Skip to content

Commit 6439dde

Browse files
Locatelli-FlorLocatelli-Flor
committed
Refactor deployment workflow to improve rollout checks and notifications
1 parent f30a96a commit 6439dde

File tree

1 file changed

+140
-116
lines changed

1 file changed

+140
-116
lines changed

‎.github/workflows/deploy.yml‎

Lines changed: 140 additions & 116 deletions
Original file line numberDiff line numberDiff line change
@@ -9,13 +9,14 @@ on:
99
env:
1010
REGISTRY: crretoxmas2024.azurecr.io
1111
NAMESPACE: reto-xmas-2025-goland-ia-backend
12-
ROLLOUT_TIMEOUT: 300
13-
READY_CHECK_INTERVAL: 10
12+
ROLLOUT_TIMEOUT: 60s
13+
READY_CHECK_RETRIES: 20
14+
READY_CHECK_SLEEP: 15
1415

1516
jobs:
1617
build-and-deploy:
1718
runs-on: ubuntu-latest
18-
timeout-minutes: 20
19+
timeout-minutes: 15
1920
strategy:
2021
fail-fast: false
2122
matrix:
@@ -30,130 +31,153 @@ jobs:
3031
deployment: rag-manager
3132

3233
steps:
33-
- name: Checkout code
34-
uses: actions/checkout@v4
35-
36-
- name: Set up Docker Buildx
37-
uses: docker/setup-buildx-action@v3
38-
39-
- name: Login to ACR
40-
uses: docker/login-action@v3
41-
with:
42-
registry: ${{ env.REGISTRY }}
43-
username: ${{ secrets.ACR_USERNAME }}
44-
password: ${{ secrets.ACR_PASSWORD }}
45-
46-
- name: Build and push Docker image
47-
uses: docker/build-push-action@v5
48-
with:
49-
context: ${{ matrix.service.path }}
50-
platforms: linux/amd64
51-
push: true
52-
tags: |
53-
${{ env.REGISTRY }}/${{ matrix.service.image }}:latest
54-
${{ env.REGISTRY }}/${{ matrix.service.image }}:${{ github.sha }}
55-
cache-from: type=registry,ref=${{ env.REGISTRY }}/${{ matrix.service.image }}:buildcache
56-
cache-to: type=registry,ref=${{ env.REGISTRY }}/${{ matrix.service.image }}:buildcache,mode=max
57-
provenance: false
58-
59-
- name: Set up kubectl
60-
uses: azure/setup-kubectl@v3
61-
with:
62-
version: 'latest'
63-
64-
- name: Configure kubectl
65-
run: |
66-
mkdir -p $HOME/.kube
67-
echo "${{ secrets.KUBECONFIG }}" | base64 -d > $HOME/.kube/config
68-
chmod 600 $HOME/.kube/config
69-
70-
- name: Update deployment image
71-
run: |
72-
kubectl set image deployment/${{ matrix.service.deployment }} \
73-
api=${{ env.REGISTRY }}/${{ matrix.service.image }}:${{ github.sha }} \
74-
-n ${{ env.NAMESPACE }}
75-
76-
- name: Wait for deployment to be ready (robust)
77-
run: |
78-
echo "Waiting for deployment ${matrix.service.deployment} to be ready..."
79-
END=$((SECONDS+ROLLOUT_TIMEOUT))
80-
81-
while [ $SECONDS -lt $END ]; do
82-
DESIRED=$(kubectl get deployment/${{ matrix.service.deployment }} -n ${{ env.NAMESPACE }} -o jsonpath='{.spec.replicas}')
83-
AVAILABLE=$(kubectl get deployment/${{ matrix.service.deployment }} -n ${{ env.NAMESPACE }} -o jsonpath='{.status.availableReplicas}')
84-
AVAILABLE=${AVAILABLE:-0}
85-
86-
echo "Available replicas: $AVAILABLE / $DESIRED"
87-
88-
if [ "$AVAILABLE" = "$DESIRED" ]; then
89-
echo "Deployment is ready!"
90-
exit 0
91-
fi
92-
93-
sleep $READY_CHECK_INTERVAL
94-
done
95-
96-
echo "Deployment did not become ready in time"
97-
kubectl describe deployment/${{ matrix.service.deployment }} -n ${{ env.NAMESPACE }}
98-
kubectl get pods -n ${{ env.NAMESPACE }} -o wide
99-
exit 1
100-
101-
- name: Get logs on failure
102-
if: failure()
103-
run: |
104-
echo "=== Deployment ==="
105-
kubectl describe deployment/${{ matrix.service.deployment }} -n ${{ env.NAMESPACE }}
106-
107-
echo "=== Pods ==="
108-
kubectl get pods -n ${{ env.NAMESPACE }} -l app=${{ matrix.service.deployment }} -o wide
109-
110-
echo "=== Pod Logs ==="
111-
kubectl logs -n ${{ env.NAMESPACE }} \
112-
-l app=${{ matrix.service.deployment }} \
113-
--tail=100 \
114-
--all-containers=true \
115-
--prefix=true || true
116-
117-
- name: Deployment Summary
118-
if: always()
119-
run: |
120-
STATUS="${{ job.status }}"
121-
122-
echo "### Deployment - ${{ matrix.service.name }}" >> $GITHUB_STEP_SUMMARY
123-
echo "" >> $GITHUB_STEP_SUMMARY
124-
echo "| Property | Value |" >> $GITHUB_STEP_SUMMARY
125-
echo "|----------|-------|" >> $GITHUB_STEP_SUMMARY
126-
echo "| Image | \`${{ env.REGISTRY }}/${{ matrix.service.image }}:${{ github.sha }}\` |" >> $GITHUB_STEP_SUMMARY
127-
echo "| Status | $STATUS |" >> $GITHUB_STEP_SUMMARY
128-
echo "" >> $GITHUB_STEP_SUMMARY
129-
130-
echo "#### Pods:" >> $GITHUB_STEP_SUMMARY
131-
echo '```' >> $GITHUB_STEP_SUMMARY
132-
kubectl get pods -n ${{ env.NAMESPACE }} -l app=${{ matrix.service.deployment }} >> $GITHUB_STEP_SUMMARY || true
133-
echo '```' >> $GITHUB_STEP_SUMMARY
34+
- name: Checkout code
35+
uses: actions/checkout@v4
36+
37+
- name: Set up Docker Buildx
38+
uses: docker/setup-buildx-action@v3
39+
40+
- name: Login to ACR
41+
uses: docker/login-action@v3
42+
with:
43+
registry: ${{ env.REGISTRY }}
44+
username: ${{ secrets.ACR_USERNAME }}
45+
password: ${{ secrets.ACR_PASSWORD }}
46+
47+
- name: Build and push Docker image
48+
uses: docker/build-push-action@v5
49+
with:
50+
context: ${{ matrix.service.path }}
51+
platforms: linux/amd64
52+
push: true
53+
tags: |
54+
${{ env.REGISTRY }}/${{ matrix.service.image }}:latest
55+
${{ env.REGISTRY }}/${{ matrix.service.image }}:${{ github.sha }}
56+
cache-from: type=registry,ref=${{ env.REGISTRY }}/${{ matrix.service.image }}:buildcache
57+
cache-to: type=registry,ref=${{ env.REGISTRY }}/${{ matrix.service.image }}:buildcache,mode=max
58+
provenance: false
59+
60+
- name: Set up kubectl
61+
uses: azure/setup-kubectl@v3
62+
with:
63+
version: 'latest'
64+
65+
- name: Configure kubectl
66+
run: |
67+
mkdir -p $HOME/.kube
68+
echo "${{ secrets.KUBECONFIG }}" | base64 -d > $HOME/.kube/config
69+
chmod 600 $HOME/.kube/config
70+
71+
- name: Update deployment image
72+
run: |
73+
kubectl set image deployment/${{ matrix.service.deployment }} \
74+
api=${{ env.REGISTRY }}/${{ matrix.service.image }}:${{ github.sha }} \
75+
-n ${{ env.NAMESPACE }}
76+
77+
- name: Wait for deployment to be ready (robust)
78+
run: |
79+
set -e
80+
81+
echo "Checking rollout status (non-blocking)..."
82+
kubectl rollout status deployment/${{ matrix.service.deployment }} \
83+
-n ${{ env.NAMESPACE }} \
84+
--timeout=${{ env.ROLLOUT_TIMEOUT }} || true
85+
86+
echo "Waiting for available replicas..."
87+
88+
for i in $(seq 1 $READY_CHECK_RETRIES); do
89+
DESIRED=$(kubectl get deployment/${{ matrix.service.deployment }} -n ${{ env.NAMESPACE }} -o jsonpath='{.spec.replicas}')
90+
AVAILABLE=$(kubectl get deployment/${{ matrix.service.deployment }} -n ${{ env.NAMESPACE }} -o jsonpath='{.status.availableReplicas}')
91+
92+
echo "Attempt $i: $AVAILABLE / $DESIRED replicas available"
93+
94+
if [ "$AVAILABLE" = "$DESIRED" ]; then
95+
echo "Deployment is ready"
96+
exit 0
97+
fi
98+
99+
sleep $READY_CHECK_SLEEP
100+
done
101+
102+
echo "Deployment did not become ready in time"
103+
kubectl describe deployment/${{ matrix.service.deployment }} -n ${{ env.NAMESPACE }}
104+
exit 1
105+
106+
- name: Verify pods
107+
if: success()
108+
run: |
109+
kubectl get pods -n ${{ env.NAMESPACE }} -l app=${{ matrix.service.deployment }} -o wide
110+
111+
- name: Get logs on failure
112+
if: failure()
113+
run: |
114+
echo "=== Deployment ==="
115+
kubectl describe deployment/${{ matrix.service.deployment }} -n ${{ env.NAMESPACE }}
116+
117+
echo "=== Pods ==="
118+
kubectl get pods -n ${{ env.NAMESPACE }} -l app=${{ matrix.service.deployment }} -o wide
119+
120+
echo "=== Pod Logs ==="
121+
kubectl logs -n ${{ env.NAMESPACE }} \
122+
-l app=${{ matrix.service.deployment }} \
123+
--tail=100 \
124+
--all-containers=true \
125+
--prefix=true || true
126+
127+
- name: Deployment Summary
128+
if: always()
129+
run: |
130+
STATUS="${{ job.status }}"
131+
132+
echo "### Deployment - ${{ matrix.service.name }}" >> $GITHUB_STEP_SUMMARY
133+
echo "" >> $GITHUB_STEP_SUMMARY
134+
echo "| Property | Value |" >> $GITHUB_STEP_SUMMARY
135+
echo "|----------|-------|" >> $GITHUB_STEP_SUMMARY
136+
echo "| Image | \`${{ env.REGISTRY }}/${{ matrix.service.image }}:${{ github.sha }}\` |" >> $GITHUB_STEP_SUMMARY
137+
echo "| Status | $STATUS |" >> $GITHUB_STEP_SUMMARY
138+
echo "" >> $GITHUB_STEP_SUMMARY
139+
140+
echo "#### Pods:" >> $GITHUB_STEP_SUMMARY
141+
echo '```' >> $GITHUB_STEP_SUMMARY
142+
kubectl get pods -n ${{ env.NAMESPACE }} -l app=${{ matrix.service.deployment }} >> $GITHUB_STEP_SUMMARY || true
143+
echo '```' >> $GITHUB_STEP_SUMMARY
134144
135145
notify-success:
136146
name: Deployment Success
137147
runs-on: ubuntu-latest
138148
needs: [build-and-deploy]
139149
if: success()
140150
steps:
141-
- name: Notify Discord
142-
run: |
143-
curl -H "Content-Type: application/json" \
144-
-X POST \
145-
-d "{\"content\": \"✅ **Deploy exitoso**\\n\\n**Repositorio:** ${{ github.repository }}\\n**Commit:** [${{ github.sha }}](${{ github.server_url }}/${{ github.repository }}/commit/${{ github.sha }})\\n**Autor:** ${{ github.actor }}\\n\\n[Ver ejecución en GitHub Actions](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }})\\n\\n🚀 Todos los servicios quedaron desplegados correctamente.\"}" \
146-
${{ secrets.DISCORD_WEBHOOK }}
151+
- name: Success Summary
152+
run: |
153+
echo "### All Services Deployed" >> $GITHUB_STEP_SUMMARY
154+
echo "" >> $GITHUB_STEP_SUMMARY
155+
echo "**Live URLs:**" >> $GITHUB_STEP_SUMMARY
156+
echo "- [DocsManager](https://goland-ia-backend-docs-manager.reto-ucu.net/docs)" >> $GITHUB_STEP_SUMMARY
157+
echo "- [RAGManager](https://goland-ia-backend-rag-manager.reto-ucu.net/docs)" >> $GITHUB_STEP_SUMMARY
158+
159+
- name: Notificar a Discord (deploy exitoso)
160+
run: |
161+
curl -H "Content-Type: application/json" \
162+
-X POST \
163+
-d "{\"content\": \"✅ **Deploy exitoso**\\n\\n**Servicio:** \\\\`${{ matrix.service.name }}\\\\`\\n**Imagen:** \\\\`${{ env.REGISTRY }}/${{ matrix.service.image }}:${{ github.sha }}\\\\`\\n**Namespace:** \\\\`${{ env.NAMESPACE }}\\\\`\\n**Autor:** ${{ github.actor }}\\n**Commit:** [${{ github.sha }}](${{ github.server_url }}/${{ github.repository }}/commit/${{ github.sha }})\\n\\n[Ver ejecución en GitHub Actions](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }})\\n\\n:rocket: ¡Todo salió bien!\"}" \
164+
https://discord.com/api/webhooks/1449147972330852463/sMs3QYOvQ6oiaPHR4PlBluwNrhWSsVmfm3_Miz6UAFrPxj1SsbqNnHXc5eK9h8ZSaumk
147165
148166
notify-failure:
149167
name: Deployment Failed
150168
runs-on: ubuntu-latest
151169
needs: [build-and-deploy]
152170
if: failure()
153171
steps:
154-
- name: Notify Discord
155-
run: |
156-
curl -H "Content-Type: application/json" \
157-
-X POST \
158-
-d "{\"content\": \"❌ **Falló el deploy**\\n\\n**Repositorio:** ${{ github.repository }}\\n**Commit:** [${{ github.sha }}](${{ github.server_url }}/${{ github.repository }}/commit/${{ github.sha }})\\n**Autor:** ${{ github.actor }}\\n\\n[Ver ejecución en GitHub Actions](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }})\\n\\n⚠️ Al menos un servicio no pudo desplegarse. Revisar logs.\"}" \
159-
${{ secrets.DISCORD_WEBHOOK }}
172+
- name: Failure Summary
173+
run: |
174+
echo "### Deployment Failed" >> $GITHUB_STEP_SUMMARY
175+
echo "" >> $GITHUB_STEP_SUMMARY
176+
echo "Please check deployment logs above." >> $GITHUB_STEP_SUMMARY
177+
178+
- name: Notificar a Discord (deploy fallido)
179+
run: |
180+
curl -H "Content-Type: application/json" \
181+
-X POST \
182+
-d "{\"content\": \"❌ **Falló el deploy**\\n\\n**Servicio:** \\\\`${{ matrix.service.name }}\\\\`\\n**Imagen:** \\\\`${{ env.REGISTRY }}/${{ matrix.service.image }}:${{ github.sha }}\\\\`\\n**Namespace:** \\\\`${{ env.NAMESPACE }}\\\\`\\n**Autor:** ${{ github.actor }}\\n**Commit:** [${{ github.sha }}](${{ github.server_url }}/${{ github.repository }}/commit/${{ github.sha }})\\n\\n[Ver ejecución en GitHub Actions](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }})\\n\\n:warning: Revisa los logs para más detalles.\"}" \
183+
https://discord.com/api/webhooks/1449147972330852463/sMs3QYOvQ6oiaPHR4PlBluwNrhWSsVmfm3_Miz6UAFrPxj1SsbqNnHXc5eK9h8ZSaumk

0 commit comments

Comments
 (0)