fix(ui): revalidate dashboard HTML on every load and answer /health/readiness

The UI image sent its HTML with no Cache-Control while the content-hashed
_next bundles are immutable for a year, so a browser could keep stale HTML
across a deploy and request chunk hashes the new image no longer ships,
failing with 404 until a hard reload. The HTML locations now send
Cache-Control: no-cache so the browser revalidates against ETag on each load.

nginx also answers 200 on /health/readiness, the path the gateway exposes,
so an ingress-wide ALB health check on that path no longer marks the UI
target unhealthy

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

tests/proxy_migration_tests/test_ui_image_serves_offline.py

@@ -18,6 +18,7 @@ Requires a working docker CLI.
 """
 
 import os
+import re
 import shutil
 import subprocess
 import time
@@ -96,6 +97,17 @@ def _probe(network: str, container: str, path: str) -> "subprocess.CompletedProc
     )
 
 
+def _response_headers(network: str, container: str, path: str) -> dict[str, str]:
+    head = _docker(
+        "run", "--rm", "--network", network, CURL_IMAGE,
+        "--silent", "--show-error", "--max-time", "10", "--head",
+        f"http://{container}:{UI_PORT}{path}",
+        check=False,
+    )
+    header_lines = (line for line in head.stdout.splitlines() if ":" in line)
+    return {name.strip().lower(): value.strip() for name, value in (line.split(":", 1) for line in header_lines)}
+
+
 def test_ui_serves_as_arbitrary_uid_read_only(ui_container: tuple[str, str]) -> None:
     """nginx boots and serves as an arbitrary uid with a read-only root fs.
 
@@ -130,3 +142,41 @@ def test_ui_serves_as_arbitrary_uid_read_only(ui_container: tuple[str, str]) ->
             f"GET {path} returned {page.stdout.strip()!r} as uid {ARBITRARY_UID}.\n"
             f"{_container_logs(container)}"
         )
+
+
+def test_readiness_path_answers_200_like_the_gateway(ui_container: tuple[str, str]) -> None:
+    network, container = ui_container
+
+    readiness = _probe(network, container, "/health/readiness")
+    assert readiness.stdout.strip() == "200", (
+        f"GET /health/readiness returned {readiness.stdout.strip()!r}; an ingress-wide ALB health check on "
+        f"that path marks the UI target unhealthy.\n{_container_logs(container)}"
+    )
+
+
+def test_html_revalidates_while_hashed_bundles_stay_immutable(ui_container: tuple[str, str]) -> None:
+    network, container = ui_container
+
+    for path in ("/ui", "/ui/", "/ui/login", "/"):
+        headers = _response_headers(network, container, path)
+        assert headers.get("cache-control") == "no-cache", (
+            f"GET {path} sent Cache-Control {headers.get('cache-control')!r}; a browser that keeps this HTML "
+            f"requests the previous deploy's chunk hashes after a rollout.\n{_container_logs(container)}"
+        )
+
+    html = _docker(
+        "run", "--rm", "--network", network, CURL_IMAGE,
+        "--silent", "--show-error", "--max-time", "10",
+        f"http://{container}:{UI_PORT}/ui/",
+        check=False,
+    )
+    chunk_paths = tuple(
+        match.group(1) for match in re.finditer(r'src="(/litellm-asset-prefix/_next/static/[^"]+\.js)"', html.stdout)
+    )
+    assert chunk_paths, f"the UI HTML references no /litellm-asset-prefix/_next/static/*.js bundle:\n{html.stdout[:2000]}"
+
+    chunk_headers = _response_headers(network, container, chunk_paths[0])
+    assert "immutable" in chunk_headers.get("cache-control", ""), (
+        f"GET {chunk_paths[0]} sent Cache-Control {chunk_headers.get('cache-control')!r}; hashed bundles must "
+        f"stay cacheable.\n{_container_logs(container)}"
+    )

ui/nginx.conf

@@ -70,8 +70,10 @@ http {
       expires 1d;
     }
 
-    # Probe target — doesn't depend on disk.
-    location = /healthz { default_type text/plain; return 200 "ok\n"; }
+    # Probe targets — don't depend on disk. /health/readiness mirrors the
+    # gateway's path so one ingress-wide ALB health check covers the UI too.
+    location = /healthz          { default_type text/plain; return 200 "ok\n"; }
+    location = /health/readiness { default_type text/plain; return 200 "ok\n"; }
 
     # Next.js App Router (output: "export") emits an RSC/flight payload
     # as <route>.txt next to <route>.html, plus __next.*.txt segment
@@ -95,14 +97,18 @@ http {
     # proxy_server: serve /ui/<page> from out/<page>.html, with App
     # Router-aware fallback (out/<page>/index.html) and a final SPA
     # fallback to out/index.html for client-side routes.
-    location = /ui  { try_files /index.html =404; }
-    location = /ui/ { try_files /index.html =404; }
+    # The HTML is the manifest for the content-hashed bundles above, so it
+    # must revalidate on every load (ETag/Last-Modified) or a browser keeps
+    # asking for chunks the previous deploy shipped and gets 404s.
+    location = /ui  { add_header Cache-Control "no-cache"; try_files /index.html =404; }
+    location = /ui/ { add_header Cache-Control "no-cache"; try_files /index.html =404; }
     location ~ ^/ui/(.+)$ {
+      add_header Cache-Control "no-cache";
       try_files /$1.html /$1/index.html /index.html =404;
     }
 
     # `/` is handy for direct-debug port-forwards.
-    location = / { try_files /index.html =404; }
+    location = / { add_header Cache-Control "no-cache"; try_files /index.html =404; }
 
     # Anything else (API calls etc.) returns 404 from the UI's
     # perspective. A reverse proxy in front of this image routes the