test: retry ambiguous setup response

Author: misunders2d

scripts/ci/ordinary-server-upgrade-e2e.sh

@@ -244,7 +244,7 @@ plan_setup() {
     server-agent setup --request "$INPUTS/server-setup.json"
 }
 
-apply_setup() {
+apply_setup_once() {
   local prefix="$1"
   local digest="$2"
   local output="$3"
@@ -257,6 +257,30 @@ apply_setup() {
     --expected-request-digest "$digest" --apply --start
 }
 
+apply_setup() {
+  local prefix="$1"
+  local digest="$2"
+  local output="$3"
+  local exit_code=0
+  apply_setup_once "$prefix" "$digest" "$output" || exit_code=$?
+  if [[ "$exit_code" -eq 0 ]]; then
+    return 0
+  fi
+  if ! jq -e '.status == "blocked" and .blocker == "systemd_start"' "$output" >/dev/null; then
+    return "$exit_code"
+  fi
+
+  # Public 0.1.50 can lose the systemctl response on a cold runner while the
+  # exact units still converge. Its documented recovery is a same-digest
+  # resume; retain the first response and prove that path rather than treating
+  # ambiguous transport status as a terminal setup failure.
+  mv "$output" "$output.systemd-response-loss"
+  if [[ -e "$output.stderr" ]]; then
+    mv "$output.stderr" "$output.systemd-response-loss.stderr"
+  fi
+  apply_setup_once "$prefix" "$digest" "$output"
+}
+
 psql_agentnet() {
   sudo -u agentnet psql -X -Atq --dbname=agentnet "$@"
 }