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
134 changes: 49 additions & 85 deletions .github/workflows/release.yml
Original file line number Diff line number Diff line change
Expand Up @@ -489,8 +489,12 @@ jobs:
group: stein-ai-production
cancel-in-progress: false
env:
DEPLOY_DIR: /srv/stein-ai
EXPECTED_HOST_FINGERPRINT: SHA256:jmzl6o5sZWlfwF2Ncg6hrEf7lluI0iDfp7t1W2MJwRk
# Production runs on the K3s cluster (k3s.seasonschange.cn), namespace stein-ai,
# Helm release "multica". KUBE_CONFIG is a kubeconfig for the namespace-scoped
# ServiceAccount stein-ai/ci-deployer; it cannot touch other namespaces.
NAMESPACE: stein-ai
RELEASE: multica
PUBLIC_URL: https://steinai.tmlab.top
BACKEND_IMAGE: docker.io/${{ secrets.DOCKER_USERNAME }}/multica-backend
WEB_IMAGE: docker.io/${{ secrets.DOCKER_USERNAME }}/multica-web
steps:
Expand All @@ -510,102 +514,62 @@ jobs:
exit 1
fi

- name: Install SSH client dependencies
run: |
sudo apt-get update
sudo apt-get install -y sshpass
- name: Set up Helm
uses: azure/setup-helm@v4
with:
version: v3.19.0

- name: Pin production SSH host key
- name: Configure cluster access
shell: bash
env:
SERVER_IP: ${{ secrets.SERVER_IP }}
SERVER_USER: ${{ secrets.SERVER_USER }}
SERVER_PWD: ${{ secrets.SERVER_PWD }}
KUBE_CONFIG: ${{ secrets.KUBE_CONFIG }}
run: |
set -Eeuo pipefail
[[ -n "$SERVER_IP" && -n "$SERVER_USER" && -n "$SERVER_PWD" ]] || {
echo "::error::SERVER_IP, SERVER_USER, and SERVER_PWD are required."
exit 1
}
[[ "$SERVER_IP" =~ ^([0-9]{1,3}\.){3}[0-9]{1,3}$ ]] || {
echo "::error::SERVER_IP must be an IPv4 address."
exit 1
}
IFS=. read -r -a address_parts <<< "$SERVER_IP"
for part in "${address_parts[@]}"; do
if ((10#$part > 255)); then
echo "::error::SERVER_IP contains an invalid IPv4 octet."
exit 1
fi
done
[[ "$SERVER_USER" =~ ^[A-Za-z_][A-Za-z0-9_-]*$ ]] || {
echo "::error::SERVER_USER contains unsupported characters."
exit 1
}
[[ -n "$KUBE_CONFIG" ]] || { echo "::error::Repository secret KUBE_CONFIG is required."; exit 1; }
install -m 0700 -d "$HOME/.kube"
printf '%s\n' "$KUBE_CONFIG" > "$HOME/.kube/config"
chmod 0600 "$HOME/.kube/config"
kubectl -n "$NAMESPACE" get deployment "$RELEASE-backend" "$RELEASE-frontend"

install -d -m 0700 "$HOME/.ssh"
scan_file=$(mktemp)
trap 'rm -f -- "$scan_file"' EXIT
ssh-keyscan -T 10 -t ed25519 "$SERVER_IP" >"$scan_file"
actual_fingerprint=$(ssh-keygen -lf "$scan_file" -E sha256 | awk 'NR == 1 { print $2 }')
if [[ "$actual_fingerprint" != "$EXPECTED_HOST_FINGERPRINT" ]]; then
echo "::error::Production SSH host fingerprint did not match the pinned value."
exit 1
fi
install -m 0600 "$scan_file" "$HOME/.ssh/known_hosts"

- name: Sync deployment files and deploy exact release
- name: Deploy release to K3s
shell: bash
env:
SERVER_IP: ${{ secrets.SERVER_IP }}
SERVER_USER: ${{ secrets.SERVER_USER }}
SSHPASS: ${{ secrets.SERVER_PWD }}
DOCKER_USERNAME: ${{ secrets.DOCKER_USERNAME }}
DOCKER_PASSWORD: ${{ secrets.DOCKER_PASSWORD }}
TAG_NAME: ${{ needs.verify.outputs.tag_name }}
run: |
set -Eeuo pipefail
[[ -n "$SERVER_IP" && -n "$SERVER_USER" && -n "$SSHPASS" ]]
[[ -n "$DOCKER_USERNAME" && -n "$DOCKER_PASSWORD" ]] || {
echo "::error::DOCKER_USERNAME and DOCKER_PASSWORD are required for production deployment."
exit 1
}
[[ "$SERVER_IP" =~ ^([0-9]{1,3}\.){3}[0-9]{1,3}$ ]]
IFS=. read -r -a address_parts <<< "$SERVER_IP"
for part in "${address_parts[@]}"; do
((10#$part <= 255))
done
[[ "$SERVER_USER" =~ ^[A-Za-z_][A-Za-z0-9_-]*$ ]]
[[ "$GITHUB_RUN_ID" =~ ^[0-9]+$ && "$GITHUB_RUN_ATTEMPT" =~ ^[0-9]+$ ]]
[[ "$DOCKER_USERNAME" =~ ^[a-z0-9]+([._-][a-z0-9]+)*$ ]] || {
echo "::error::DOCKER_USERNAME is not a valid Docker Hub namespace."
exit 1
}
[[ "$TAG_NAME" =~ ^v[0-9]+\.[0-9]+\.[0-9]+(-[0-9A-Za-z.-]+)?$ ]]
echo "Deploying $TAG_NAME (previous: $(helm -n "$NAMESPACE" get values "$RELEASE" -a -o json | jq -r '.images.backend.tag'))"
# The backend runs `migrate up` on start; the chart uses strategy Recreate.
# On failure: helm -n stein-ai rollback multica
if ! helm upgrade "$RELEASE" deploy/helm/multica \
--namespace "$NAMESPACE" \
-f deploy/k3s/values.yaml \
--set images.backend.repository="$BACKEND_IMAGE" \
--set images.backend.tag="$TAG_NAME" \
--set images.frontend.repository="$WEB_IMAGE" \
--set images.frontend.tag="$TAG_NAME" \
--wait --timeout 10m; then
kubectl -n "$NAMESPACE" get pods -o wide || true
kubectl -n "$NAMESPACE" logs "deployment/$RELEASE-backend" --tail=80 || true
echo "::error::Deployment of $TAG_NAME failed. Roll back with: helm -n $NAMESPACE rollback $RELEASE"
exit 1
fi
kubectl -n "$NAMESPACE" apply -f deploy/k3s/ingress.yaml
helm -n "$NAMESPACE" history "$RELEASE" --max 3

target="${SERVER_USER}@${SERVER_IP}"
suffix="${GITHUB_RUN_ID}.${GITHUB_RUN_ATTEMPT}"
remote_compose="${DEPLOY_DIR}/.docker-compose.selfhost.yml.${suffix}"
remote_script="${DEPLOY_DIR}/.deploy-selfhost.sh.${suffix}"
ssh_options=(
-o StrictHostKeyChecking=yes
-o UserKnownHostsFile="$HOME/.ssh/known_hosts"
-o PasswordAuthentication=yes
-o PubkeyAuthentication=no
-o ConnectTimeout=15
)

sshpass -e ssh "${ssh_options[@]}" "$target" \
"install -d -m 0750 '$DEPLOY_DIR'"
sshpass -e scp "${ssh_options[@]}" docker-compose.selfhost.yml \
"$target:$remote_compose"
sshpass -e scp "${ssh_options[@]}" scripts/deploy-selfhost.sh \
"$target:$remote_script"
sshpass -e ssh "${ssh_options[@]}" "$target" \
"install -m 0644 '$remote_compose' '$DEPLOY_DIR/docker-compose.selfhost.yml' && install -m 0750 '$remote_script' '$DEPLOY_DIR/deploy-selfhost.sh' && rm -f -- '$remote_compose' '$remote_script'"

printf '%s\n' "$DOCKER_PASSWORD" | sshpass -e ssh "${ssh_options[@]}" "$target" \
"$DEPLOY_DIR/deploy-selfhost.sh" "$TAG_NAME" "$BACKEND_IMAGE" "$WEB_IMAGE" "$DOCKER_USERNAME"
- name: Check public endpoint
shell: bash
run: |
set -Eeuo pipefail
for attempt in $(seq 1 12); do
if curl -fsS -o /dev/null "$PUBLIC_URL/api/config" && curl -fsS -o /dev/null "$PUBLIC_URL/"; then
echo "$PUBLIC_URL is healthy"
exit 0
fi
sleep 5
done
echo "::error::$PUBLIC_URL did not become healthy after the deployment."
exit 1

helm-chart:
needs: [verify, docker-backend-merge, docker-web-merge]
Expand Down
13 changes: 13 additions & 0 deletions deploy/k3s/ingress.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
# Single public host: /ws goes straight to the backend, everything else to the frontend
# (the frontend proxies /api to the backend via REMOTE_API_URL).
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata: {name: steinai, namespace: stein-ai}
spec:
ingressClassName: traefik
rules:
- host: steinai.tmlab.top
http:
paths:
- {path: /ws, pathType: Prefix, backend: {service: {name: multica-backend, port: {number: 8080}}}}
- {path: /, pathType: Prefix, backend: {service: {name: multica-frontend, port: {number: 3000}}}}
42 changes: 42 additions & 0 deletions deploy/k3s/values.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,42 @@
# Production values for the K3s cluster (namespace: stein-ai, release: multica).
# Used by .github/workflows/release.yml (deploy-production). Image tags are set by CI
# with --set; everything else lives here so the deployed config is versioned.
#
# Secrets are NOT stored here. The backend reads them from the pre-created
# Secret `multica-secrets` (JWT_SECRET, DATABASE_URL, SMTP_*, ...).
# The database is the shared CloudNativePG cluster `pg-main` in namespace `database`.
existingSecret: multica-secrets
images:
backend: {repository: docker.io/tmllab/multica-backend, tag: v0.1.9}
frontend: {repository: docker.io/tmllab/multica-web, tag: v0.1.9}
postgres:
external: {enabled: true}
backend:
replicas: 1
uploads:
persistence: {enabled: true, size: 5Gi, storageClass: local-path}
config:
appEnv: production
appUrl: https://steinai.tmlab.top
frontendOrigin: https://steinai.tmlab.top
corsAllowedOrigins: https://steinai.tmlab.top
cookieDomain: ""
allowSignup: false
disableWorkspaceCreation: false
googleRedirectUri: https://steinai.tmlab.top/auth/callback
localUploadBaseUrl: ""
resources:
requests: {cpu: 50m, memory: 96Mi}
limits: {memory: 1Gi}
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions: [{key: role, operator: In, values: [legacy]}]
frontend:
replicas: 1
resources:
requests: {cpu: 50m, memory: 160Mi}
limits: {memory: 1Gi}
ingress:
enabled: false
Loading