ci(release): neutralize → sync-notice; native build/deploy is .hanzo/workflows/deploy.yml (GitHub is a mirror)

This commit is contained in:
2026-07-24 15:13:32 -07:00
parent 911969ede6
commit 4e6ec52afd
+8 -678
View File
@@ -1,684 +1,14 @@
name: release
# Cuts a release of ghcr.io/hanzoai/cloud. The invariant this workflow exists to
# enforce:
#
# a git tag v<X.Y.Z> exists ⇔ an image ghcr.io/hanzoai/cloud:v<X.Y.Z>
# was pushed AND booted to "listening" in the smoke test.
#
# The tag is a RECEIPT for a proven image, minted only AFTER a successful push —
# never a trigger for a build that might fail. The prior design triggered builds
# FROM pushed tags, so a tag could exist with no image behind it (a failed or
# never-run build) — universe would then try to roll that tag and the pods went
# ImagePullBackOff (phantom v1.786.42/43). Here the order is inverted:
#
# main push → compute next version → build → SMOKE → push image → tag → notify
#
# so a push/smoke/build failure fails the run BEFORE the tag step and leaves no
# tag; universe is only ever notified of a version whose image is proven present.
#
# DO NOT push v* tags by hand anymore. This workflow OWNS them. A hand-cut tag has
# no image behind it (exactly the phantom this prevents) and won't build (there is
# no `tags:` trigger). Every merge to main IS the release; skip one with the usual
# `[skip ci]` in the commit/merge message (a docs-only change need not ship).
#
# concurrency: a single serialized lane (cancel-in-progress:false — a
# mid-flight push must finish, never be killed between "image pushed" and "tag
# created"). Two main pushes can therefore never compute the same next number:
# the queued run starts only after the running one tags, re-reads the tags, and
# lands on the next patch. Monotonic by construction.
#
# The next version is max(highest git tag, highest pushed container tag) + 1 (patch
# bump only — never a major/minor jump). Folding in the container tags means we
# never reuse a number that already has a pushed image, even if some earlier run
# pushed an image but died before tagging.
#
# ── Infra notes (unchanged, still true) ─────────────────────────────────────────
# Self-hosted arcd amd64 scale set — NEVER GitHub-hosted runners (this org's
# GitHub-hosted Actions are billing-frozen). GHCR login uses GH_PAT, not the repo
# GITHUB_TOKEN: the ghcr.io/hanzoai/cloud package is linked to a DIFFERENT repo
# (hanzoai/ai, from the cloud→ai module rename), so this repo's GITHUB_TOKEN is
# denied write to it (permission_denied: write_package). GH_PAT (admin:org +
# write:packages) writes any hanzoai package regardless of package-repo linkage,
# and is the BuildKit gh_token the Dockerfile uses to fetch private cross-org Go
# modules. amd64-only: the cluster is amd64; one platform completes on the live
# scale set without waiting on the arm64 pool.
# Neutralized — the ONE canonical build/publish/deploy pipeline for
# ghcr.io/hanzoai/cloud is now .hanzo/workflows/deploy.yml (Hanzo Git push →
# in-cluster act_runner → BuildKit builds ./Dockerfile → ghcr.io/hanzoai/cloud:<sha>
# → kubectl patch app {cloud,cloud-iam2-canary,cloud-reader} → operator reconcile →
# Hanzo CD). GitHub is a mirror; this workflow no longer builds, pushes, tags, or
# promotes the image.
on:
push:
branches: [main]
workflow_dispatch:
# Cut tags on the cloud repo (git tag push) → contents: write. packages: write
# to push the image; id-token for provenance.
permissions:
contents: write
packages: write
id-token: write
# One serialized release lane. Never cancel in-flight: a run killed between
# "image pushed" and "tag created" is exactly the drift we are preventing.
concurrency:
group: release-cloud
cancel-in-progress: false
jobs:
build-amd64:
# ARC ephemeral runners match jobs targeting the scale-set NAME as a label.
runs-on: [hanzo-build-linux-amd64]
outputs:
# The FINAL assigned version comes from the Tag step (atomic free-version
# assignment), NOT the compute step — under concurrency the compute value may
# have been superseded. notify-universe must roll the version that was actually
# tagged + whose image was actually retagged.
version: ${{ steps.tag.outputs.version }}
version_v: ${{ steps.tag.outputs.version_v }}
steps:
- name: Checkout (full history + all tags — the version floor is read from tags)
uses: actions/checkout@v4
with:
fetch-depth: 0
fetch-tags: true
- name: Compute next version (monotonic patch bump over git + container tags)
id: ver
env:
GH_PAT: ${{ secrets.GH_PAT }}
run: |
set -euo pipefail
git fetch --tags --force --quiet
# Highest semver git tag (vX.Y.Z), normalised without the leading v.
git_max="$(git tag -l 'v[0-9]*.[0-9]*.[0-9]*' \
| sed 's/^v//' | grep -E '^[0-9]+\.[0-9]+\.[0-9]+$' | sort -V | tail -1 || true)"
# Best-effort: highest ALREADY-PUSHED container tag, so a number that has
# an image (even from a run that died before tagging) is never reused.
cont_max=""
if command -v gh >/dev/null 2>&1; then
cont_max="$(GH_TOKEN="$GH_PAT" gh api \
'/orgs/hanzoai/packages/container/cloud/versions?per_page=100' \
--jq '.[].metadata.container.tags[]?' 2>/dev/null \
| sed 's/^v//' | grep -E '^[0-9]+\.[0-9]+\.[0-9]+$' | sort -V | tail -1 || true)"
fi
# Floor = highest of the two; fall back to 1.786.0 only if the repo has
# no tags at all (first release ever).
max="$(printf '%s\n%s\n%s\n' "1.786.0" "$git_max" "$cont_max" \
| grep -E '^[0-9]+\.[0-9]+\.[0-9]+$' | sort -V | tail -1)"
major="${max%%.*}"; rest="${max#*.}"; minor="${rest%%.*}"; patch="${rest##*.}"
version="${major}.${minor}.$((patch + 1))"
# This version is a STARTING HINT only. It labels the smoke-built image and
# seeds the OCI metadata; the FINAL version is assigned atomically in the Tag
# step (which recomputes + retries on collision), so a taken number here is
# NOT fatal — the Tag step finds the next free one. Just note it and proceed.
if git rev-parse -q --verify "refs/tags/v${version}" >/dev/null; then
echo "note: hint v${version} already tagged — the Tag step will assign the next free version"
fi
echo "version=${version}" >> "$GITHUB_OUTPUT"
echo "version_v=v${version}" >> "$GITHUB_OUTPUT"
echo "major_minor=${major}.${minor}" >> "$GITHUB_OUTPUT"
echo "sha_short=$(git rev-parse --short "$GITHUB_SHA")" >> "$GITHUB_OUTPUT"
echo "Next release: v${version} (git_max='${git_max:-none}' container_max='${cont_max:-none}')"
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@v3
with:
driver: docker-container
driver-opts: network=host
- name: Mirror credential (registry.hanzo.ai)
# Dual-host: pull the KMS deploy kubeconfig (same Universal Auth flow
# the hanzoai/ci reusable uses), read the cluster-synced
# registry-credentials dockerconfig, and log in. Best-effort: absent
# creds → GHCR-only release, never a blocked tag.
env:
KMS_CLIENT_ID: ${{ secrets.KMS_CLIENT_ID }}
KMS_CLIENT_SECRET: ${{ secrets.KMS_CLIENT_SECRET }}
KMS_ENDPOINT: ${{ vars.KMS_ENDPOINT || 'https://kms.hanzo.ai' }}
REGISTRY_USER: ${{ secrets.REGISTRY_USER }}
REGISTRY_PASSWORD: ${{ secrets.REGISTRY_PASSWORD }}
run: |
set -uo pipefail
# Direct credential first (repo/org secret — works on private repos,
# where the Free plan hides org KMS secrets); KMS kubeconfig fallback.
if [ -n "${REGISTRY_USER:-}" ] && [ -n "${REGISTRY_PASSWORD:-}" ]; then
if echo "$REGISTRY_PASSWORD" | docker login registry.hanzo.ai -u "$REGISTRY_USER" --password-stdin; then
echo "MIRROR_OK=1" >> "$GITHUB_ENV"
else
echo "mirror login failed (registry.hanzo.ai unreachable) — mirror skipped, release continues"
fi
exit 0
fi
[ -z "${KMS_CLIENT_ID:-}" ] && { echo "no KMS creds — mirror skipped"; exit 0; }
TOKEN=$(curl -sf "$KMS_ENDPOINT/v1/kms/auth/login" -H 'Content-Type: application/json' -d "{\"clientId\":\"$KMS_CLIENT_ID\",\"clientSecret\":\"$KMS_CLIENT_SECRET\"}" | jq -r '.accessToken // empty')
[ -z "$TOKEN" ] && { echo "KMS login failed — mirror skipped"; exit 0; }
KC=$(curl -sf "$KMS_ENDPOINT/v1/kms/orgs/hanzo/secrets/deploy/KUBECONFIG?env=prod" -H "Authorization: Bearer $TOKEN" | jq -r '.secret.value // empty')
[ -z "$KC" ] && { echo "no KUBECONFIG in KMS — mirror skipped"; exit 0; }
echo "$KC" | base64 -d > "$RUNNER_TEMP/kubeconfig"
command -v kubectl >/dev/null 2>&1 || {
KVER=$(curl -fsSL https://dl.k8s.io/release/stable.txt)
mkdir -p "$HOME/.local/bin"
curl -fsSL "https://dl.k8s.io/release/${KVER}/bin/linux/amd64/kubectl" -o "$HOME/.local/bin/kubectl" && chmod +x "$HOME/.local/bin/kubectl"
export PATH="$HOME/.local/bin:$PATH"
}
CFG=$(KUBECONFIG="$RUNNER_TEMP/kubeconfig" kubectl -n hanzo get secret registry-credentials -o jsonpath='{.data.\.dockerconfigjson}' 2>/dev/null | base64 -d || true)
[ -z "$CFG" ] && { echo "registry-credentials unreadable — mirror skipped"; exit 0; }
UP=$(echo "$CFG" | jq -r '.auths["registry.hanzo.ai"].auth // empty' | base64 -d)
[ -z "$UP" ] && { echo "no registry auth — mirror skipped"; exit 0; }
echo "::add-mask::${UP#*:}"
if echo "${UP#*:}" | docker login registry.hanzo.ai -u "${UP%%:*}" --password-stdin; then
echo "MIRROR_OK=1" >> "$GITHUB_ENV"
else
echo "mirror login failed (registry.hanzo.ai unreachable) — mirror skipped, release continues"
fi
- name: Log in to ghcr.io (GH_PAT — writes the cloud package despite its ai-repo linkage)
uses: docker/login-action@v3
with:
registry: ghcr.io
username: hanzo-dev
password: ${{ secrets.GH_PAT }}
- name: Resolve decomplection artifact digests (the Go-only build's prebuilt inputs)
id: artifacts
run: |
set -euo pipefail
# cloud compiles ONLY Go; it pulls three prebuilt artifacts (console SPA,
# agent-skills catalog, native flags staticlib). Resolve each published
# :latest to an IMMUTABLE digest so THIS release is reproducible (pinned,
# not floating :latest) AND a console/skills/flags change is picked up —
# its CI republished :latest, so this resolves to the NEW digest. A MISSING
# artifact FAILS the release HERE, before build/smoke/push/tag: the receipt
# invariant means we never tag an image that couldn't embed the real console.
command -v crane >/dev/null 2>&1 || {
mkdir -p "$HOME/.local/bin"
curl -fsSL "https://github.com/google/go-containerregistry/releases/latest/download/go-containerregistry_Linux_x86_64.tar.gz" \
| tar -xz -C "$HOME/.local/bin" crane
}
export PATH="$HOME/.local/bin:$PATH"
resolve() {
local repo="$1" d
d="$(crane digest "ghcr.io/hanzoai/${repo}:latest" 2>/dev/null || true)"
[ -n "$d" ] || { echo "::error::decomplection artifact ghcr.io/hanzoai/${repo}:latest is not published — refusing to cut a release that would embed a stale/placeholder ${repo}"; return 1; }
printf 'ghcr.io/hanzoai/%s@%s' "$repo" "$d"
}
CONSOLE_IMAGE="$(resolve console-embed)" || exit 1
SKILLS_IMAGE="$(resolve agent-skills)" || exit 1
FLAGS_IMAGE="$(resolve cloud-flags)" || exit 1
{
echo "console_image=${CONSOLE_IMAGE}"
echo "skills_image=${SKILLS_IMAGE}"
echo "flags_image=${FLAGS_IMAGE}"
} >> "$GITHUB_OUTPUT"
echo "resolved: console=${CONSOLE_IMAGE} skills=${SKILLS_IMAGE} flags=${FLAGS_IMAGE}"
- name: OCI labels
id: meta
uses: docker/metadata-action@v5
with:
images: ghcr.io/hanzoai/cloud
tags: type=raw,value=${{ steps.ver.outputs.version_v }}
# ── Build → SMOKE → push → tag ───────────────────────────────────────────
# 1. Build once to a LOCAL tag (load into the daemon, do NOT push). Warms
# the BuildKit cache — the expensive console/npm + Go layers land here.
# 2. Boot that exact image and assert it reaches "listening" with no
# startup-crash signature (the gate).
# 3. Re-run build with push:true and the real tags: identical context /
# platform / secrets, so every layer is a cache hit from step 1 and the
# step only exports + pushes the already-tested image. Nothing that failed
# the smoke test can reach the registry.
# 4. Only after the push succeeds, mint + push the git tag (the receipt).
- name: Build (load locally for the smoke test)
uses: docker/build-push-action@v6
with:
context: .
platforms: linux/amd64
push: false
load: true
tags: cloud:smoke
labels: ${{ steps.meta.outputs.labels }}
# cloud compiles ONLY Go: pull the three prebuilt artifacts pinned to the
# digests resolved above (reproducible, and fresh — a console/skills/flags
# change is a new digest). No node/python/rust toolchain in this build.
build-args: |
CONSOLE_IMAGE=${{ steps.artifacts.outputs.console_image }}
SKILLS_IMAGE=${{ steps.artifacts.outputs.skills_image }}
FLAGS_IMAGE=${{ steps.artifacts.outputs.flags_image }}
# GIT_AUTH_TOKEN: BuildKit secret the Dockerfile consumes to fetch private
# cross-org Go modules (hanzoai/*, luxfi/*) over authenticated git.
secrets: |
GIT_AUTH_TOKEN=${{ secrets.GH_PAT }}
- name: Smoke test — the binary MUST boot to "listening" with no crash signature
run: |
set -euo pipefail
IMAGE=cloud:smoke
CID=""
cleanup() { [ -n "$CID" ] && docker rm -f "$CID" >/dev/null 2>&1 || true; }
trap cleanup EXIT
# Minimal, production-representative boot env:
# • a writable ephemeral /data root — the audit store, the embedded
# KMS secrets plane and every per-tenant SQLite open files under
# CLOUD_DATA_DIR; an unwritable dir would fail EVERY image before
# MountAll and the gate would stop discriminating good from bad; and
# • a throwaway 32-byte KMS master key so the KMS plane mounts on its
# normal ready path exactly as prod does (no real secret is used).
# The subsystem that crashed the incident (metrics, mount order 40)
# mounts AFTER kms (order 10), so the boot must get past kms for the
# gate to observe the panic — this env does exactly that.
KEY="$(head -c 32 /dev/urandom | base64 | tr -d '\n')"
CID="$(docker run -d \
--tmpfs /data:rw,size=64m \
-e CLOUD_DATA_DIR=/data \
-e CLOUD_ENV=smoke \
-e CLOUD_KMS_MASTER_KEY_REF="$KEY" \
"$IMAGE")"
# Poll up to 60s for boot to either finish ("listening" is logged once
# every subsystem has mounted and both transports are about to bind) or
# die (a Mount panic exits the process). A healthy boot is ~1-2s; the
# ceiling only guards a cold daemon.
listening=0
for _ in $(seq 1 60); do
logs="$(docker logs "$CID" 2>&1 || true)"
if printf '%s' "$logs" | grep -q '"message":"listening"'; then listening=1; break; fi
if [ "$(docker inspect -f '{{.State.Running}}' "$CID" 2>/dev/null || echo false)" != "true" ]; then break; fi
sleep 1
done
logs="$(docker logs "$CID" 2>&1 || true)"
echo "::group::cloud:smoke boot logs"
printf '%s\n' "$logs"
echo "::endgroup::"
# (1) No startup-crash signature. Catches the incident's Mount
# type-assert panic AND any generic Go panic — case-insensitive so a
# re-worded variant can't slip through — BEFORE a byte is pushed.
if printf '%s' "$logs" | grep -Eiq 'metrics\.Mount|mount metrics|panic|want \*zip\.App'; then
echo "SMOKE FAIL: startup-crash signature in boot logs (see above)"
exit 1
fi
# (2) Reached "listening" — proof that MountAll returned for every
# enabled subsystem (a failed Mount returns before this line).
if [ "$listening" -ne 1 ]; then
echo "SMOKE FAIL: binary never reached \"listening\" (a subsystem did not mount)"
exit 1
fi
# (3) Still alive — a server that logged "listening" then exited (e.g. a
# listener bind failure) is not a healthy image.
if [ "$(docker inspect -f '{{.State.Running}}' "$CID" 2>/dev/null || echo false)" != "true" ]; then
echo "SMOKE FAIL: process exited after \"listening\""
exit 1
fi
echo "SMOKE PASS: cloud:smoke booted to \"listening\" with no crash signature"
# ── Functional smoke — authenticated per-subsystem probe (the REAL gate) ─────
# The boot check above proves the process REACHES "listening"; this proves the
# mounted HTTP surface actually WORKS. /smoke (cmd/smoke, baked into the image)
# hits ONE side-effect-free read per core subsystem and FAILS the release on any
# broken code — above all a 402 on a READ (the balance-gate-over-blocks-reads
# regression) or a 5xx (a crash, e.g. the /v1/billing/usage self-dispatch 500).
# So a release can never ship with chat/billing/projects/kms/... down.
- name: Functional smoke — per-subsystem probe (fails the release if a core endpoint is broken)
env:
KMS_CLIENT_ID: ${{ secrets.KMS_CLIENT_ID }}
KMS_CLIENT_SECRET: ${{ secrets.KMS_CLIENT_SECRET }}
KMS_ENDPOINT: ${{ vars.KMS_ENDPOINT || 'https://kms.hanzo.ai' }}
# A KMS-provisioned short-lived smoke bearer, injected as a secret (NEVER
# hardcoded). Absent → the anonymous matrix still gates public/authed and
# catches every 402-on-read / 5xx.
SMOKE_TOKEN: ${{ secrets.SMOKE_TOKEN }}
run: |
set -euo pipefail
IMAGE=cloud:smoke
CID=""
cleanup() { [ -n "$CID" ] && docker rm -f "$CID" >/dev/null 2>&1 || true; }
trap cleanup EXIT
KEY="$(head -c 32 /dev/urandom | base64 | tr -d '\n')"
CID="$(docker run -d \
--tmpfs /data:rw,size=64m \
-e CLOUD_DATA_DIR=/data -e CLOUD_ENV=smoke -e CLOUD_KMS_MASTER_KEY_REF="$KEY" \
"$IMAGE")"
# Wait for the HTTP listener to bind (or the process to die).
up=0
for _ in $(seq 1 60); do
lg="$(docker logs "$CID" 2>&1 || true)"
printf '%s' "$lg" | grep -q '"message":"listening"' && { up=1; break; }
[ "$(docker inspect -f '{{.State.Running}}' "$CID" 2>/dev/null || echo false)" != "true" ] && break
sleep 1
done
if [ "$up" != 1 ]; then
echo "::group::boot logs"; docker logs "$CID" 2>&1 || true; echo "::endgroup::"
echo "FUNCTIONAL SMOKE INFRA FAIL: image never reached \"listening\""
exit 1
fi
# Token: prefer the injected secret; else mint from KMS (a provisioned smoke
# identity); else run the anonymous matrix. Never hardcoded.
if [ -z "${SMOKE_TOKEN:-}" ] && [ -n "${KMS_CLIENT_ID:-}" ]; then
KT=$(curl -sf "$KMS_ENDPOINT/v1/kms/auth/login" -H 'Content-Type: application/json' \
-d "{\"clientId\":\"$KMS_CLIENT_ID\",\"clientSecret\":\"$KMS_CLIENT_SECRET\"}" | jq -r '.accessToken // empty' || true)
[ -n "$KT" ] && SMOKE_TOKEN=$(curl -sf "$KMS_ENDPOINT/v1/kms/orgs/hanzo/secrets/smoke/TOKEN?env=prod" \
-H "Authorization: Bearer $KT" | jq -r '.secret.value // empty' || true)
fi
if [ -n "${SMOKE_TOKEN:-}" ]; then echo "::add-mask::$SMOKE_TOKEN"; echo "smoke: AUTHENTICATED matrix"; else echo "smoke: ANONYMOUS matrix (no SMOKE_TOKEN wired)"; fi
# /smoke is baked into the image (Dockerfile) — exec it INSIDE the container,
# so it probes the real mounted surface at localhost:8080 with no port/network
# plumbing. A non-zero exit here fails the release BEFORE any image is pushed.
docker exec \
-e SMOKE_BASE_URL=http://127.0.0.1:8080 \
-e SMOKE_TOKEN="${SMOKE_TOKEN:-}" \
"$CID" /smoke
# ── Migration smoke — the gate the v1.800.1 crashloop would have tripped ─────
# The plain smoke above boots on a FRESH /data, so every subsystem's migrate()
# takes its CREATE-TABLE path and no forward-migration is exercised — which is
# exactly why a DDL valid on a fresh store but broken on a pre-existing one (an
# index over a not-yet-ADDed column: affiliates referrer_org in v1.800.1, wallets
# project/agent before it) sailed through CI and took api.hanzo.ai down. This
# step reproduces the REAL prod upgrade path: boot the PRIOR released image to lay
# its on-disk (cek-encrypted) SQLite schema into a persistent volume, then boot
# the candidate over that SAME volume and require it to still reach "listening".
# A migrate() that assumes a fresh store dies here, before any image is pushed.
- name: Migration smoke — candidate MUST boot over the PRIOR release's on-disk schema
env:
# The image whose on-disk schema a prod upgrade migrates FROM — the tag the
# fleet runs today. Bump to the last-DEPLOYED tag as releases roll (override
# without a code change via the SMOKE_MIGRATION_BASELINE repo/org variable).
BASELINE: ${{ vars.SMOKE_MIGRATION_BASELINE }}
run: |
set -euo pipefail
BASELINE="${BASELINE:-ghcr.io/hanzoai/cloud:v1.799.19}"
CANDIDATE=cloud:smoke
VOL="cloudmig-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
B1=""; B2=""
cleanup() { docker rm -f "$B1" "$B2" >/dev/null 2>&1 || true; docker volume rm "$VOL" >/dev/null 2>&1 || true; }
trap cleanup EXIT
docker volume create "$VOL" >/dev/null
# ONE throwaway 32-byte master key for BOTH boots: cek seals each per-db DEK
# under it on the baseline boot and unwraps it on the candidate boot. A
# mismatched key fails closed (never opens), so sharing it is what puts the
# MIGRATE path — not a decrypt error — under test.
KEY="$(head -c 32 /dev/urandom | base64 | tr -d '\n')"
boot() { # $1=image $2=name -> prints container id
docker run -d --name "$2" \
-v "$VOL":/data \
-e CLOUD_DATA_DIR=/data \
-e CLOUD_ENV=smoke \
-e CLOUD_KMS_MASTER_KEY_REF="$KEY" \
"$1"
}
wait_listen() { # $1=container -> 0 if "listening", 1 if it died / timed out
for _ in $(seq 1 90); do
lg="$(docker logs "$1" 2>&1 || true)"
printf '%s' "$lg" | grep -q '"message":"listening"' && return 0
[ "$(docker inspect -f '{{.State.Running}}' "$1" 2>/dev/null || echo false)" != "true" ] && return 1
sleep 1
done
return 1
}
# A fresh docker named volume is root:root 0755, but the cloud image runs
# NON-ROOT, so it cannot create cek's <db>.cek.lock under /data (the single-boot
# smoke only worked because --tmpfs is world-writable). Make the shared volume
# writable for BOTH boots, re-opening it between them so the candidate can read
# the baseline's files even if their runtime UIDs differ.
chmod_vol() { docker run --rm --user 0 -v "$VOL":/data --entrypoint sh "$CANDIDATE" -c 'chmod -R 0777 /data'; }
# Boot 1 — the prior release writes its real schema into the volume. It must
# reach "listening" (proof every subsystem migrated + its DB is on disk); if
# the pinned baseline can't boot in this env the gate is blind, so fail loud.
echo "migration baseline: $BASELINE"
pulled=0; for _ in 1 2 3; do if docker pull "$BASELINE"; then pulled=1; break; fi; sleep 5; done
[ "$pulled" = 1 ] || { echo "MIGRATION SMOKE INFRA FAIL: cannot pull baseline $BASELINE"; exit 1; }
chmod_vol
B1="$(boot "$BASELINE" cloudmig_base)"
if ! wait_listen "$B1"; then
echo "::group::baseline boot logs"; docker logs "$B1" 2>&1 || true; echo "::endgroup::"
echo "MIGRATION SMOKE INFRA FAIL: baseline $BASELINE did not reach \"listening\" — cannot stage the prior schema (inspect/bump SMOKE_MIGRATION_BASELINE)"
exit 1
fi
docker stop "$B1" >/dev/null
chmod_vol
# Boot 2 — the candidate migrates that on-disk schema IN PLACE. This is the gate.
B2="$(boot "$CANDIDATE" cloudmig_cand)"
listening=0; wait_listen "$B2" && listening=1
logs="$(docker logs "$B2" 2>&1 || true)"
echo "::group::candidate migration boot logs"; printf '%s\n' "$logs"; echo "::endgroup::"
# 'no such column'/'no such table' is the exact index-before-ADD-COLUMN crash;
# 'panic' catches any generic Mount failure. The load-bearing check is the
# "listening" assertion below — a migrate() crash exits before it.
if printf '%s' "$logs" | grep -Eiq 'panic|no such column|no such table'; then
echo "MIGRATION SMOKE FAIL: candidate logged a DDL/migration error over the prior schema (the v1.800.1-class regression)"
exit 1
fi
if [ "$listening" -ne 1 ]; then
echo "MIGRATION SMOKE FAIL: candidate did NOT reach \"listening\" over the $BASELINE schema — a subsystem's migrate() crashes on a pre-existing store"
exit 1
fi
echo "MIGRATION SMOKE PASS: candidate booted to \"listening\" over the $BASELINE on-disk schema"
- name: Push (cache hit from the smoke build — publishes the tested image)
id: push
uses: docker/build-push-action@v6
with:
context: .
platforms: linux/amd64
push: true
# Push ONLY the immutable, per-commit sha- tag (always unique — never races)
# and the floating latest. The v<X.Y.Z> version is NOT pushed here: the
# compute-step version may be claimed by a concurrent release between compute
# and now, and pushing it would clobber that release's :vX image (mutable tag
# corruption). The version is assigned + the proven sha-image retagged to it
# ATOMICALLY in the Tag step below, so :vX exists iff its git tag exists.
tags: |
ghcr.io/hanzoai/cloud:sha-${{ steps.ver.outputs.sha_short }}
ghcr.io/hanzoai/cloud:latest
labels: ${{ steps.meta.outputs.labels }}
# SAME artifact digests as the smoke build → every layer is a cache hit from
# step 1 and the pushed image is byte-identical to the one smoke proved.
build-args: |
CONSOLE_IMAGE=${{ steps.artifacts.outputs.console_image }}
SKILLS_IMAGE=${{ steps.artifacts.outputs.skills_image }}
FLAGS_IMAGE=${{ steps.artifacts.outputs.flags_image }}
secrets: |
GIT_AUTH_TOKEN=${{ secrets.GH_PAT }}
# THE RECEIPT + ATOMIC VERSION ASSIGNMENT (race-safe). Reached only because
# build + smoke + push all succeeded, so a proven image exists under the unique
# sha- tag. Here we assign the next FREE v<X.Y.Z> and retag that proven image to
# it — atomically, with the git-tag push as the serialization point:
# * Recompute the next version FRESH (compute-step's value may have been claimed
# by a concurrent release in the build window).
# * If that version's git tag already exists, bump and retry.
# * Retag the proven sha-image → :vX (+ :X.Y.Z + :X.Y) via imagetools (metadata
# only, NO rebuild — byte-identical to the smoke-passed image).
# * Push the git tag; the FIRST pusher of vX wins, a loser deletes its local tag
# and recomputes. So concurrent releases each grab a distinct free number and
# the invariant "git tag vX ⇔ image :vX pushed+smoke-passed" holds under race.
- name: Tag the proven image (atomic free-version assignment — race-safe)
id: tag
env:
GH_PAT: ${{ secrets.GH_PAT }}
run: |
set -euo pipefail
git config user.name "hanzo-dev"
git config user.email "dev@hanzo.ai"
SHA_IMG="ghcr.io/hanzoai/cloud:sha-${{ steps.ver.outputs.sha_short }}"
PUSH_URL="https://x-access-token:${GH_PAT}@github.com/${GITHUB_REPOSITORY}.git"
for attempt in $(seq 1 8); do
git fetch --tags --force --quiet
git_max="$(git tag -l 'v[0-9]*.[0-9]*.[0-9]*' \
| sed 's/^v//' | grep -E '^[0-9]+\.[0-9]+\.[0-9]+$' | sort -V | tail -1 || true)"
# Newest page only — NOT --paginate. Container versions are created
# newest-first and version tags are monotonic, so the highest version
# is always among the most-recent versions; paginating the WHOLE
# registry history is what livelocked this step as tags accumulated.
# Fail-CLOSED. An ORPHANED container tag — image pushed by a run that
# died or was cancelled after imagetools-create but before its git tag —
# MUST raise the floor, or a later run reassigns that same number to a
# different image (an ambiguous mutable prod tag; the v1.801.50 flip). A
# git-only floor can't see the orphan, so if the container-tag lookup
# ERRORS (vs legitimately returning no tags) we retry the whole attempt
# rather than silently proceeding — a version with a pushed image is never
# reused. (Reordering git-tag before imagetools-create is the WRONG fix: it
# reintroduces the phantom "tag ⇔ no image" this workflow exists to prevent.)
cont_max=""
if command -v gh >/dev/null 2>&1; then
if cont_raw="$(GH_TOKEN="$GH_PAT" gh api \
'/orgs/hanzoai/packages/container/cloud/versions?per_page=100' \
--jq '.[].metadata.container.tags[]?' 2>/dev/null)"; then
cont_max="$(printf '%s\n' "$cont_raw" \
| sed 's/^v//' | grep -E '^[0-9]+\.[0-9]+\.[0-9]+$' | sort -V | tail -1 || true)"
else
echo " container-tag lookup failed — retry so an orphaned tag can't be reused (attempt $attempt)"; sleep 3; continue
fi
fi
max="$(printf '%s\n%s\n%s\n' "1.786.0" "$git_max" "$cont_max" \
| grep -E '^[0-9]+\.[0-9]+\.[0-9]+$' | sort -V | tail -1)"
major="${max%%.*}"; rest="${max#*.}"; minor="${rest%%.*}"; patch="${rest##*.}"
VER="${major}.${minor}.$((patch + 1))"; V="v${VER}"
if git rev-parse -q --verify "refs/tags/$V" >/dev/null; then
echo " $V already tagged — recomputing (attempt $attempt)"; sleep 3; continue
fi
docker buildx imagetools create \
-t "ghcr.io/hanzoai/cloud:${V}" \
-t "ghcr.io/hanzoai/cloud:${VER}" \
-t "ghcr.io/hanzoai/cloud:${major}.${minor}" \
"$SHA_IMG"
# Dual-host: mirror the release tags to OUR fleet registry (server-
# side copy) so the cluster never depends on GHCR to deploy. crane,
# not buildx imagetools: the IAM token realm doesn't answer buildx's
# multi-scope token request (spec gap, tracked), crane's single-scope
# flow works. Best-effort — a mirror hiccup never blocks the receipt.
if [ "${MIRROR_OK:-}" = "1" ]; then
command -v crane >/dev/null 2>&1 || {
curl -fsSL https://github.com/google/go-containerregistry/releases/download/v0.20.2/go-containerregistry_Linux_x86_64.tar.gz \
| tar -xz -C "$HOME/.local/bin" crane 2>/dev/null || {
mkdir -p "$HOME/.local/bin"
curl -fsSL https://github.com/google/go-containerregistry/releases/download/v0.20.2/go-containerregistry_Linux_x86_64.tar.gz \
| tar -xz -C "$HOME/.local/bin" crane
}
export PATH="$HOME/.local/bin:$PATH"
}
for MT in "${V}" "${VER}" "${major}.${minor}"; do
# Bounded: registry.hanzo.ai can *hang* (not just fail), and this
# is best-effort — an unbounded crane copy once livelocked the whole
# tag step and held the serialized release lane. timeout makes the
# mirror truly best-effort so the git-tag receipt below always runs.
timeout 120 crane copy "$SHA_IMG" "registry.hanzo.ai/hanzoai/cloud:${MT}" \
|| echo "::warning::mirror registry.hanzo.ai/hanzoai/cloud:${MT} failed or timed out"
done
fi
git tag -a "$V" -m "release $V — image ghcr.io/hanzoai/cloud:$V (retagged from sha-${{ steps.ver.outputs.sha_short }}, smoke-passed ${GITHUB_SHA})"
if git push "$PUSH_URL" "$V" 2>/dev/null; then
echo "Tagged $V → ghcr.io/hanzoai/cloud:$V"
echo "version=${VER}" >> "$GITHUB_OUTPUT"
echo "version_v=${V}" >> "$GITHUB_OUTPUT"
exit 0
fi
echo " push of $V lost the race — recomputing (attempt $attempt)"
git tag -d "$V" >/dev/null 2>&1 || true
sleep 3
done
echo "::error::could not acquire a free version tag after 8 attempts"
exit 1
# ── Promote: the declared-tag bump that makes the release DEPLOY ─────────────
# The tag minted above is the receipt for a pushed, smoke-passed image; THIS job
# records it as the desired state Hanzo CD reconciles. The universe-crs ArgoCD
# Application (ns hanzo-cd, `automated` sync + selfHeal) syncs
# infra/k8s/operator/crs/*.yaml → cluster and the operator rolls the Deployment,
# so a tag bump committed here reaches api.hanzo.ai with NO hand-dispatch and NO
# hand-edit of the CR.
#
# This is the SAME yq-bump → `deploy(<svc>): <tag>` universe commit the hanzoai/ci
# reusable (build.yml deploy step) does for every other service. cloud owns it
# HERE because its image is built by this workflow, not the ci reusable — its
# hanzo.yml carries no main `images:` entry and `# NO deploy`, so the shared
# deploy step never bumps cloud's CR. A direct in-cluster CR patch is NOT enough:
# ArgoCD selfHeal reverts any live edit not also recorded in git within ~45s.
# The retired notify-universe repository_dispatch had no receiver after the
# image-update.yml deploy hub was deleted in the Hanzo CD cutover; the git commit
# IS the sanctioned path now.
promote:
needs: build-amd64
# Only a real release promotes: build+smoke+push+tag all succeeded, so a
# proven v* image exists. A failure earlier leaves version_v empty → skipped.
if: ${{ needs.build-amd64.outputs.version_v != '' }}
sync-notice:
runs-on: [hanzo-build-linux-amd64]
steps:
- name: Record the proven tag in universe crs/cloud.yaml (Hanzo CD rolls it)
env:
# GH_PAT already pushes this repo's git tags above (contents:write on the
# hanzoai org), so it writes hanzoai/universe too — the SAME token the ci
# reusable falls back to for the universe deploy commit.
GH_PAT: ${{ secrets.GH_PAT }}
VERSION_V: ${{ needs.build-amd64.outputs.version_v }}
run: |
set -euo pipefail
[ -n "${GH_PAT:-}" ] || { echo "::error::no GH_PAT — cannot record the declared-tag bump in universe"; exit 1; }
# Bare arc runners ship no yq — provision the static binary (sudo-free,
# same pattern the ci reusable and this workflow's kubectl/crane fetches use).
if ! command -v yq >/dev/null 2>&1; then
mkdir -p "$HOME/.local/bin"; export PATH="$HOME/.local/bin:$PATH"
curl -fsSL https://github.com/mikefarah/yq/releases/latest/download/yq_linux_amd64 \
-o "$HOME/.local/bin/yq" && chmod +x "$HOME/.local/bin/yq"
fi
git clone -q --depth 1 \
"https://x-access-token:${GH_PAT}@github.com/hanzoai/universe.git" \
"$RUNNER_TEMP/universe"
CR="$RUNNER_TEMP/universe/infra/k8s/operator/crs/cloud.yaml"
[ -f "$CR" ] || { echo "::error::crs/cloud.yaml not found in universe"; exit 1; }
CUR="$(yq -r '.spec.image.tag // ""' "$CR")"
echo "cloud CR: ${CUR:-<empty>} → ${VERSION_V}"
# Monotonic guard: never roll the CR BACKWARD. Release runs finish under a
# serialized lane but a slow older run must never overwrite a newer promote.
# Skip iff the CR already holds a semver >= the version we just cut.
CURN="${CUR#v}"; NEWN="${VERSION_V#v}"
if printf '%s' "$CURN" | grep -qE '^[0-9]+\.[0-9]+\.[0-9]+$'; then
top="$(printf '%s\n%s\n' "$CURN" "$NEWN" | sort -V | tail -1)"
if [ "$top" = "$CURN" ] && [ "$CURN" != "$NEWN" ]; then
echo "::notice::cloud CR already at v${CURN} (≥ ${VERSION_V}) — not rolling back"; exit 0
fi
fi
yq -i ".spec.image.tag = \"${VERSION_V}\"" "$CR"
if git -C "$RUNNER_TEMP/universe" diff --quiet; then
echo "::notice::crs/cloud.yaml already at ${VERSION_V} — nothing to record"; exit 0
fi
git -C "$RUNNER_TEMP/universe" -c user.name=hanzo-ci -c user.email=dev@hanzo.ai \
commit -qam "deploy(cloud): ${VERSION_V} (${GITHUB_REPOSITORY}@$(echo "${GITHUB_SHA}" | cut -c1-7))"
# Rebase-safe push: universe main advances on every service's deploy, so a
# concurrent commit must not make cloud's promote lose the whole roll. Retry
# a few times, rebasing between attempts.
for attempt in $(seq 1 5); do
if git -C "$RUNNER_TEMP/universe" push -q origin HEAD:main; then
echo "recorded deploy(cloud): ${VERSION_V} — Hanzo CD (universe-crs) will roll it to api.hanzo.ai"
exit 0
fi
echo " universe push lost the race — rebasing (attempt ${attempt})"
git -C "$RUNNER_TEMP/universe" pull -q --rebase origin main || true
sleep 3
done
echo "::error::could not record the cloud tag bump in universe after 5 attempts"; exit 1
- run: echo "native pipeline is .hanzo/workflows/deploy.yml; GitHub is a mirror"