its workflow-runs connector reads each run and the billable time GitHub reports for it, then attributes minutes by repo, workflow and runner. As described above, history we've already synced stays in our database after GitHub's retention deletes it. A first-time backfill can't reach past your current retention, though. We can't see network allowlists, so the multi-cloud and blob-host changes are configuration work on your side. For more, see how to read the GitHub Actions page, how to read the runners page and Actions costs hiding in plain sight.

Frequently asked questions

Does GitHub Actions retention now delete workflow runs?

Yes. Since October 1, 2026, checks, workflow runs and statuses follow the same retention setting as artifacts and logs. That includes checks from third-party apps. Records older than the configured period are cleaned up automatically. Public repositories cap at 90 days. Lengthening the setting later doesn't restore anything already removed.

Will my GitHub Actions jobs run on AWS?

Possibly, if they use standard 2-core or 4-core Linux runners or compatible larger runners without specialized configurations. It's automatic on Free, Pro and Team. Enterprise Cloud customers opt in under Enterprise settings → Actions → Hosted compute cloud providers. Windows runners and private-networking runners aren't included.

What replaced the macOS 14 runner?

GitHub retired macos-14, macos-14-large and macos-14-xlarge on November 2, 2026. The suggested arm64 labels are macos-latest (macos-26), macos-15, macos-latest-xlarge and macos-15-xlarge. For pinning a toolchain, the Xcode 27 image (xcode-27, xcode-27-xlarge) is generally available on Apple Silicon only.

Sources

If you want your Actions run history kept somewhere retention can't reach, connect your GitHub org — read-only and free during beta — or explore the demo dashboard first.

github-actionsci-cdrunnerscost-optimizationgovernancefinops

See it in your own org

FinOpsAid meters the GitHub spend no invoice itemizes — read-only, and free during beta.