ci: fail fast on kaniko job failure with logs, longer job TTL for post-mortem

Co-authored-by: Cursor <cursoragent@cursor.com>

.gitea/workflows/build-deploy.yaml

@@ -44,6 +44,33 @@ jobs:
           echo "IMAGE_TAG=${TAG}" >> "$GITHUB_ENV"
           echo "Build tag: ${TAG}"
 
+      - name: Define job waiter
+        shell: sh
+        run: |
+          # wait_for_job <name>: exit 0 on Complete, exit 1 (with kaniko logs) on Failed/timeout
+          cat > wait_for_job.sh <<'ENDSCRIPT'
+          #!/bin/sh
+          JOB="$1"
+          DEADLINE=$(( $(date +%s) + 2400 ))
+          while :; do
+            CONDS="$(kubectl -n ${BUILD_NS} get job/${JOB} -o jsonpath='{range .status.conditions[*]}{.type}={.status} {end}' 2>/dev/null)"
+            case "$CONDS" in
+              *Complete=True*) echo "Job ${JOB} completed"; exit 0 ;;
+              *Failed=True*)
+                echo "Job ${JOB} FAILED — kaniko logs:"
+                kubectl -n ${BUILD_NS} logs job/${JOB} -c kaniko --tail=100 || true
+                exit 1 ;;
+            esac
+            if [ "$(date +%s)" -gt "$DEADLINE" ]; then
+              echo "Timed out waiting for job ${JOB} — kaniko logs:"
+              kubectl -n ${BUILD_NS} logs job/${JOB} -c kaniko --tail=100 || true
+              exit 1
+            fi
+            sleep 15
+          done
+          ENDSCRIPT
+          chmod +x wait_for_job.sh
+
       - name: Build backend image (Kaniko Job)
         shell: sh
         run: |
@@ -55,7 +82,7 @@ jobs:
             name: ${JOB_NAME}
             namespace: ${BUILD_NS}
           spec:
-            ttlSecondsAfterFinished: 600
+            ttlSecondsAfterFinished: 3600
             backoffLimit: 0
             template:
               spec:
@@ -104,11 +131,7 @@ jobs:
                           path: config.json
           ENDJOB
           echo "Waiting for backend build job: ${JOB_NAME}"
-          if ! kubectl -n ${BUILD_NS} wait --for=condition=complete job/${JOB_NAME} --timeout=40m; then
-            echo "--- kaniko logs (tail) ---"
-            kubectl -n ${BUILD_NS} logs job/${JOB_NAME} -c kaniko --tail=100 || true
-            exit 1
-          fi
+          ./wait_for_job.sh "${JOB_NAME}"
           echo "Backend build done"
 
       - name: Build frontend image (Kaniko Job)
@@ -122,7 +145,7 @@ jobs:
             name: ${JOB_NAME}
             namespace: ${BUILD_NS}
           spec:
-            ttlSecondsAfterFinished: 600
+            ttlSecondsAfterFinished: 3600
             backoffLimit: 0
             template:
               spec:
@@ -168,11 +191,7 @@ jobs:
                           path: config.json
           ENDJOB
           echo "Waiting for frontend build job: ${JOB_NAME}"
-          if ! kubectl -n ${BUILD_NS} wait --for=condition=complete job/${JOB_NAME} --timeout=40m; then
-            echo "--- kaniko logs (tail) ---"
-            kubectl -n ${BUILD_NS} logs job/${JOB_NAME} -c kaniko --tail=100 || true
-            exit 1
-          fi
+          ./wait_for_job.sh "${JOB_NAME}"
           echo "Frontend build done"
 
       - name: Update GitOps repo and push