ci(e2e): add run-id label to infra resources for cleanup

Signed-off-by: Anton Yachmenev <[email protected]>

Author: yachmenevas

.github/workflows/e2e-matrix.yml

@@ -177,17 +177,23 @@ jobs:
         uses: actions/download-artifact@v4
         with:
           name: nested-run-${{ env.RUN_ID }}
-          path: .
+          path: ci/dvp-e2e/tmp/runs/${{ env.RUN_ID }}
+          merge-multiple: true
 
       - name: Configure kubeconfig env
+        working-directory: ci/dvp-e2e
+        env:
+          RUN_ID: ${{ env.RUN_ID }}
         run: |
-          set -euo pipefail
-          NESTED_KUBECONFIG="${GITHUB_WORKSPACE}/ci/dvp-e2e/tmp/runs/${RUN_ID}/nested/kubeconfig"
-          if [ ! -s "$NESTED_KUBECONFIG" ]; then
-            echo "[ERR] Nested kubeconfig not found at $NESTED_KUBECONFIG" >&2
-            exit 1
-          fi
-          echo "KUBECONFIG=$NESTED_KUBECONFIG" >> "$GITHUB_ENV"
+          task ci:kubeconfig:ensure
+
+      - name: Pause for manual inspection
+        working-directory: ci/dvp-e2e
+        env:
+          RUN_ID: ${{ env.RUN_ID }}
+          MANUAL_WAIT_SECONDS: ${{ vars.MANUAL_WAIT_SECONDS || '7200' }}
+        run: |
+          task ci:manual-wait
 
       - name: Run E2E tests
         working-directory: test/e2e

ci/dvp-e2e/Taskfile.yaml

@@ -623,3 +623,71 @@ tasks:
           TMP_DIR: "{{ .TMP_DIR }}"
           NESTED_KUBECONFIG: "{{ .NESTED_KUBECONFIG }}"
           SDS_SC_NAME: "{{ .SDS_SC_NAME }}"
+  ci:kubeconfig:ensure:
+    desc: Ensure nested kubeconfig exists at the expected path
+    vars:
+      RUN_ID: '{{ .RUN_ID | default (env "RUN_ID") | default "" }}'
+      WORKSPACE:
+        sh: git rev-parse --show-toplevel 2>/dev/null || pwd
+    cmds:
+      - |
+        set -euo pipefail
+        RUN_ID="{{ .RUN_ID }}"
+        if [ -z "$RUN_ID" ]; then
+          echo "[ERR] RUN_ID must be provided to locate nested kubeconfig" >&2
+          exit 1
+        fi
+        WORKSPACE="${GITHUB_WORKSPACE:-{{ .WORKSPACE }}}"
+        TARGET_PATH="$WORKSPACE/ci/dvp-e2e/tmp/runs/$RUN_ID/nested/kubeconfig"
+        if [ ! -s "$TARGET_PATH" ]; then
+          echo "[ERR] Nested kubeconfig not found at $TARGET_PATH" >&2
+          exit 1
+        fi
+        echo "[INFO] Using nested kubeconfig at $TARGET_PATH"
+        [ -n "${GITHUB_ENV:-}" ] && echo "KUBECONFIG=$TARGET_PATH" >> "$GITHUB_ENV"
+  ci:manual-wait:
+    desc: Pause execution to allow manual SSH inspection of nested cluster
+    vars:
+      RUN_ID: '{{ .RUN_ID | default (env "RUN_ID") | default "" }}'
+      WORKSPACE:
+        sh: git rev-parse --show-toplevel 2>/dev/null || pwd
+      WAIT_SECONDS: '{{ .WAIT_SECONDS | default (env "MANUAL_WAIT_SECONDS") | default "7200" }}'
+    cmds:
+      - |
+        set -euo pipefail
+        RUN_ID="{{ .RUN_ID }}"
+        WAIT="{{ .WAIT_SECONDS }}"
+        if [ -z "$RUN_ID" ]; then
+          echo "[ERR] RUN_ID must be set for ci:manual-wait" >&2
+          exit 1
+        fi
+        if ! [[ "$WAIT" =~ ^[0-9]+$ ]]; then
+          echo "[ERR] WAIT_SECONDS must be numeric (got '$WAIT')" >&2
+          exit 1
+        fi
+        if [ "$WAIT" -le 0 ]; then
+          echo "[INFO] Manual wait skipped (WAIT_SECONDS=$WAIT)"
+          exit 0
+        fi
+        WORKSPACE='{{ .WORKSPACE }}'
+        PARENT_KUBECONFIG="$WORKSPACE/ci/dvp-e2e/tmp/runs/$RUN_ID/parent.kubeconfig"
+        echo "[INFO] Pausing for $WAIT seconds before running tests."
+        echo "[INFO] Use parent kubeconfig for SSH tunneling:"
+        echo "       export KUBECONFIG=$PARENT_KUBECONFIG"
+        echo "       d8 v ssh --namespace $RUN_ID --username ubuntu <vm-name>"
+        echo "[INFO] Press Ctrl+C in the workflow run to cancel wait early."
+        START_TS=$(date +%s)
+        END=$((START_TS + WAIT))
+        LAST_NOTE=$START_TS
+        while true; do
+          NOW=$(date +%s)
+          [ "$NOW" -ge "$END" ] && break
+          REM=$((END - NOW))
+          printf '[INFO] Manual wait: %d seconds remaining...\n' "$REM"
+          if [ $((NOW - LAST_NOTE)) -ge 300 ]; then
+            echo "[INFO] Cluster should be ready for manual SSH troubleshooting."
+            LAST_NOTE=$NOW
+          fi
+          sleep 60 || true
+        done
+        echo "[INFO] Manual wait finished; proceeding to tests."