Summary

When E2E tests fail in CI, we lack the information needed to diagnose root causes. A recent api-clients failure returned HTTP 500 instead of 201 during workload creation, but we couldn't determine why because:

1. Response bodies aren't logged on failure — tests assert status codes but don't dump the response body when assertions fail
2. thv serve logs aren't captured — the API server's stdout/stderr isn't saved as a CI artifact
3. Ginkgo report files aren't generated — the upload step consistently finds no files at test/e2e/ginkgo-report.xml

Proposed improvements

1. Log response bodies on assertion failure

Test helpers like createWorkload, registerClient, etc. should log the response body when the status code doesn't match expectations. Example pattern:

resp := createWorkload(apiServer, workloadReq)
body, _ := io.ReadAll(resp.Body)
resp.Body.Close()
Expect(resp.StatusCode).To(Equal(http.StatusCreated),
    "Unexpected status code. Response body: %s", string(body))

2. Capture thv serve logs as CI artifacts

The E2E test runner or helpers should redirect thv serve stderr/stdout to a log file, and the CI workflow should upload it alongside test results.

3. Fix Ginkgo report generation

The workflow expects test/e2e/ginkgo-report.xml and test/e2e/junit-report.xml but they're never created. Either:

• Pass --junit-report=junit-report.xml to ginkgo in run_tests.sh
• Or remove the upload step if reports aren't needed

4. Consider adding --vv output on failure

When a test fails, having verbose Ginkgo output (--vv) would show the full step timeline, making it easier to trace what happened.

Context

• Example failure: https://github.com/stacklok/toolhive/actions/runs/22908568453/job/66475961041
• The test should successfully register client with default group got HTTP 500 from createWorkload after ~60 seconds (likely a container image pull timeout), but we couldn't confirm without server logs or the error response body.

Generated with Claude Code