cleanup + concurrency test

This commit is contained in:
STJ 2026-03-16 19:41:42 -07:00
parent 48ea4d15d6
commit 0a5d0f288b
5 changed files with 87 additions and 33 deletions

View file

@ -92,12 +92,6 @@ class BrowserSession:
await self.browser.start()
# [info] Really annoying discovery: despite being isolated in
# different internal sessions, browser use does not
# *actually* isolate the cookies and internals.
#
# [fix] We use CDP browser contexts manually to make the
# sessions unique per CDP connection. Nothing groundbreaking
cdp = self.browser.cdp_client
ctx = await cdp.send.Target.createBrowserContext(params={"disposeOnDetach": True})
self.browser_context_id = ctx["browserContextId"]
@ -116,13 +110,16 @@ class BrowserSession:
target_id = await _scoped(new_window=True)
await self.browser.get_or_create_cdp_session(target_id, focus=True)
async def close(self) -> None:
async def dispose_context(self) -> None:
if self.browser_context_id and self.browser and self.browser.is_cdp_connected:
with contextlib.suppress(Exception):
# [fix] properly dispose the session on closure.
await self.browser.cdp_client.send.Target.disposeBrowserContext(
params={"browserContextId": self.browser_context_id},
)
self.browser_context_id = None
async def close(self) -> None:
await self.dispose_context()
await _close_browser(self.browser)
self.browser = None

View file

@ -39,7 +39,6 @@ def _preflight():
_preflight()
# in pretty mode, silence all loggers and suppress pytest's own terminal output
if ui.is_pretty():
for _name in (
"strix.tests.integration",
@ -147,10 +146,12 @@ def browsers(agent_state, request):
if "error" in result:
pytest.fail(f"Browser launch failed for {aid}: {result}")
yield [Browser(aid) for aid in agent_ids]
yield [Browser(aid, agent_state) for aid in agent_ids]
for aid in agent_ids:
_manager.sessions.pop(aid, None)
session = _manager.remove(aid)
if session:
_run_in_bg(session.dispose_context())
set_current_agent_id(_SESSION_AGENT_ID)

View file

@ -18,8 +18,9 @@ def setup_screenshots_dir():
class Browser:
def __init__(self, agent_id):
def __init__(self, agent_id, agent_state=None):
self._agent_id = agent_id
self._agent_state = agent_state
def __getattr__(self, action):
import asyncio
@ -32,12 +33,17 @@ class Browser:
def call(**kwargs):
async def _run():
set_current_agent_id(self._agent_id)
return await browser_action(agent_state=None, action=action, **kwargs)
return await browser_action(
agent_state=self._agent_state,
action=action,
**kwargs,
)
future = asyncio.run_coroutine_threadsafe(_run(), _bg_loop)
result = future.result(timeout=120)
if "error" in result:
Fail(result).error(result["error"])
_strip_screenshot(result, _caller_test_name())
return result
return call
@ -62,20 +68,20 @@ def _caller_test_name():
return "unknown"
def _save_screenshot(result, name):
b64 = result.get("screenshot")
def _strip_screenshot(result, name):
b64 = result.pop("screenshot", None)
if not b64 or not isinstance(b64, str) or len(b64) < 100:
return None
return
path = SCREENSHOTS_DIR / f"{name}.png"
path.write_bytes(base64.b64decode(b64))
return str(path)
result["screenshot_path"] = str(path)
class Fail:
def __init__(self, result=None):
self._result = result
self._name = _caller_test_name()
self._screenshot = _save_screenshot(result, self._name) if result else None
self._screenshot = result.get("screenshot_path") if result else None
def expected(self, value):
self._expected = value

View file

@ -1,5 +1,4 @@
import base64
import binascii
from pathlib import Path
import pytest
@ -89,15 +88,13 @@ def test_screenshot(browser):
browser.navigate(url="https://example.com")
result = browser.screenshot()
screenshot = result.get("screenshot", "")
ui.log(f"screenshot → {len(screenshot)} bytes base64")
path = result.get("screenshot_path")
ui.log(f"screenshot → {path}")
if len(screenshot) <= 100:
Fail(result).expected("> 100 bytes").got(f"{len(screenshot)} bytes")
try:
base64.b64decode(screenshot)
except (ValueError, binascii.Error) as e:
Fail(result).error(f"invalid base64: {e}")
if not path:
Fail(result).error("no screenshot saved")
elif not Path(path).exists():
Fail(result).error(f"screenshot file missing: {path}")
def test_input(browser):

View file

@ -1,13 +1,13 @@
import pytest
from . import console as ui
from .helpers import Fail, act_parallel
from .helpers import Browser, Fail, act_parallel
pytestmark = [pytest.mark.integration, pytest.mark.browsers(2)]
def test_session_objects_are_distinct(browsers):
def test_session_objects_are_distinct(browsers: list[Browser]) -> None:
from strix.tools.browser.browser_manager import _manager
a, b = browsers
@ -25,7 +25,7 @@ def test_session_objects_are_distinct(browsers):
)
def test_parallel_navigate_isolated(browsers):
def test_parallel_navigate_isolated(browsers: list[Browser]) -> None:
a, b = browsers
ui.status("isolation → navigating both agents concurrently")
result_a, result_b = act_parallel(
@ -55,7 +55,7 @@ def test_parallel_navigate_isolated(browsers):
Fail(state_b).expected("agent B still on iana.org").got(state_b.get("url"))
def test_cookie_isolation(browsers):
def test_cookie_isolation(browsers: list[Browser]) -> None:
a, b = browsers
a.navigate(url="https://example.com")
a.evaluate(code="document.cookie = 'agent=A; path=/'")
@ -76,7 +76,7 @@ def test_cookie_isolation(browsers):
Fail(result_a).expected("cookie 'agent=A' present").got(cookie_a)
def test_concurrent_mixed_actions(browsers):
def test_concurrent_mixed_actions(browsers: list[Browser]) -> None:
a, b = browsers
a.navigate(url="https://example.com")
b.navigate(url="https://www.iana.org")
@ -94,3 +94,56 @@ def test_concurrent_mixed_actions(browsers):
for r in results:
if "error" in r:
Fail(r).error(r["error"])
def test_session_close_does_not_affect_other(browsers: list[Browser]) -> None:
a, b = browsers
a.navigate(url="https://example.com")
b.navigate(url="https://www.iana.org")
a.close_browser()
ui.log("agent A closed")
result = b.navigate(url="https://example.com")
ui.log(f"agent B navigation result: {result}")
if "error" in result:
Fail(result).error(f"agent B broken after A closed: {result['error']}")
if "example.com" not in result.get("url", ""):
Fail(result).expected("url containing 'example.com'").got(result.get("url"))
def test_relaunch_parallel_no_side_effects(browsers: list[Browser]) -> None:
a, b = browsers
a.navigate(url="https://example.com")
b.navigate(url="https://www.iana.org")
# close A, verify B is unaffected
a.close_browser()
b_check = b.screenshot()
if "iana.org" not in b_check.get("url", ""):
Fail(b_check).expected("agent B still on iana.org").got(b_check.get("url"))
# relaunch A
a.launch()
a.navigate(url="https://example.com")
# parallel actions on both after relaunch
results = act_parallel(
[
(a, {"action": "evaluate", "code": "document.title"}),
(b, {"action": "evaluate", "code": "document.title"}),
(a, {"action": "screenshot"}),
(b, {"action": "screenshot"}),
]
)
for r in results:
if "error" in r:
Fail(r).error(r["error"])
# verify no cross-contamination after relaunch (use url, titles vary)
a_state, b_state = results[2], results[3]
if "example.com" not in a_state.get("url", ""):
Fail(a_state).expected("agent A on example.com").got(a_state.get("url"))
if "iana.org" not in b_state.get("url", ""):
Fail(b_state).expected("agent B on iana.org").got(b_state.get("url"))