fix(ci): verify deployment socket and rollback image

scripts/deploy-dimensionlab-website.sh

@@ -241,7 +241,6 @@ container_running() {
 
 wait_for_container_restart() {
   local expected_image
-  local deadline
 
   if "$dry_run"; then
     log "DRY-RUN: would wait for $CONTAINER_NAME to run $latest_tag"
@@ -250,18 +249,27 @@ wait_for_container_restart() {
 
   expected_image="$(latest_image_id)"
   [ -n "$expected_image" ] || fail "could not resolve image id for $latest_tag"
+
+  wait_for_container_image "$expected_image" "new image" || fail "$CONTAINER_NAME did not restart on $latest_tag within ${DEPLOY_CONTAINER_START_TIMEOUT_SECONDS}s"
+}
+
+wait_for_container_image() {
+  local expected_image="$1"
+  local label="$2"
+  local deadline
+
   deadline=$((SECONDS + DEPLOY_CONTAINER_START_TIMEOUT_SECONDS))
 
   while [ "$SECONDS" -lt "$deadline" ]; do
     if container_running && [ "$(container_image_id)" = "$expected_image" ]; then
-      log "$CONTAINER_NAME is running the new image"
+      log "$CONTAINER_NAME is running the $label"
       return 0
     fi
 
     sleep 2
   done
 
-  fail "$CONTAINER_NAME did not restart on $latest_tag within ${DEPLOY_CONTAINER_START_TIMEOUT_SECONDS}s"
+  return 1
 }
 
 smoke_get() {
@@ -303,6 +311,8 @@ wait_for_smoke() {
 }
 
 rollback() {
+  local rollback_image
+
   if [ "$deployment_started" != "true" ] || [ -z "$rollback_tag" ] || [ "$rollback_done" = "true" ]; then
     return 0
   fi
@@ -310,8 +320,18 @@ rollback() {
   rollback_done=true
   rollback_in_progress=true
   printf '[deploy:%s] rolling back to %s\n' "$APP_NAME" "$rollback_tag" >&2
+  rollback_image="$("$container_cli" image inspect "$rollback_tag" --format '{{.Id}}' 2>/dev/null || true)"
   "$container_cli" tag "$rollback_tag" "$latest_tag" || true
-  restart_service || true
+  if container_running; then
+    restart_service || true
+  else
+    printf '[deploy:%s] waiting for %s to recover with rollback image\n' "$APP_NAME" "$SERVICE_NAME" >&2
+  fi
+  if [ -n "$rollback_image" ] && wait_for_container_image "$rollback_image" "rollback image"; then
+    printf '[deploy:%s] rollback image is running\n' "$APP_NAME" >&2
+  else
+    printf '[deploy:%s] ERROR: rollback image did not become healthy\n' "$APP_NAME" >&2
+  fi
   rollback_in_progress=false
 }