Update workflow yml

Author: tuhaihe

.github/workflows/sandbox-validation.yml

@@ -71,6 +71,7 @@ jobs:
     runs-on: ubuntu-latest
 
     strategy:
+      fail-fast: false
       matrix:
         version: ['main', '2.0.0']
 
@@ -91,23 +92,28 @@ jobs:
           cd devops/sandbox
           ./run.sh -c ${{ matrix.version }}
 
-          echo "Waiting for Apache Cloudberry deployment to complete..."
-          timeout 120 bash -c 'until docker logs cbdb-cdw 2>&1 | grep -q "DEPLOYMENT SUCCESSFUL"; do sleep 10; echo "Still waiting for deployment..."; done'
-          echo "Deployment completed successfully!"
-        timeout-minutes: 30
-
-      - name: Test database operations
-        run: |
-          echo "Waiting for database readiness (up to 60s)..."
-          for i in {1..12}; do
-            if docker exec -u gpadmin cbdb-cdw bash -l -c "pg_isready -h localhost -p 5432 -U gpadmin" >/dev/null 2>&1; then
-              echo "Database is ready."
+          echo "Waiting for deployment to complete..."
+          for i in {1..30}; do
+            if docker logs cbdb-cdw 2>&1 | grep -q "DEPLOYMENT SUCCESSFUL"; then
+              echo "✅ Deployment completed successfully"
               break
             fi
-            if [ "$i" -eq 12 ]; then
-              echo "Database did not become ready in time."
+            if ! docker ps | grep -q cbdb-cdw; then
+              echo "ERROR: Container stopped unexpectedly"
+              docker logs cbdb-cdw
               exit 1
             fi
+            [ $i -eq 30 ] && echo "ERROR: Deployment timeout" && docker logs cbdb-cdw && exit 1
+            sleep 10
+          done
+        timeout-minutes: 30
+
+      - name: Test database operations
+        run: |
+          echo "Waiting for database readiness..."
+          for i in {1..24}; do
+            docker exec -u gpadmin cbdb-cdw bash -l -c "pg_isready -h localhost -p 5432 -U gpadmin" >/dev/null 2>&1 && break
+            [ $i -eq 24 ] && echo "ERROR: Database not ready" && exit 1
             sleep 5
           done
 
@@ -116,10 +122,10 @@ jobs:
 
           echo "Verifying segment configuration..."
           docker exec -u gpadmin cbdb-cdw bash -l -c "psql -c 'SELECT * FROM gp_segment_configuration'"
-
+          
           echo "Checking available extensions..."
           docker exec -u gpadmin cbdb-cdw bash -l -c "psql -c 'SELECT * FROM pg_available_extensions'"
-
+          
           echo "✅ Single node test (${{ matrix.version }}) completed successfully"
 
       - name: Check container logs on failure
@@ -131,18 +137,21 @@ jobs:
           docker logs --tail 100 cbdb-cdw || true
           echo "=== Container Resource Usage ==="
           docker stats --no-stream cbdb-cdw || true
+          echo "=== Disk Status ==="
+          df -h
 
       - name: Cleanup
         if: always()
         run: |
           docker rm -f cbdb-cdw || true
-          docker system prune -af || true
+          docker system prune -af --volumes || true
 
   test-multi-node-build:
     name: Test Multi Node Build
     runs-on: ubuntu-latest
 
     strategy:
+      fail-fast: false
       matrix:
         version: ['main', '2.0.0']
 
@@ -163,32 +172,36 @@ jobs:
           cd devops/sandbox
           ./run.sh -c ${{ matrix.version }} -m
 
-          sleep 120  # Wait for containers to initialize
           echo "Waiting for Apache Cloudberry multi-node deployment to complete..."
-          timeout 120 bash -c 'until docker logs cbdb-cdw 2>&1 | grep -q "DEPLOYMENT SUCCESSFUL"; do sleep 15; echo "Still waiting for multi-node deployment..."; done'
-          echo "Multi-node deployment completed successfully!"
+          for i in {1..24}; do
+            if docker logs cbdb-cdw 2>&1 | grep -q "DEPLOYMENT SUCCESSFUL"; then
+              echo "✅ Multi-node deployment completed"
+              break
+            fi
+            if ! docker ps | grep -q cbdb-cdw; then
+              echo "ERROR: Coordinator stopped unexpectedly"
+              docker logs cbdb-cdw
+              exit 1
+            fi
+            [ $i -eq 24 ] && echo "ERROR: Deployment timeout" && docker logs cbdb-cdw && exit 1
+            sleep 15
+          done
         timeout-minutes: 30
 
       - name: Test SSH connectivity between nodes
         run: |
           echo "Testing SSH connectivity from coordinator to all nodes..."
-          docker exec -u gpadmin cbdb-cdw bash -l -c "ssh -o StrictHostKeyChecking=no sdw1 'hostname'"
-          docker exec -u gpadmin cbdb-cdw bash -l -c "ssh -o StrictHostKeyChecking=no sdw2 'hostname'"
-          docker exec -u gpadmin cbdb-cdw bash -l -c "ssh -o StrictHostKeyChecking=no scdw 'hostname'"
-          echo "✅ SSH connectivity test passed"
+          docker exec -u gpadmin cbdb-cdw bash -l -c "ssh -o StrictHostKeyChecking=no -o ConnectTimeout=10 sdw1 'hostname'"
+          docker exec -u gpadmin cbdb-cdw bash -l -c "ssh -o StrictHostKeyChecking=no -o ConnectTimeout=10 sdw2 'hostname'"
+          docker exec -u gpadmin cbdb-cdw bash -l -c "ssh -o StrictHostKeyChecking=no -o ConnectTimeout=10 scdw 'hostname'"
+          echo "✅ SSH connectivity verified"
 
       - name: Test database operations
         run: |
-          echo "Waiting for database readiness (up to 60s)..."
-          for i in {1..12}; do
-            if docker exec -u gpadmin cbdb-cdw bash -l -c "pg_isready -h localhost -p 5432 -U gpadmin" >/dev/null 2>&1; then
-              echo "Database is ready."
-              break
-            fi
-            if [ "$i" -eq 12 ]; then
-              echo "Database did not become ready in time."
-              exit 1
-            fi
+          echo "Waiting for database readiness..."
+          for i in {1..24}; do
+            docker exec -u gpadmin cbdb-cdw bash -l -c "pg_isready -h localhost -p 5432 -U gpadmin" >/dev/null 2>&1 && break
+            [ $i -eq 24 ] && echo "ERROR: Database not ready" && exit 1
             sleep 5
           done
 
@@ -218,13 +231,16 @@ jobs:
           docker logs --tail 50 cbdb-sdw2 || true
           echo "=== Network Status ==="
           docker network ls || true
+          echo "=== Disk Status ==="
+          df -h
 
       - name: Cleanup
         if: always()
         run: |
           cd devops/sandbox
-          docker compose -f docker-compose-rockylinux9.yml down -v || true
-          docker system prune -af || true
+          docker compose -f docker-compose-rockylinux9.yml down -v --remove-orphans || true
+          docker network rm cbdb-interconnect || true
+          docker system prune -af --volumes || true
 
   notify-results:
     name: Notify Results