stacklok / toolhive / 20781514468

50%

Build Type

push

github

Committed by web-flow

Commit Message

Add K8s manager infrastructure for vMCP dynamic backend discovery (#3192)

* Add K8s manager infrastructure for vMCP dynamic backend discovery

Implements controller-runtime manager wrapper for VirtualMCP servers
running in Kubernetes with dynamic backend discovery mode. This provides
the foundation for live backend updates without pod restarts.

Closes: #3000

* fixes from review

* fix ci

* Fix flaky auth discovery E2E test by adding pod readiness check

The "discovered auth behavior with real MCP requests" test was failing
intermittently on k8s 1.34.3 with "connection reset by peer" errors.

Root cause:
The nested Context's BeforeAll only checked that the Service had a NodePort
assigned, but did not verify the VirtualMCPServer pod was still running and
ready. Between the outer BeforeAll completing and the nested test executing,
the pod could have crashed or restarted, especially on k8s 1.34.3.

Fix:
Add explicit pod readiness checks in the nested BeforeAll to ensure:
1. VirtualMCPServer status is Ready
2. vMCP pods are running and ready
3. Only then proceed to get NodePort and run tests

This prevents the test from attempting HTTP connections to pods that may
have crashed or are in CrashLoopBackOff state.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

* Fixes from review

---------

Co-authored-by: taskbot <taskbot@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>

Run Details

98 of 186 new or added lines in 4 files covered. (52.69%)

4 existing lines in 2 files now uncovered.

28866 of 57145 relevant lines covered (50.51%)

59.04 hits per line