ci: add upterm debugging as an alternative to tmate

This enables interactive debugging of ci checks
with the upterm tool as an alternative to the tmate tool which does not
work entirely anymore.
see https://github.com/rook/rook/issues/16131

addition to tmate, upterm debugging is now enabled via the debug-ci label on
PRs

Signed-off-by: Michael Adam <obnox@samba.org>

.github/workflows/upterm_debug/action.yml

@@ -0,0 +1,29 @@
+name: "Upterm debugging tests"
+description: "Setup an upterm session for interactive debugging of failing tests"
+inputs:
+  debug-ci:
+    description: "boolean for debug-ci label in PR"
+    required: false
+runs:
+  using: "composite"
+  steps:
+    - name: consider upterm debugging
+      shell: bash --noprofile --norc -eo pipefail -x {0}
+      if: runner.debug || inputs.debug-ci == 'true'
+      run: |
+        # Enable upterm only in the main Rook repo, where the USE_TMATE secret is set in the repo, or if the action is re-run
+        if [ "$GITHUB_REPOSITORY_OWNER" = "rook" ] || [ "$GITHUB_RUN_ATTEMPT" -gt 1 ]; then
+          echo ENABLE_UPTERM=1 >> $GITHUB_ENV
+        fi
+
+    - name: set up upterm session
+      if: ${{ env.ENABLE_UPTERM }}
+      uses: owenthereal/action-upterm@v1
+      with:
+        # detached: true
+        ## whether to limit ssh access and adds the ssh public key for the user which triggered the workflow:
+        ## detached is not implemented (yet?)
+        ## see: https://github.com/owenthereal/action-upterm/issues/12
+        limit-access-to-actor: false
+        ## If no one connects within five minutes, shut down the server.
+        wait-timeout-minutes: 5