Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Frontend test: give the cold Vite SSR render in reasoning-source-render room on a loaded runner (#12903)

Daniel Han3000d91ac885
diff --git a/studio/frontend/tests/reasoning-source-render.test.ts b/studio/frontend/tests/reasoning-source-render.test.ts
index 860ee478fb7..59a6b986e15 100644
--- a/studio/frontend/tests/reasoning-source-render.test.ts
+++ b/studio/frontend/tests/reasoning-source-render.test.ts
@@ -8,6 +8,12 @@ import { promisify } from "node:util";
const execFileAsync = promisify(execFile);
+// Each render cold-starts Vite and transforms the Markdown component graph with no
+// prebundle: about 9 s on an idle Linux box. The Windows frontend job runs it beside
+// thousands of other tests and went past a 30 s cap, failing PRs that never touched
+// the renderer. The cap only stops a hung child; the assertions are what is tested.
+const RENDER_TIMEOUT_MS = 120_000;
+
// Vite's Markdown plugin graph retains close hooks after SSR. Run the real
// renderer in an isolated process so the assertion owns its complete lifetime
// and can terminate those hooks after the markup has been produced.
@@ -115,7 +121,7 @@ test("source Markdown renders in an assistant root scope without a message part"
{
cwd: process.cwd(),
maxBuffer: 1_000_000,
- timeout: 30_000,
+ timeout: RENDER_TIMEOUT_MS,
},
);
@@ -133,7 +139,7 @@ test("source Markdown suppresses a duplicate HTML card when render_html is prese
{
cwd: process.cwd(),
maxBuffer: 1_000_000,
- timeout: 30_000,
+ timeout: RENDER_TIMEOUT_MS,
},
);