stacklok / toolhive-studio / 26233238545

70%

main: 70%

Pull #2282

github

test(e2e): use LONG_TIMEOUT on registry install/uninstall flow

The `install and uninstall server from registry` test consistently
times out on CI at the `getByRole('link', { name: /^view$/i }).click()`
step (default 30s action timeout). Between clicking the dialog's
`Install server` and the "View" link rendering, the backend has to:

1. pull `docker.io/mcp/everything:latest` (cold cache on most runners),
2. start the egress squid proxy (the server's manifest declares no
   outbound network, so a proxy container is required),
3. start the workload container,
4. wait for it to become healthy.

On the default CI shared runners that chain regularly exceeds 30s,
producing a flaky `TimeoutError: locator.click: Timeout 30000ms
exceeded` on `view`. The fixture already exports `LONG_TIMEOUT`
(180s) for "operations that may take a while, like LLM inference on
CI" — the registry-install path is in the same class of slow
container-orchestration operations, so reuse it for the two waits
that depend on the spawned containers being up:

  - the click on the "View" link (rendered once install completes),
  - the `Running` text appearing on the workload detail page.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

5194 of 7997 branches covered (64.95%)

7706 of 10933 relevant lines covered (70.48%)

114.91 hits per line