Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
343 changes: 318 additions & 25 deletions .github/workflows/candidate.yml

Large diffs are not rendered by default.

6 changes: 6 additions & 0 deletions docs/dress-rehearsal.md
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,12 @@ previous candidate, and rollback result/time. Historical attempts are retained
root-only beside it; detailed install logs are root-only under
`/var/log/1helm-candidate`.

Phase 4 retains this private status as the Linux dress-rehearsal proof and also
runs an ephemeral hosted Linux acceptance lane for clean install, distinct
prior Stable to candidate update, state identity, systemd health, and a forced
failure rollback. The Phase 3 bundle requires both records. Phase 4 does not
change or repurpose this guest or its unique runner.

Integrate this read-only evidence with Phase 0 without tracking private
coordinates:

Expand Down
138 changes: 138 additions & 0 deletions docs/phase4-platform-acceptance.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,138 @@
# Phase 4 cross-platform candidate acceptance

1. Build the ready-to-run Linux TGZ on the hosted builder and retain its
GitHub-hosted provenance attestation.
2. Build the Apple Silicon DMG and updater ZIP on the dedicated Mac runner;
Developer ID-sign, notarize, staple, and Gatekeeper-check both exact
payloads. Missing Apple credentials or runner capacity blocks the matrix.
3. Run Linux, macOS, and Windows 11 acceptance concurrently after their exact
bytes exist. Each job rechecks repository, workflow, push event, main ref,
candidate SHA, CI run, and candidate run before repository code executes.
4. Retain normalized JSON with exact artifact SHA-256 and byte counts, candidate
and CI run identities, machine/runner identity, check timestamps, state
digests, and rollback or scoped-uninstall outcome. Windows binds behavior to
the Linux TGZ and publishes no artifact or signing claim.
5. Assemble Phase 3's promotion bundle only after all builds, the private Linux
dress rehearsal, and all three acceptance records pass. The assembler copies
retained bytes and cannot build. Missing, failed, skipped, or unavailable
lanes remain blockers in the retained candidate matrix status.

## One-time dedicated runners

Register repository-scoped runners with `--no-default-labels` and exactly one
label: `1helm-macos-phase4` or `1helm-windows-phase4`. Do not put Proxmox,
Stable, website, production, or broad repository-write credentials on either
machine. The runner account is a dedicated ordinary user with no production
data, no broad `sudo`/administrator membership, and no other workloads.

Install the root-owned `ops/platform-acceptance/runner-job-started.sh` on macOS
or the administrator-owned native PowerShell
`ops/platform-acceptance/runner-job-started.ps1` on Windows as
`ACTIONS_RUNNER_HOOK_JOB_STARTED`. The hook accepts only `gitcommit90/1Helm`, the Candidate dress
rehearsal workflow, its allowlisted job, a successful CI `push` to trusted
`main`, and the exact SHA. Do not set the enabling variables until the label,
hook, account isolation, and prerequisites below have been inspected:

- repository variable `HELM_PHASE4_MACOS_ENABLED=1` enables the Mac build and
acceptance jobs;
- repository variable `HELM_PHASE4_WINDOWS_ENABLED=1` enables Windows
acceptance.

Leaving either variable absent is the safe default: the lane is skipped by
GitHub but recorded as **blocked**, and no complete promotion artifact exists.
Registration tokens are one-use and transient; never save one in an image,
shell profile, workflow, or repository file.

### macOS runner and credentials

Use an Apple Silicon Mac retained solely for release acceptance. Install the
repository runner, Node 22, Xcode command-line tools, and the pinned npm
dependencies. In the dedicated account's login keychain install:

- one Developer ID Application certificate/private key;
- `APPLE_TEAM_ID` and optionally `APPLE_SIGN_IDENTITY` as runner service
environment values;
- an `APPLE_NOTARY_PROFILE` created with `xcrun notarytool store-credentials`.

The workflow intentionally does not use GitHub repository secrets for signing:
the secure credential and private key remain on the retained Mac. Configure the
runner service to unlock only that dedicated user's keychain before a job and
lock it in a job-completed hook; do not store the keychain password in the
repository, runner labels, workflow environment, or shell history. The account's
`~/Library/Application Support/1Helm-OCI-v1` must be absent before every job;
residue is a hard failure. The job validates signature, notarization ticket,
staple, Gatekeeper, exact versions/digests, clean launch/loopback health, updater
ZIP replacement from the latest digest-qualified prior Stable DMG, and
Application Support identity. Signing requirements are never downgraded to
ad-hoc signing.

### Windows 11 runner and reboot resume

Use a dedicated Windows 11 x64 VM/account with no production data. The runner
is non-elevated. The one-time operator proof starts from both WSL optional
features disabled, invokes the tracked installer from an ordinary PowerShell
window, records the single UAC approval, expected restart, and same-user resume,
then stores an administrator-owned, user-readable
`C:\ProgramData\1Helm-Phase4\provisioning-evidence.json`:

```json
{
"schema": 1,
"kind": "1helm-windows-runner-provisioning",
"features_initially_disabled": true,
"single_uac": true,
"restart_required": true,
"same_user_resume": true,
"keepalive_after_sign_in": true,
"service_active": true,
"localhost_health": true,
"snapshot_baseline": true,
"dedicated": true,
"production_data": false
}
```

After that real provisioning exercise, use the scoped uninstaller, confirm the
named product distribution and install root are absent, and capture the clean,
WSL-enabled dedicated-VM snapshot used before every acceptance job. The runner
registration, hook, pinned rootfs, and the administrator-owned provisioning
record remain; 1Helm product state does not.

Also stage the pinned Ubuntu rootfs at
`C:\ProgramData\1Helm-Phase4\ubuntu-noble-wsl-amd64.rootfs.tar.gz`.

The workflow does not claim an in-job Windows reboot: a self-hosted job cannot
honestly survive one. Instead, it records the allowed snapshot-assisted
equivalent. The provisioner restores the accepted clean snapshot before the
job routes; the exact candidate job then stops the same-user limited logon
keepalive, cold-stops only its named WSL distribution, restarts that keepalive,
and requires systemd, localhost health, and retained state to recover. The JSON
record names this mode `snapshot-assisted-equivalent`; it never labels the
exercise as a real reboot. The administrator-owned record separately preserves
the one-time real disabled-features/UAC/restart/same-user-resume proof.

Windows uses the repository's tracked `install.ps1`, `install.sh`, keepalive,
and `uninstall.ps1` with documented local candidate inputs, so it is a private
site-equivalent path without touching the public website. It creates an
unrelated WSL control distribution and proves uninstall retains it. It never
claims Windows artifact creation or signing.

## Recovery and teardown

First disable the applicable repository variable so new jobs cannot route.
Wait for any active job to finish, archive only the retained JSON and exact
candidate artifacts needed by the owner, then remove the repo runner
registration and stop its service. Remove only the dedicated account/VM after
confirming its identity; never use a wildcard or a shared host.

Mac cleanup quits the test app and removes transient app copies and Application
Support after success; the dedicated Application Support directory is retained
on failure for diagnosis and must be removed before re-enabling the runner. Windows cleanup
uses the scoped product uninstaller for the exact `1helm-phase4` distribution
and separately unregisters only its named unrelated test control. If teardown
is interrupted, leave the runner disabled and inspect both exact names before
retrying. Linux hosted runners are ephemeral; the Phase 2 private runner and

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Hyphenate the compound modifier.

Change Linux hosted runners to Linux-hosted runners at Line 134.

🧰 Tools
🪛 LanguageTool

[grammar] ~134-~134: Use a hyphen to join words.
Context: ... both exact names before retrying. Linux hosted runners are ephemeral; the Phase ...

(QB_NEW_EN_HYPHEN)

🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/phase4-platform-acceptance.md` at line 134, Update the “Linux hosted
runners” phrase in the Phase 4 platform acceptance documentation to
“Linux-hosted runners,” preserving the surrounding text.

Source: Linters/SAST tools

its retained evidence are not teardown targets for Phase 4.

No Phase 4 automation publishes Stable, creates a tag/Release, deploys the
website, or changes production data/services.
10 changes: 8 additions & 2 deletions docs/release-lifecycle.md
Original file line number Diff line number Diff line change
Expand Up @@ -156,8 +156,14 @@ current `main`; successful
CI and candidate workflow identities; Linux attestation, archive digest, and
embedded commit; private dress-rehearsal health for that digest; all three
retained artifact records; and retained macOS, Linux, and Windows acceptance.
Phase 3 does not manufacture platform records: until Phase 4 supplies them, the
dry run reports them as blockers and publication remains paused.
The trusted-main candidate workflow now retains the complete Phase 4 matrix:
hosted Linux candidate and acceptance, signed/notarized Mac bytes and Apple
Silicon acceptance, and Windows 11 behavior acceptance bound to the Linux TGZ.
It assembles the Phase 3 promotion bundle only when every exact byte and record
passes. Missing credentials, disabled/offline runners, failed checks, or absent
evidence appear in the retained per-platform status as blockers; they never
become skipped success. Provisioning and recovery are documented in
[phase4-platform-acceptance.md](./phase4-platform-acceptance.md).

Publication additionally requires `mode=publish`, the exact identity-bound
confirmation printed by the dry run, and owner approval in the protected
Expand Down
113 changes: 113 additions & 0 deletions ops/platform-acceptance/linux.sh
Original file line number Diff line number Diff line change
@@ -0,0 +1,113 @@
#!/usr/bin/env bash
set -euo pipefail

ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
ARCHIVE="${HELM_CANDIDATE_ARCHIVE:?exact Linux candidate archive is required}"
MANIFEST="${HELM_CANDIDATE_MANIFEST:?exact candidate manifest is required}"
PROVENANCE="${HELM_CANDIDATE_PROVENANCE:?exact hosted provenance bundle is required}"
OUTPUT="${HELM_ACCEPTANCE_OUTPUT:?acceptance output is required}"
STARTED_AT="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
VERSION="$(node -p 'JSON.parse(require("fs").readFileSync(process.argv[1])).version' "$MANIFEST")"
DIGEST="$(node -p 'JSON.parse(require("fs").readFileSync(process.argv[1])).artifact.sha256' "$MANIFEST")"
COMMIT="$(node -p 'JSON.parse(require("fs").readFileSync(process.argv[1])).source.commit' "$MANIFEST")"
CI_RUN="$(node -p 'JSON.parse(require("fs").readFileSync(process.argv[1])).ci.run_id' "$MANIFEST")"

[[ "$GITHUB_REPOSITORY" == "gitcommit90/1Helm" && "$GITHUB_EVENT_NAME" == "workflow_run" \
&& "$GITHUB_REF" == "refs/heads/main" && "$GITHUB_SHA" == "$COMMIT" \
&& "${HELM_EXPECTED_COMMIT:?}" == "$COMMIT" && "${HELM_EXPECTED_CI_RUN_ID:?}" == "$CI_RUN" ]] \
|| { echo "Linux acceptance refused an untrusted candidate identity." >&2; exit 1; }
export HELM_ACCEPTANCE_PLATFORM=linux
export HELM_ACCEPTANCE_STARTED_AT="$STARTED_AT"
export HELM_PHASE4_RUNNER_LABEL=ubuntu-latest
node "$ROOT/scripts/pending-acceptance-evidence.mjs"
[[ "$(id -u)" -ne 0 ]] || { echo "Linux acceptance must begin as the hosted ordinary runner user." >&2; exit 1; }
[[ "$(sha256sum "$ARCHIVE" | awk '{print $1}')" == "$DIGEST" ]] \
|| { echo "Linux candidate digest mismatch." >&2; exit 1; }
node -e 'import("./scripts/candidate-manifest.mjs").then(({candidateIdentityFromArchive})=>{const x=candidateIdentityFromArchive(process.argv[1]); if(x.commit!==process.argv[2]) process.exit(2)})' "$ARCHIVE" "$COMMIT"
gh attestation verify "$ARCHIVE" --bundle "$PROVENANCE" \
--repo gitcommit90/1Helm --signer-workflow gitcommit90/1Helm/.github/workflows/candidate.yml \
--source-ref refs/heads/main --source-digest "$COMMIT" --deny-self-hosted-runners

work="$(mktemp -d)"
trap 'rm -rf -- "$work"' EXIT
prefix="$(tar -tzf "$ARCHIVE" | awk -F/ '/^[^/]+\/site\/public\/install\.sh$/ && !found { print $1; found=1 }')"
[[ -n "$prefix" ]] || { echo "Candidate installer is missing." >&2; exit 1; }
tar -xzf "$ARCHIVE" -C "$work" "$prefix/site/public/install.sh"

# The hosted VM is disposable and contains no user or production data. Its
# first installation is therefore an actual clean systemd installation.
sudo env HELM_RELEASE_SHA256="$DIGEST" bash "$work/$prefix/site/public/install.sh" "$ARCHIVE"
sudo systemctl is-active --quiet 1helm.service
curl -fsS http://127.0.0.1:8123/api/setup/status >"$work/clean-health.json"
[[ "$(readlink -f /opt/1helm/current)" == "/opt/1helm/releases/$VERSION-$DIGEST" ]]

# Resolve the newest immutable public Stable release distinct from this
# candidate version. Candidate versions normally remain unchanged during
# preview phases, so `releases/latest` can legitimately be the same version.
gh api "repos/$GITHUB_REPOSITORY/releases?per_page=20" >"$work/releases.json"
node - "$work/releases.json" "$VERSION" >"$work/previous-release.json" <<'NODE'
const fs = require("fs");
const releases = JSON.parse(fs.readFileSync(process.argv[2], "utf8"));
const current = process.argv[3];
const release = releases.find((item) => !item.draft && !item.prerelease
&& /^v\d+\.\d+\.\d+$/.test(String(item.tag_name || ""))
&& String(item.tag_name).slice(1) !== current);
if (!release) process.exit(2);
process.stdout.write(JSON.stringify(release));
NODE
PREVIOUS_VERSION="$(node -p 'String(JSON.parse(require("fs").readFileSync(process.argv[1])).tag_name||"").replace(/^v/,"")' "$work/previous-release.json")"
[[ "$PREVIOUS_VERSION" =~ ^[0-9]+\.[0-9]+\.[0-9]+$ && "$PREVIOUS_VERSION" != "$VERSION" ]] \
|| { echo "A distinct previous Stable version is required for Linux updater acceptance." >&2; exit 1; }
PREVIOUS_NAME="1Helm-$PREVIOUS_VERSION-linux-node.tgz"
readarray -t previous_asset < <(node - "$work/previous-release.json" "$PREVIOUS_NAME" <<'NODE'
const fs = require("fs");
const release = JSON.parse(fs.readFileSync(process.argv[2], "utf8"));
const asset = (release.assets || []).find((item) => item.name === process.argv[3]);
if (!asset || !/^sha256:[a-f0-9]{64}$/.test(String(asset.digest || ""))) process.exit(2);
console.log(asset.browser_download_url);
console.log(asset.digest.slice(7));
NODE
)
[[ "${#previous_asset[@]}" -eq 2 ]] || { echo "Prior Stable Linux asset lacks digest-qualified metadata." >&2; exit 1; }
PREVIOUS_URL="${previous_asset[0]}"
curl -fsSL --proto '=https' --tlsv1.2 --retry 3 -o "$work/$PREVIOUS_NAME" "$PREVIOUS_URL"
PREVIOUS_DIGEST="$(sha256sum "$work/$PREVIOUS_NAME" | awk '{print $1}')"
[[ "$PREVIOUS_DIGEST" == "${previous_asset[1]}" ]] || { echo "Prior Stable Linux digest mismatch." >&2; exit 1; }
previous_prefix="$(tar -tzf "$work/$PREVIOUS_NAME" | awk -F/ '/^[^/]+\/site\/public\/install\.sh$/ && !found { print $1; found=1 }')"
tar -xzf "$work/$PREVIOUS_NAME" -C "$work" "$previous_prefix/site/public/install.sh"
sudo env HELM_RELEASE_SHA256="$PREVIOUS_DIGEST" bash "$work/$previous_prefix/site/public/install.sh" "$work/$PREVIOUS_NAME"
[[ "$(node -p 'require("/opt/1helm/current/package.json").version')" == "$PREVIOUS_VERSION" ]]

MARKER=/var/lib/1helm-oci-v1/phase4-acceptance-state
openssl rand -hex 32 | sudo tee "$MARKER" >/dev/null
sudo chown 1helm:1helm "$MARKER"
STATE_BEFORE="$(sudo sha256sum "$MARKER" | awk '{print $1}')"
CANDIDATE_RELEASE="/opt/1helm/releases/$VERSION-$DIGEST"
[[ -d "$CANDIDATE_RELEASE" ]] || { echo "Clean candidate release was not retained for updater acceptance." >&2; exit 1; }
sudo "$CANDIDATE_RELEASE/site/public/apply-linux-release.sh" "$CANDIDATE_RELEASE" "$VERSION"
[[ "$(node -p 'require("/opt/1helm/current/package.json").version')" == "$VERSION" ]]
sudo systemctl is-active --quiet 1helm.service
curl -fsS http://127.0.0.1:8123/api/setup/status >"$work/update-health.json"

# Exercise the candidate's real atomic host transaction with a derived local
# startup-failure fixture. The fixture is never uploaded as candidate bytes.
[[ "$(readlink -f /opt/1helm/current)" == "$CANDIDATE_RELEASE" ]]
FAILURE_RELEASE="/opt/1helm/releases/$VERSION-$DIGEST-phase4-failure"
sudo cp -a "$CANDIDATE_RELEASE" "$FAILURE_RELEASE"
printf '%s\n' 'throw new Error("Phase 4 controlled startup failure");' | sudo tee "$FAILURE_RELEASE/src/server/index.ts" >/dev/null
if sudo "$CANDIDATE_RELEASE/site/public/apply-linux-release.sh" "$FAILURE_RELEASE" "$VERSION"; then
echo "Controlled failure unexpectedly passed." >&2
exit 1
fi
[[ "$(readlink -f /opt/1helm/current)" == "$CANDIDATE_RELEASE" ]]
sudo systemctl is-active --quiet 1helm.service
curl -fsS http://127.0.0.1:8123/api/setup/status >"$work/rollback-health.json"
STATE_AFTER="$(sudo sha256sum "$MARKER" | awk '{print $1}')"
[[ "$STATE_BEFORE" == "$STATE_AFTER" ]]
sudo rm -rf -- "$FAILURE_RELEASE"

export HELM_PREVIOUS_VERSION="$PREVIOUS_VERSION"
export HELM_STATE_BEFORE_SHA256="$STATE_BEFORE"
export HELM_STATE_AFTER_SHA256="$STATE_AFTER"
export HELM_MACHINE_OS_VERSION="$(. /etc/os-release; printf '%s' "$PRETTY_NAME")"
node "$ROOT/scripts/linux-acceptance-evidence.mjs"
Loading
Loading