orca/.github/workflows/daily-mac-build.yml

472 lines
23 KiB
YAML

name: Daily macOS Dev Build
# Why: gives developers a signed macOS build of main once a day that the in-app
# updater can install directly, without waiting for an RC cut — less noise than
# hourly.
#
# Schedule is a single UTC cron (GH Actions has no timezone-aware schedules).
# 14:15 UTC is early morning Pacific year-round (6:15am PST / 7:15am PDT). Minute
# 15 avoids stacking with hourly-mac-build, which fires at minute 0 every hour.
#
# Deliberately narrow scope:
# - macOS only. Other platforms keep using RC/stable.
# - No tests, no lint, no e2e. This channel trades safety for latency; PR CI
# and release-cut remain the gates that matter.
# - Signed AND notarized, exactly like a release. The notary round trip is the
# one slow step kept: macOS anchors a notarized app's TCC grants on identifier
# + team rather than on its cdhash, so those grants survive an update. Without
# a ticket every daily reads as a new client and silently loses file access
# under Documents/Desktop/Downloads.
#
# Artifacts publish to stablyai/orca-daily, never to stablyai/orca: the main
# repo's releases atom feed exposes only its 10 newest entries, so high-volume
# dev tags would evict every stable/RC entry and break updates for real users.
# Separate from orca-hourly so someone riding main's hourlies never sees the
# sparse daily series mixed into that list.
#
# GITHUB_TOKEN is scoped to this repo and cannot publish there, so writes use the
# same GitHub App as hourly (installed on orca-daily with Contents: Read and
# write). Provision the App secrets with
# `bash config/scripts/setup-hourly-release-token.sh`, then grant the App access
# to orca-daily with `bash config/scripts/setup-daily-release-repo.sh`:
# HOURLY_RELEASE_APP_ID the App's numeric id (shared with hourly/adhoc)
# HOURLY_RELEASE_APP_PRIVATE_KEY the App's .pem private key
#
# Installation tokens live one hour, which is why this mints three times. Install
# and build need no token at all, and notarization can hold the publish step for
# tens of minutes; minting again once the build is done starts the clock at the
# first call that actually uses it rather than burning a third of it on
# `pnpm install`. The upload step's own retry budget (2x45m) can outlive that
# second token, so a third is minted after it for verify/publish/prune/cleanup —
# without it a stuck notary run would strand a draft that cleanup gets a 401 on.
# The release stays an unpublished draft until the manifest check passes, so the
# worst case is still an invisible draft — and the build-number query counts
# drafts, so it holds its number and the next run does not reuse it.
on:
schedule:
# Once a day, early morning Pacific. Single cron — no DST twin, no clock gate.
- cron: '15 14 * * *'
workflow_dispatch:
inputs:
force:
description: Build even if main has not moved since the last daily
required: false
default: false
type: boolean
permissions:
contents: read
concurrency:
group: daily-mac-build
cancel-in-progress: false
env:
DAILY_REPO: stablyai/orca-daily
# Keep ~30 days so a regression can be bisected without holding a full year.
DAILY_RETAIN_COUNT: 30
jobs:
build-daily-mac:
if: github.repository == 'stablyai/orca'
runs-on: blacksmith-6vcpu-macos-15
# Why 150: it must exceed the worst case the retry budgets below can produce
# (install 3x10 + publish 2x45 = 120, plus ~25 for checkout/build/verify), or
# the job is killed mid-retry and no cleanup step runs at all. A typical run
# is far shorter — this is the notary queue's tail, not its median.
timeout-minutes: 150
env:
NODE_OPTIONS: --max-old-space-size=4096
steps:
- name: Checkout
uses: actions/checkout@v6
with:
ref: main
fetch-depth: 0
# Why: this job only reads stablyai/orca and never pushes; every write
# goes to the daily repo through a minted App token passed by env.
# Not persisting the checkout credential shrinks the blast radius if a
# build step is compromised (zizmor: artipacked).
persist-credentials: false
- name: Mint daily repo token
id: app_token
uses: actions/create-github-app-token@v2
with:
app-id: ${{ secrets.HOURLY_RELEASE_APP_ID }}
private-key: ${{ secrets.HOURLY_RELEASE_APP_PRIVATE_KEY }}
owner: stablyai
repositories: orca-daily
# Why: main is often idle. Rebuilding an unchanged commit burns a runner
# hour and adds a redundant tag to the retention window.
- name: Check whether main moved since the last daily
id: freshness
shell: bash
env:
GH_TOKEN: ${{ steps.app_token.outputs.token }}
FORCED: ${{ github.event_name == 'workflow_dispatch' && inputs.force }}
run: |
set -euo pipefail
head_sha="$(git rev-parse HEAD)"
echo "head_sha=$head_sha" >>"$GITHUB_OUTPUT"
if [[ "$FORCED" == "true" ]]; then
echo "should_build=true" >>"$GITHUB_OUTPUT"
echo "Forced dispatch; building $head_sha."
exit 0
fi
# The previous daily records its source commit in the release body.
# Drafts are excluded: an unpublished leftover never shipped, so treating
# it as "the last build" would skip a build that never actually happened.
last_body="$(gh release list --repo "$DAILY_REPO" --limit 20 --json tagName,isDraft \
--jq 'map(select(.isDraft | not)) | .[0].tagName // empty' 2>/dev/null || true)"
if [[ -z "$last_body" ]]; then
echo "should_build=true" >>"$GITHUB_OUTPUT"
echo "No prior daily release found; building $head_sha."
exit 0
fi
last_sha="$(gh release view "$last_body" --repo "$DAILY_REPO" --json body \
--jq '.body | capture("commit `(?<sha>[0-9a-f]{7,40})`") | .sha' 2>/dev/null || true)"
if [[ -n "$last_sha" && "$head_sha" == "$last_sha"* ]]; then
echo "should_build=false" >>"$GITHUB_OUTPUT"
echo "main is unchanged since $last_body ($last_sha); skipping."
else
echo "should_build=true" >>"$GITHUB_OUTPUT"
echo "main moved to $head_sha (last daily built $last_sha); building."
fi
- name: Setup pnpm
if: steps.freshness.outputs.should_build == 'true'
uses: pnpm/action-setup@v6
with:
run_install: false
- name: Setup Node.js
if: steps.freshness.outputs.should_build == 'true'
uses: actions/setup-node@v6
with:
node-version-file: package.json
cache: pnpm
- name: Cache electron-builder downloads
if: steps.freshness.outputs.should_build == 'true'
uses: actions/cache@v5
with:
path: |
~/Library/Caches/electron
~/Library/Caches/electron-builder
key: electron-builder-mac-${{ hashFiles('pnpm-lock.yaml') }}
restore-keys: |
electron-builder-mac-
- name: Install dependencies
if: steps.freshness.outputs.should_build == 'true'
uses: nick-fields/retry@v4
with:
timeout_minutes: 10
max_attempts: 3
retry_wait_seconds: 30
command: pnpm install --frozen-lockfile
# Why: signing is what makes a daily installable over an existing Orca, so
# a missing cert must fail here rather than after a 20-minute build.
- name: Verify macOS signing environment
if: steps.freshness.outputs.should_build == 'true'
run: node config/scripts/verify-macos-release-env.mjs
env:
CSC_LINK: ${{ secrets.MAC_CERTS }}
CSC_KEY_PASSWORD: ${{ secrets.MAC_CERTS_PASSWORD }}
APPLE_ID: ${{ secrets.APPLE_ID }}
APPLE_APP_SPECIFIC_PASSWORD: ${{ secrets.APPLE_APP_SPECIFIC_PASSWORD }}
APPLE_TEAM_ID: ${{ secrets.APPLE_TEAM_ID }}
- name: Compute daily version
id: daily
if: steps.freshness.outputs.should_build == 'true'
shell: bash
env:
GH_TOKEN: ${{ steps.app_token.outputs.token }}
MAIN_REPO_TOKEN: ${{ github.token }}
run: |
set -euo pipefail
# Existing titles, which carry the build number this series continues
# from. The script picks the number, because it restarts per base version
# and only the script knows which base this build resolved to.
#
# Why drafts count here but not in the freshness check: that check asks
# "did this commit ship", where a draft is a no. This one asks "is the
# number free", where a stranded draft still holds one.
names="$(gh release list --repo "$DAILY_REPO" --limit 200 --json name \
--jq '.[].name // empty')"
# Why the main repo's tags decide the base version rather than
# package.json: main's version only moves on `release:` commits, and
# stable patches are cut from release branches that never merge back, so
# package.json can sit several patches behind what users are running. A
# separate token because GH_TOKEN above is the App's, scoped to the
# daily repo. Empty on failure — the script then falls back to
# package.json, which is stale but never wrong enough to fail a build.
published="$(GH_TOKEN="$MAIN_REPO_TOKEN" gh release list \
--repo "$GITHUB_REPOSITORY" --limit 100 --exclude-drafts \
--json tagName --jq '.[].tagName' || true)"
echo "Highest published tag seen: $(head -1 <<<"$published")"
ORCA_PUBLISHED_VERSIONS="$published" ORCA_DAILY_RELEASE_NAMES="$names" \
node config/scripts/daily-build-version.mjs \
>"$RUNNER_TEMP/daily-identity.txt"
grep -E '^(version|build_number)=' "$RUNNER_TEMP/daily-identity.txt"
# Why check rather than trust: the checkout above pins `ref: main`, but a
# workflow_dispatch runs this file from whatever branch was dispatched. A
# branch that edits this step while main still has the old script yields
# an empty name and an untitled release — silent, and only visible once
# someone opens the releases page. Fail here instead.
if ! grep -q '^name=' "$RUNNER_TEMP/daily-identity.txt"; then
echo "::error::daily-build-version.mjs emitted no release name; this workflow and main's copy of the script are out of sync."
exit 1
fi
cat "$RUNNER_TEMP/daily-identity.txt" >>"$GITHUB_OUTPUT"
- name: Build app
if: steps.freshness.outputs.should_build == 'true'
run: pnpm build:release
env:
NODE_OPTIONS: --max-old-space-size=4096
# Why: daily builds are not an official channel — telemetry's transport
# gate accepts only 'stable' or 'rc', so leaving this unset keeps them
# silent, which is correct for unvetted dev artifacts.
ORCA_DIAGNOSTICS_TOKEN_URL: https://www.onorca.dev/diagnostics/token
# Why a second mint: everything from here on writes to the daily repo, and
# the notary round trip inside the publish step can be tens of minutes. The
# token minted at the top has already spent install + build of its one hour
# on steps that never touched it; restarting the clock here gives the slow
# part the full budget.
- name: Re-mint daily repo token for publish
id: app_token_publish
if: steps.freshness.outputs.should_build == 'true'
uses: actions/create-github-app-token@v2
with:
app-id: ${{ secrets.HOURLY_RELEASE_APP_ID }}
private-key: ${{ secrets.HOURLY_RELEASE_APP_PRIVATE_KEY }}
owner: stablyai
repositories: orca-daily
- name: Create daily release
id: release
if: steps.freshness.outputs.should_build == 'true'
shell: bash
env:
GH_TOKEN: ${{ steps.app_token_publish.outputs.token }}
TAG: v${{ steps.daily.outputs.version }}
NAME: ${{ steps.daily.outputs.name }}
SHA: ${{ steps.freshness.outputs.head_sha }}
run: |
set -euo pipefail
# Kept at 12 even though the title shows 7: the freshness check above
# parses this back out of the body to decide whether main has moved.
short_sha="${SHA:0:12}"
# Why a file rather than an inline string: the publish step re-asserts
# this same body, and the freshness check only works if the two agree
# exactly. One source, written once, read twice.
notes_file="$RUNNER_TEMP/daily-release-notes.md"
cat >"$notes_file" <<EOF
Automated daily macOS dev build from commit \`$short_sha\`.
Built from [\`stablyai/orca@$short_sha\`](https://github.com/stablyai/orca/commit/$SHA).
**Unvetted.** No tests ran. Signed and notarized like a release, so it
installs through Orca's in-app updater and opens from a manual download
without a Gatekeeper prompt — but nothing here has been reviewed.
EOF
# Why create it up front: electron-builder then uploads into a known tag
# rather than inferring one from package.json.
#
# Why --draft: everything between here and the manifest check is a window
# where the release exists but has no installable assets. A draft is
# absent from the releases list and from listReleaseBuilds, so a job that
# dies in that window — including a hard kill by the job timeout, which
# runs no cleanup step at all — leaves something invisible rather than a
# tag the picker offers and the download 404s on. It is flipped live only
# after the manifest is verified.
gh release create "$TAG" \
--repo "$DAILY_REPO" \
--title "$NAME" \
--draft \
--notes-file "$notes_file"
echo "tag=$TAG" >>"$GITHUB_OUTPUT"
echo "notes_file=$notes_file" >>"$GITHUB_OUTPUT"
- name: Publish daily macOS artifacts
if: steps.freshness.outputs.should_build == 'true'
uses: nick-fields/retry@v4
with:
# Why 45 like the release pipeline: an attempt is pack + notarize +
# upload, and the notary queue is the unbounded part. Why 2 attempts and
# not 3: a missed daily costs a day, and a third attempt buys less than
# it costs in runner time once the notary is that stuck.
timeout_minutes: 45
max_attempts: 2
retry_wait_seconds: 30
command: node config/scripts/ensure-native-runtime.mjs --runtime=electron && ORCA_MAC_DAILY=1 pnpm exec electron-builder --config config/electron-builder.config.cjs --mac --publish always
env:
# Why: electron-builder's github publisher targets the repo named in the
# config; the token must therefore carry write access to orca-daily.
GH_TOKEN: ${{ steps.app_token_publish.outputs.token }}
ORCA_DAILY_BUILD_VERSION: ${{ steps.daily.outputs.version }}
ORCA_BUILD_COMMIT: ${{ steps.daily.outputs.commit }}
CSC_LINK: ${{ secrets.MAC_CERTS }}
CSC_KEY_PASSWORD: ${{ secrets.MAC_CERTS_PASSWORD }}
# Why all three: electron-builder's notarize step authenticates to the
# Apple notary service with the app-specific password, not with the
# signing cert. Omitting them fails the build rather than skipping it,
# since `notarize` is now on for this path.
APPLE_ID: ${{ secrets.APPLE_ID }}
APPLE_APP_SPECIFIC_PASSWORD: ${{ secrets.APPLE_APP_SPECIFIC_PASSWORD }}
APPLE_TEAM_ID: ${{ secrets.APPLE_TEAM_ID }}
# Why a third mint: the upload step above allows two 45-minute attempts, so
# it can outlive the one-hour token minted before it. Every remaining step
# writes to the daily repo, including the failure path that discards the
# draft — a 401 there is exactly the stranded draft nobody can clean up.
#
# Why always(): a failed or cancelled upload is the case that needs this
# most, since the cleanup step below runs only on that path.
- name: Re-mint daily repo token for verify and cleanup
id: app_token_final
if: always() && steps.freshness.outputs.should_build == 'true'
uses: actions/create-github-app-token@v2
with:
app-id: ${{ secrets.HOURLY_RELEASE_APP_ID }}
private-key: ${{ secrets.HOURLY_RELEASE_APP_PRIVATE_KEY }}
owner: stablyai
repositories: orca-daily
# Why: the updater resolves a tag, then fetches latest-mac.yml from it. A
# release missing that manifest is a tag the picker offers and the download
# 404s on, so fail loudly instead of leaving a broken entry.
- name: Verify update manifest published
if: steps.freshness.outputs.should_build == 'true'
shell: bash
env:
GH_TOKEN: ${{ steps.app_token_final.outputs.token }}
TAG: ${{ steps.release.outputs.tag }}
run: |
set -euo pipefail
assets="$(gh release view "$TAG" --repo "$DAILY_REPO" --json assets --jq '.assets[].name')"
echo "Published assets:"
echo "$assets"
# Why exit 1 without deleting here: the release is still a draft, so it is
# already invisible to users, and the failure handler below owns cleanup.
# Deleting inline under `set -e` would also let the delete's exit code
# preempt this explicit failure.
for required in latest-mac.yml; do
if ! grep -qx "$required" <<<"$assets"; then
echo "::error::Daily draft $TAG is missing $required; the updater could not install it."
exit 1
fi
done
if ! grep -q '\.zip$' <<<"$assets"; then
echo "::error::Daily draft $TAG has no ZIP artifact for the updater to download."
exit 1
fi
# Why this is the last mutating step: publishing the draft is what makes the
# build visible to listReleaseBuilds. Doing it only after the manifest check
# means the picker can never offer a release whose assets are incomplete.
- name: Publish the verified release
id: publish_live
if: steps.freshness.outputs.should_build == 'true'
shell: bash
env:
GH_TOKEN: ${{ steps.app_token_final.outputs.token }}
TAG: ${{ steps.release.outputs.tag }}
NAME: ${{ steps.daily.outputs.name }}
NOTES_FILE: ${{ steps.release.outputs.notes_file }}
run: |
set -euo pipefail
# --title again: electron-builder resolves this draft by tag and may
# rewrite its title on upload. Re-asserting here means the name the
# picker reads is the one composed above, whatever it did in between.
# --notes-file for the same reason, and it matters more: the next run's
# freshness check parses the source commit back out of this body, so a
# body the publisher overwrote would rebuild an unchanged main daily.
gh release edit "$TAG" --repo "$DAILY_REPO" --draft=false --prerelease \
--title "$NAME" --notes-file "$NOTES_FILE"
echo "Published $TAG as \"$NAME\""
# Why: a draft left behind by a failed publish is invisible to users but still
# holds its tag name, so the next run for the same minute would collide.
#
# Why it is gated on publish_live not having succeeded: a later failure (the
# prune step) must not delete a release that already went live and that users
# may already be installing. A job killed by the outer timeout runs no steps
# at all — which is exactly why the release stays a draft until verified.
# Why cancelled() too: a run stopped from the Actions UI is not a failure(),
# so without it a manual cancel mid-publish would strand the draft.
- name: Discard the draft release on failure
if: >-
(failure() || cancelled()) && steps.release.outputs.tag != '' &&
steps.publish_live.outcome != 'success'
shell: bash
env:
# Fall back to the publish token: if the re-mint itself is what failed,
# the older token is the only one left and may still have time on it.
GH_TOKEN: ${{ steps.app_token_final.outputs.token || steps.app_token_publish.outputs.token }}
TAG: ${{ steps.release.outputs.tag }}
run: |
set -uo pipefail
# No --cleanup-tag: an unpublished draft never created a git tag.
echo "Run failed before publish; discarding draft $TAG"
gh release delete "$TAG" --repo "$DAILY_REPO" --yes ||
echo "::warning::Could not discard draft $TAG; remove it manually."
- name: Prune old daily releases
# Only after a live publish: $TAG is then a non-draft we must not delete,
# and failed runs should not reshuffle retention around a draft that the
# failure path is about to discard.
if: steps.publish_live.outcome == 'success'
shell: bash
env:
GH_TOKEN: ${{ steps.app_token_final.outputs.token }}
# Protect the tag this run just shipped; at the retain cap, a bad sort
# can otherwise mark the newest release as stale and delete it.
TAG: ${{ steps.release.outputs.tag }}
run: |
set -euo pipefail
# Why: --cleanup-tag so pruning does not leave orphan tags behind that
# keep showing up in tag lists with no release or assets attached.
# Drafts are excluded so retention counts shipped builds only; a stale
# draft is handled by the failure path, not by the retention window.
#
# Sort by publishedAt (not createdAt). Many non-draft dailies can share
# one createdAt (bulk import / re-create), so createdAt ranking is
# unstable. publishedAt is real recency; tagName (...YYYYMMDDHHMM) is
# the deterministic tie-break.
#
# Why force $TAG to the front before slicing: a hard retain-window seat
# for this run's release. Dropping $TAG from the list *before* the slice
# would permanently keep retain+1 releases; skipping it only in the
# delete loop would under-prune when the sort is still wrong. Partition
# keeps relative order of every other tag.
jq_filter='map(select(.isDraft | not)) | sort_by(.publishedAt // "", .tagName) | reverse'
if [[ -n "${TAG:-}" ]]; then
jq_filter+=" | (map(select(.tagName == \"${TAG//\"/\\\"}\")) + map(select(.tagName != \"${TAG//\"/\\\"}\")))"
fi
jq_filter+=" | .[${DAILY_RETAIN_COUNT}:] | .[].tagName"
stale="$(gh release list --repo "$DAILY_REPO" --limit 200 --json tagName,publishedAt,isDraft \
--jq "$jq_filter")"
if [[ -z "$stale" ]]; then
echo "Nothing to prune; at or under $DAILY_RETAIN_COUNT retained builds."
exit 0
fi
while read -r tag; do
[[ -n "$tag" ]] || continue
# Belt-and-suspenders: partition above should already exclude $TAG.
if [[ -n "${TAG:-}" && "$tag" == "$TAG" ]]; then
echo "::warning::Prune list still included just-published $tag after protect; skipping delete."
continue
fi
echo "Pruning $tag"
gh release delete "$tag" --repo "$DAILY_REPO" --yes --cleanup-tag || \
echo "::warning::Could not prune $tag"
done <<<"$stale"