fix(kubernetes): Concurrent request failures from shared ApiClient

code-interpreter/tests/e2e/test_concurrent_execute.py

@@ -0,0 +1,93 @@
+"""Reproduce: concurrent /v1/execute requests fail with "Handshake status 200 OK".
+
+The KubernetesExecutor uses a single CoreV1Api client for both REST and
+streaming operations.  stream.stream() temporarily monkey-patches the
+shared api_client.request to use WebSocket.  Under concurrent load, a
+REST call from one request can land during another request's patch window,
+causing a WebSocket handshake against a non-WebSocket endpoint.
+
+This test fires multiple concurrent requests at the code interpreter and
+asserts that all succeed.  With the current bug, at least one will fail
+with an error containing "Handshake status" or a 500 status.
+
+After the fix (separate ApiClient instances for REST vs streaming), all
+requests should succeed.
+"""
+
+from __future__ import annotations
+
+import concurrent.futures
+from typing import Any, Final
+
+import httpx
+import pytest
+
+BASE_URL: Final[str] = "http://localhost:8000"
+# Number of concurrent requests — enough to reliably trigger the race.
+CONCURRENCY: Final[int] = 5
+
+
+def _execute_request(index: int) -> dict[str, Any]:
+    """Send a single /v1/execute request and return the parsed result.
+
+    Raises on transport errors or non-200 status so the caller can
+    collect failures.
+    """
+    timeout = httpx.Timeout(60.0, connect=10.0)
+    with httpx.Client(base_url=BASE_URL, timeout=timeout) as client:
+        response = client.post(
+            "/v1/execute",
+            json={
+                "code": f"print('request {index}')",
+                "timeout_ms": 30000,
+            },
+        )
+        response.raise_for_status()
+        return {"index": index, "result": response.json()}
+
+
+def test_concurrent_execute_requests_all_succeed() -> None:
+    """Fire N concurrent /v1/execute requests.
+
+    With the shared-client bug, overlapping stream.stream() calls cause
+    REST calls to be routed through the WebSocket path, producing errors
+    like "Handshake status 200 OK".
+
+    All N requests must return exit_code == 0 for this test to pass.
+    """
+    # Verify the service is reachable first
+    timeout = httpx.Timeout(10.0, connect=5.0)
+    with httpx.Client(base_url=BASE_URL, timeout=timeout) as client:
+        try:
+            health = client.get("/health")
+        except httpx.TransportError as exc:
+            pytest.fail(f"Code interpreter not reachable at {BASE_URL}: {exc}")
+        assert health.status_code == 200 and health.json()["status"] == "ok"
+
+    # Fire concurrent requests
+    results: list[dict[str, Any]] = []
+    errors: list[str] = []
+
+    with concurrent.futures.ThreadPoolExecutor(max_workers=CONCURRENCY) as pool:
+        futures = {pool.submit(_execute_request, i): i for i in range(CONCURRENCY)}
+
+        for future in concurrent.futures.as_completed(futures):
+            idx = futures[future]
+            try:
+                results.append(future.result())
+            except Exception as exc:
+                errors.append(f"request {idx}: {exc}")
+
+    # Report all failures together for easier debugging
+    assert not errors, f"{len(errors)}/{CONCURRENCY} concurrent requests failed:\n" + "\n".join(
+        errors
+    )
+
+    # Every successful response should have exit_code == 0
+    for r in results:
+        result = r["result"]
+        assert result["exit_code"] == 0, (
+            f"request {r['index']} failed: "
+            f"stdout={result.get('stdout')!r} "
+            f"stderr={result.get('stderr')!r}"
+        )