fix(cli): run one-shot query cleanup before lease release (#43036)

* fix(cli): run one-shot query cleanup before lease release

Signed-off-by: mnajafian-nv <mnajafian@nvidia.com>

* test(cli): cover quiet one-shot cleanup finalization

Signed-off-by: mnajafian-nv <mnajafian@nvidia.com>

---------

Signed-off-by: mnajafian-nv <mnajafian@nvidia.com>
This commit is contained in:
mnajafian-nv 2026-06-09 21:52:13 -07:00 committed by GitHub
parent 96af61b6ef
commit d03cdd63eb
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 257 additions and 8 deletions

49
cli.py
View file

@ -952,7 +952,7 @@ def _prepare_deferred_agent_startup() -> None:
exc_info=True,
)
def _run_cleanup():
def _run_cleanup(*, notify_session_finalize: bool = True):
"""Run resource cleanup exactly once."""
global _cleanup_done
if _cleanup_done:
@ -988,16 +988,12 @@ def _run_cleanup():
pass
# Shut down memory provider (on_session_end + shutdown_all) at actual
# session boundary — NOT per-turn inside run_conversation().
try:
from hermes_cli.plugins import invoke_hook as _invoke_hook
_invoke_hook(
"on_session_finalize",
if notify_session_finalize:
_notify_session_finalize(
session_id=_active_agent_ref.session_id if _active_agent_ref else None,
platform="cli",
reason="shutdown",
)
except Exception:
pass
try:
if _active_agent_ref and hasattr(_active_agent_ref, 'shutdown_memory_provider'):
# Forward the agent's own transcript so memory providers'
@ -1015,6 +1011,24 @@ def _run_cleanup():
pass
def _notify_session_finalize(
*,
session_id: str | None,
platform: str = "cli",
reason: str = "shutdown",
) -> None:
try:
from hermes_cli.plugins import invoke_hook as _invoke_hook
_invoke_hook(
"on_session_finalize",
session_id=session_id,
platform=platform,
reason=reason,
)
except Exception:
pass
def _emit_interrupted_session_end(cli, *, reason: str = "keyboard_interrupt") -> None:
"""Best-effort on_session_end hook for interrupted non-interactive runs."""
agent = getattr(cli, "agent", None)
@ -1051,6 +1065,25 @@ def _emit_interrupted_session_end(cli, *, reason: str = "keyboard_interrupt") ->
pass
def _notify_single_query_session_finalize(cli, *, reason: str = "shutdown") -> None:
agent = getattr(cli, "agent", None)
session_id = getattr(agent, "session_id", None) or getattr(cli, "session_id", None)
_notify_session_finalize(
session_id=session_id,
platform=getattr(agent, "platform", None) or "cli",
reason=reason,
)
def _finalize_single_query(cli) -> None:
"""Close one-shot CLI resources before releasing the active session lease."""
try:
_notify_single_query_session_finalize(cli)
_run_cleanup(notify_session_finalize=False)
finally:
cli._release_active_session()
def _reset_terminal_input_modes_on_exit() -> None:
"""Best-effort: disable focus reporting + mouse tracking on TUI exit so they
don't leak into the next shell session sharing the tab.
@ -13583,7 +13616,7 @@ def main(
cli.chat(query, images=single_query_images or None)
cli._print_exit_summary()
finally:
cli._release_active_session()
_finalize_single_query(cli)
return
# Run interactive mode

View file

@ -0,0 +1,216 @@
from types import SimpleNamespace
import pytest
import cli
def test_finalize_single_query_runs_cleanup_without_reemitting_finalize_before_release(monkeypatch):
calls = []
fake_cli = SimpleNamespace(_release_active_session=lambda: calls.append(("release", {})))
def cleanup(**kwargs):
calls.append(("cleanup", kwargs))
monkeypatch.setattr(
cli,
"_notify_single_query_session_finalize",
lambda _cli: calls.append(("finalize", {})),
)
monkeypatch.setattr(cli, "_run_cleanup", cleanup)
cli._finalize_single_query(fake_cli)
assert calls == [
("finalize", {}),
("cleanup", {"notify_session_finalize": False}),
("release", {}),
]
def test_finalize_single_query_releases_session_when_cleanup_fails(monkeypatch):
calls = []
fake_cli = SimpleNamespace(_release_active_session=lambda: calls.append("release"))
def cleanup(**kwargs):
calls.append("cleanup")
raise RuntimeError("cleanup failed")
monkeypatch.setattr(
cli,
"_notify_single_query_session_finalize",
lambda _cli: calls.append("finalize"),
)
monkeypatch.setattr(cli, "_run_cleanup", cleanup)
with pytest.raises(RuntimeError, match="cleanup failed"):
cli._finalize_single_query(fake_cli)
assert calls == ["finalize", "cleanup", "release"]
def test_finalize_single_query_runs_cleanup_when_finalize_hook_fails(monkeypatch):
calls = []
fake_agent = SimpleNamespace(session_id="agent-session", platform="cli")
fake_cli = SimpleNamespace(
agent=fake_agent,
session_id="cli-session",
_release_active_session=lambda: calls.append("release"),
)
def invoke_hook(name, **kwargs):
calls.append("finalize")
raise RuntimeError("hook failed")
monkeypatch.setattr("hermes_cli.plugins.invoke_hook", invoke_hook)
monkeypatch.setattr(cli, "_run_cleanup", lambda **kwargs: calls.append("cleanup"))
cli._finalize_single_query(fake_cli)
assert calls == ["finalize", "cleanup", "release"]
def test_notify_single_query_session_finalize_uses_agent_session(monkeypatch):
calls = []
fake_agent = SimpleNamespace(session_id="agent-session", platform="cli")
fake_cli = SimpleNamespace(agent=fake_agent, session_id="cli-session")
def invoke_hook(name, **kwargs):
calls.append((name, kwargs))
monkeypatch.setattr("hermes_cli.plugins.invoke_hook", invoke_hook)
cli._notify_single_query_session_finalize(fake_cli)
assert calls == [
(
"on_session_finalize",
{
"session_id": "agent-session",
"platform": "cli",
"reason": "shutdown",
},
)
]
def test_human_single_query_main_finalizes_after_query(monkeypatch):
calls = []
import cli as cli_mod
class _Console:
def print(self, *_args, **_kwargs):
calls.append("query-label")
class FakeCLI:
def __init__(self, **_kwargs):
self.console = _Console()
self.session_id = "single-query-session"
self.agent = SimpleNamespace(
session_id="single-query-session",
platform="cli",
)
def _claim_active_session(self, surface, *, stderr=False):
calls.append(("claim", surface, stderr))
return True
def _show_security_advisories(self):
calls.append("advisories")
def chat(self, query, images=None):
calls.append(("chat", query, images))
return "done"
def _print_exit_summary(self):
calls.append("summary")
monkeypatch.setattr(cli_mod, "HermesCLI", FakeCLI)
monkeypatch.setattr(cli_mod.atexit, "register", lambda *_args, **_kwargs: None)
monkeypatch.setattr(
cli_mod,
"_finalize_single_query",
lambda fake_cli: calls.append(("finalize", fake_cli.session_id)),
)
cli_mod.main(query="hello", quiet=False, toolsets="terminal")
assert calls == [
("claim", "cli", False),
"query-label",
"advisories",
("chat", "hello", None),
"summary",
("finalize", "single-query-session"),
]
def test_quiet_single_query_main_finalizes_while_preserving_exit_code(monkeypatch):
calls = []
import cli as cli_mod
def run_conversation(*, user_message, conversation_history):
calls.append(("run", user_message, conversation_history))
return {
"final_response": "",
"error": "provider failed",
"failed": True,
}
class FakeCLI:
def __init__(self, **_kwargs):
self.provider = "test-provider"
self.model = "test-model"
self.session_id = "quiet-session"
self.conversation_history = []
self._active_agent_route_signature = "same-route"
self.agent = SimpleNamespace(
session_id="quiet-session",
platform="cli",
quiet_mode=False,
suppress_status_output=False,
stream_delta_callback=object(),
tool_gen_callback=object(),
run_conversation=run_conversation,
)
def _claim_active_session(self, surface, *, stderr=False):
calls.append(("claim", surface, stderr))
return True
def _ensure_runtime_credentials(self):
calls.append("credentials")
return True
def _resolve_turn_agent_config(self, effective_query):
calls.append(("resolve", effective_query))
return {
"signature": "same-route",
"model": None,
"runtime": None,
"request_overrides": None,
}
def _init_agent(self, **kwargs):
calls.append(("init", kwargs))
return True
monkeypatch.delenv("HERMES_KANBAN_TASK", raising=False)
monkeypatch.delenv("HERMES_KANBAN_GOAL_MODE", raising=False)
monkeypatch.setattr(cli_mod, "HermesCLI", FakeCLI)
monkeypatch.setattr(cli_mod.atexit, "register", lambda *_args, **_kwargs: None)
monkeypatch.setattr(
cli_mod,
"_finalize_single_query",
lambda fake_cli: calls.append(("finalize", fake_cli.session_id)),
)
with pytest.raises(SystemExit) as exc_info:
cli_mod.main(query="hello", quiet=True, toolsets="terminal")
assert exc_info.value.code == 1
assert ("claim", "cli", True) in calls
assert ("run", "hello", []) in calls
assert calls[-1] == ("finalize", "quiet-session")