{{ message }}
Repository navigation
Expand file tree
/
Copy pathtest_multi_client_e2e.py
More file actions
489 lines (409 loc) · 19.2 KB
/
Copy pathtest_multi_client_e2e.py
File metadata and controls
489 lines (409 loc) · 19.2 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
"""E2E Multi-Client Broadcast Tests
Tests that verify the protocol v3 broadcast model works correctly when
multiple clients are connected to the same CLI server session.
"""
import asyncio
import contextlib
import os
import shutil
import tempfile
import pytest
import pytest_asyncio
from pydantic import BaseModel, Field
from copilot import CopilotClient, RuntimeConnection, define_tool
from copilot.rpc import (
PermissionDecisionApproveOnce,
PermissionDecisionReject,
)
from copilot.session import PermissionHandler, PermissionNoResult
from copilot.tools import ToolInvocation
from .testharness import wait_for_event
from .testharness.proxy import CapiProxy
pytestmark = pytest.mark.asyncio(loop_scope="module")
class MultiClientContext:
"""Extended test context that manages two clients connected to the same CLI server."""
def __init__(self):
self.cli_path: str = ""
self.home_dir: str = ""
self.work_dir: str = ""
self.proxy_url: str = ""
self._proxy: CapiProxy | None = None
self._client1: CopilotClient | None = None
self._client2: CopilotClient | None = None
async def setup(self):
from .testharness.context import get_cli_path_for_tests
self.cli_path = get_cli_path_for_tests()
self.home_dir = os.path.realpath(tempfile.mkdtemp(prefix="copilot-multi-config-"))
self.work_dir = os.path.realpath(tempfile.mkdtemp(prefix="copilot-multi-work-"))
self._proxy = CapiProxy()
self.proxy_url = await self._proxy.start()
github_token = (
"fake-token-for-e2e-tests" if os.environ.get("GITHUB_ACTIONS") == "true" else None
)
# Client 1 uses TCP mode so a second client can connect to the same server
self._client1 = CopilotClient(
connection=RuntimeConnection.for_tcp(
path=self.cli_path, connection_token="py-tcp-shared-test-token"
),
working_directory=self.work_dir,
env=self.get_env(),
github_token=github_token,
)
# Trigger connection by creating and disconnecting an init session
init_session = await self._client1.create_session(
on_permission_request=PermissionHandler.approve_all
)
await init_session.disconnect()
# Read the actual port from client 1 and create client 2
actual_port = self._client1.runtime_port
assert actual_port is not None, "Client 1 should have an actual port after connecting"
self._client2 = CopilotClient(
connection=RuntimeConnection.for_uri(
f"localhost:{actual_port}", connection_token="py-tcp-shared-test-token"
)
)
async def teardown(self, test_failed: bool = False):
if self._client2:
try:
await self._client2.stop()
except Exception:
pass
self._client2 = None
if self._client1:
try:
await self._client1.stop()
except Exception:
pass
self._client1 = None
if self._proxy:
await self._proxy.stop(skip_writing_cache=test_failed)
self._proxy = None
if self.home_dir and os.path.exists(self.home_dir):
shutil.rmtree(self.home_dir, ignore_errors=True)
if self.work_dir and os.path.exists(self.work_dir):
shutil.rmtree(self.work_dir, ignore_errors=True)
async def configure_for_test(self, test_file: str, test_name: str):
import re
sanitized_name = re.sub(r"[^a-zA-Z0-9]", "_", test_name).lower()
# Use the same snapshot directory structure as the standard context
from .testharness.context import SNAPSHOTS_DIR
snapshot_path = SNAPSHOTS_DIR / test_file / f"{sanitized_name}.yaml"
abs_snapshot_path = str(snapshot_path.resolve())
if self._proxy:
await self._proxy.configure(abs_snapshot_path, self.work_dir)
# Clear temp directories between tests; tolerate Windows holding the
# SQLite session-store.db open briefly after the CLI subprocess exits.
from pathlib import Path
for base_dir in (self.home_dir, self.work_dir):
for item in Path(base_dir).iterdir():
if item.is_dir():
shutil.rmtree(item, ignore_errors=True)
else:
with contextlib.suppress(OSError):
item.unlink(missing_ok=True)
def get_env(self) -> dict:
env = os.environ.copy()
env.update(
{
"COPILOT_API_URL": self.proxy_url,
"COPILOT_HOME": self.home_dir,
"XDG_CONFIG_HOME": self.home_dir,
"XDG_STATE_HOME": self.home_dir,
}
)
return env
@property
def client1(self) -> CopilotClient:
if not self._client1:
raise RuntimeError("Context not set up")
return self._client1
@property
def client2(self) -> CopilotClient:
if not self._client2:
raise RuntimeError("Context not set up")
return self._client2
@pytest.hookimpl(tryfirst=True, hookwrapper=True)
def pytest_runtest_makereport(item, call):
outcome = yield
rep = outcome.get_result()
if rep.when == "call" and rep.failed:
item.session.stash.setdefault("any_test_failed", False)
item.session.stash["any_test_failed"] = True
@pytest_asyncio.fixture(scope="module", loop_scope="module")
async def mctx(request):
"""Multi-client test context fixture."""
context = MultiClientContext()
await context.setup()
yield context
any_failed = request.session.stash.get("any_test_failed", False)
await context.teardown(test_failed=any_failed)
@pytest_asyncio.fixture(autouse=True, loop_scope="module")
async def configure_multi_test(request, mctx):
"""Automatically configure the proxy for each test."""
module_name = request.module.__name__.split(".")[-1]
test_file = module_name[5:] if module_name.startswith("test_") else module_name
if test_file.endswith("_e2e"):
test_file = test_file[:-4] # Snapshot-folder compatibility with pre-rename layout
test_name = request.node.name
if test_name.startswith("test_"):
test_name = test_name[5:]
await mctx.configure_for_test(test_file, test_name)
yield
class TestMultiClientBroadcast:
async def test_both_clients_see_tool_request_and_completion_events(
self, mctx: MultiClientContext
):
"""Both clients see tool request and completion events."""
class SeedParams(BaseModel):
seed: str = Field(description="A seed value")
@define_tool("magic_number", description="Returns a magic number")
def magic_number(params: SeedParams, invocation: ToolInvocation) -> str:
return f"MAGIC_{params.seed}_42"
# Client 1 creates a session with a custom tool
session1 = await mctx.client1.create_session(
on_permission_request=PermissionHandler.approve_all, tools=[magic_number]
)
# Client 2 resumes with NO tools — should not overwrite client 1's tools
session2 = await mctx.client2.resume_session(
session1.session_id, on_permission_request=PermissionHandler.approve_all
)
waiters = []
try:
client1_requested = wait_for_event(
session1, lambda event: event.type.value == "external_tool.requested"
)
client2_requested = wait_for_event(
session2, lambda event: event.type.value == "external_tool.requested"
)
client1_completed = wait_for_event(
session1, lambda event: event.type.value == "external_tool.completed"
)
client2_completed = wait_for_event(
session2, lambda event: event.type.value == "external_tool.completed"
)
waiters = [client1_requested, client2_requested, client1_completed, client2_completed]
# Send a prompt that triggers the custom tool
# Use a longer timeout: first multi-client TCP test on Windows CI needs extra time
response = await session1.send_and_wait(
"Use the magic_number tool with seed 'hello' and tell me the result",
timeout=30.0,
)
assert response is not None
assert "MAGIC_hello_42" in (response.data.content or "")
# Both clients should have seen the external_tool.requested and completed events
await asyncio.gather(*waiters)
finally:
for waiter in waiters:
if not waiter.done():
waiter.cancel()
await asyncio.gather(*waiters, return_exceptions=True)
await session2.disconnect()
async def test_one_client_approves_permission_and_both_see_the_result(
self, mctx: MultiClientContext
):
"""One client approves a permission request and both see the result."""
permission_requests = []
# Client 1 creates a session and manually approves permission requests
session1 = await mctx.client1.create_session(
on_permission_request=lambda request, invocation: (
permission_requests.append(request) or PermissionDecisionApproveOnce()
),
)
# Client 2 observes the permission request but leaves the decision to client 1.
session2 = await mctx.client2.resume_session(
session1.session_id,
on_permission_request=lambda request, invocation: PermissionNoResult(),
)
waiters = []
try:
client1_requested = wait_for_event(
session1, lambda event: event.type.value == "permission.requested"
)
client2_requested = wait_for_event(
session2, lambda event: event.type.value == "permission.requested"
)
client1_completed = wait_for_event(
session1, lambda event: event.type.value == "permission.completed"
)
client2_completed = wait_for_event(
session2, lambda event: event.type.value == "permission.completed"
)
waiters = [client1_requested, client2_requested, client1_completed, client2_completed]
# Send a prompt that triggers a write operation (requires permission)
response = await session1.send_and_wait(
"Create a file called hello.txt containing the text 'hello world'", timeout=10.0
)
assert response is not None
assert response.data.content
# Client 1 should have handled permission requests
assert len(permission_requests) > 0
# Both clients should have seen permission.requested events
await asyncio.gather(client1_requested, client2_requested)
# Both clients should have seen permission.completed events with approved result
completed_events = await asyncio.gather(client1_completed, client2_completed)
for event in completed_events:
assert event.data.result.kind == "approved"
finally:
for waiter in waiters:
if not waiter.done():
waiter.cancel()
await asyncio.gather(*waiters, return_exceptions=True)
await session2.disconnect()
async def test_one_client_rejects_permission_and_both_see_the_result(
self, mctx: MultiClientContext
):
"""One client rejects a permission request and both see the result."""
# Client 1 creates a session and denies all permission requests
session1 = await mctx.client1.create_session(
on_permission_request=lambda request, invocation: PermissionDecisionReject(),
)
# Client 2 observes the permission request but leaves the decision to client 1.
session2 = await mctx.client2.resume_session(
session1.session_id,
on_permission_request=lambda request, invocation: PermissionNoResult(),
)
waiters = []
try:
client1_requested = wait_for_event(
session1, lambda event: event.type.value == "permission.requested"
)
client2_requested = wait_for_event(
session2, lambda event: event.type.value == "permission.requested"
)
client1_completed = wait_for_event(
session1, lambda event: event.type.value == "permission.completed"
)
client2_completed = wait_for_event(
session2, lambda event: event.type.value == "permission.completed"
)
session1_idle = wait_for_event(
session1, lambda event: event.type.value == "session.idle"
)
waiters = [
client1_requested,
client2_requested,
client1_completed,
client2_completed,
session1_idle,
]
# Create a file that the agent will try to edit
test_file = os.path.join(mctx.work_dir, "protected.txt")
with open(test_file, "w") as f:
f.write("protected content")
await session1.send("Edit protected.txt and replace 'protected' with 'hacked'.")
# Both clients should have seen permission.requested and permission.completed
await asyncio.gather(client1_requested, client2_requested)
# Both clients should see the denial
completed_events = await asyncio.gather(client1_completed, client2_completed)
for event in completed_events:
assert event.data.result.kind == "denied-interactively-by-user"
assert (await session1_idle).type.value == "session.idle"
# Verify the file was NOT modified (permission was denied)
with open(test_file) as f:
content = f.read()
assert content == "protected content"
finally:
for waiter in waiters:
if not waiter.done():
waiter.cancel()
await asyncio.gather(*waiters, return_exceptions=True)
await session2.disconnect()
@pytest.mark.timeout(90)
async def test_two_clients_register_different_tools_and_agent_uses_both(
self, mctx: MultiClientContext
):
"""Two clients register different tools and agent uses both."""
class CountryCodeParams(BaseModel):
model_config = {"populate_by_name": True}
country_code: str = Field(alias="countryCode", description="A two-letter country code")
@define_tool("city_lookup", description="Returns a city name for a given country code")
def city_lookup(params: CountryCodeParams, invocation: ToolInvocation) -> str:
return f"CITY_FOR_{params.country_code}"
@define_tool("currency_lookup", description="Returns a currency for a given country code")
def currency_lookup(params: CountryCodeParams, invocation: ToolInvocation) -> str:
return f"CURRENCY_FOR_{params.country_code}"
# Client 1 creates a session with tool A
session1 = await mctx.client1.create_session(
on_permission_request=PermissionHandler.approve_all, tools=[city_lookup]
)
# Client 2 resumes with tool B (different tool, union should have both)
session2 = await mctx.client2.resume_session(
session1.session_id,
on_permission_request=PermissionHandler.approve_all,
tools=[currency_lookup],
)
# Send prompts sequentially to avoid nondeterministic tool_call ordering
response1 = await session1.send_and_wait(
"Use the city_lookup tool with countryCode 'US' and tell me the result.",
timeout=10.0,
)
assert response1 is not None
assert "CITY_FOR_US" in (response1.data.content or "")
response2 = await session1.send_and_wait(
"Now use the currency_lookup tool with countryCode 'US' and tell me the result.",
timeout=10.0,
)
assert response2 is not None
assert "CURRENCY_FOR_US" in (response2.data.content or "")
await session2.disconnect()
@pytest.mark.timeout(90)
@pytest.mark.skip(
reason="Flaky on CI: Python TCP socket close detection is too slow for snapshot replay"
)
async def test_disconnecting_client_removes_its_tools(self, mctx: MultiClientContext):
"""Disconnecting a client removes its tools from the session."""
class InputParams(BaseModel):
input: str = Field(description="Input value")
@define_tool("stable_tool", description="A tool that persists across disconnects")
def stable_tool(params: InputParams, invocation: ToolInvocation) -> str:
return f"STABLE_{params.input}"
@define_tool(
"ephemeral_tool",
description="A tool that will disappear when its client disconnects",
)
def ephemeral_tool(params: InputParams, invocation: ToolInvocation) -> str:
return f"EPHEMERAL_{params.input}"
# Client 1 creates a session with stable_tool
session1 = await mctx.client1.create_session(
on_permission_request=PermissionHandler.approve_all, tools=[stable_tool]
)
# Client 2 resumes with ephemeral_tool
await mctx.client2.resume_session(
session1.session_id,
on_permission_request=PermissionHandler.approve_all,
tools=[ephemeral_tool],
)
# Verify both tools work before disconnect.
# Sequential prompts avoid nondeterministic tool_call ordering.
stable_response = await session1.send_and_wait(
"Use the stable_tool with input 'test1' and tell me the result.", timeout=10.0
)
assert stable_response is not None
assert "STABLE_test1" in (stable_response.data.content or "")
ephemeral_response = await session1.send_and_wait(
"Use the ephemeral_tool with input 'test2' and tell me the result.", timeout=10.0
)
assert ephemeral_response is not None
assert "EPHEMERAL_test2" in (ephemeral_response.data.content or "")
# Force disconnect client 2 without destroying the shared session
await mctx.client2.force_stop()
# Give the server time to process the connection close and remove tools
await asyncio.sleep(0.5)
# Recreate client2 for future tests (but don't rejoin the session)
actual_port = mctx.client1.runtime_port
mctx._client2 = CopilotClient(
connection=RuntimeConnection.for_uri(
f"localhost:{actual_port}", connection_token="py-tcp-shared-test-token"
)
)
# Now only stable_tool should be available
after_response = await session1.send_and_wait(
"Use the stable_tool with input 'still_here'."
" Also try using ephemeral_tool"
" if it is available.",
timeout=10.0,
)
assert after_response is not None
assert "STABLE_still_here" in (after_response.data.content or "")
# ephemeral_tool should NOT have produced a result
assert "EPHEMERAL_" not in (after_response.data.content or "")
You can’t perform that action at this time.