refactor(agent): make resolver sole runtime owner
This commit is contained in:
chengyongru
Xubin Ren
@@ -30,6 +30,7 @@ from nanobot.nanobot import (
|
||||
StreamEvent,
|
||||
StreamEventType,
|
||||
)
|
||||
from nanobot.utils.llm_runtime import runtime_from_provider_snapshot
|
||||
|
||||
|
||||
def _write_config(tmp_path: Path, overrides: dict | None = None) -> Path:
|
||||
@@ -504,45 +505,58 @@ async def test_run_allows_parallel_sessions_without_model_override(tmp_path):
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_run_model_overrides_are_serialized_before_snapshot_build(tmp_path):
|
||||
async def test_run_model_overrides_can_overlap_without_default_mutation(tmp_path):
|
||||
from nanobot.bus.events import OutboundMessage
|
||||
from nanobot.providers.factory import ProviderSnapshot
|
||||
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
original_model = bot._loop.model
|
||||
active_models: list[str] = []
|
||||
snapshot_base_models: list[str] = []
|
||||
assert not hasattr(bot, "_runtime_overrides")
|
||||
original_runtime = bot._loop.runtime_resolver.runtime
|
||||
active_runs: list[tuple[str, str]] = []
|
||||
first_entered = asyncio.Event()
|
||||
both_entered = asyncio.Event()
|
||||
release_first = asyncio.Event()
|
||||
|
||||
def fake_snapshot(*, model, model_preset):
|
||||
def fake_resolve(*, model, model_preset, config):
|
||||
assert model is not None
|
||||
assert model_preset is None
|
||||
snapshot_base_models.append(bot._loop.model)
|
||||
return ProviderSnapshot(
|
||||
assert config is bot._config
|
||||
return runtime_from_provider_snapshot(ProviderSnapshot(
|
||||
provider=_fake_provider(model, max_tokens=2048),
|
||||
model=model,
|
||||
context_window_tokens=4096,
|
||||
signature=("sdk", model),
|
||||
)
|
||||
))
|
||||
|
||||
bot._runtime_overrides.model_override_snapshot = MagicMock(side_effect=fake_snapshot)
|
||||
bot._loop.runtime_resolver.resolve_override = MagicMock(side_effect=fake_resolve)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, hooks):
|
||||
active_models.append(bot._loop.model)
|
||||
async def fake_process_direct(message, *, session_key, hooks, runtime):
|
||||
active_runs.append((session_key, runtime.model))
|
||||
assert bot._loop.runtime_resolver.runtime is original_runtime
|
||||
if message == "first":
|
||||
first_entered.set()
|
||||
await asyncio.wait_for(release_first.wait(), timeout=1)
|
||||
if len(active_runs) == 2:
|
||||
both_entered.set()
|
||||
await asyncio.wait_for(release_first.wait(), timeout=1)
|
||||
return OutboundMessage(channel="cli", chat_id="direct", content=message)
|
||||
|
||||
bot._loop.process_direct = fake_process_direct
|
||||
|
||||
first = asyncio.create_task(bot.run("first", model="model:first"))
|
||||
first = asyncio.create_task(bot.run(
|
||||
"first",
|
||||
session_key="sdk:first",
|
||||
model="model:first",
|
||||
))
|
||||
await asyncio.wait_for(first_entered.wait(), timeout=1)
|
||||
|
||||
second = asyncio.create_task(bot.run("second", model="model:second"))
|
||||
await asyncio.sleep(0)
|
||||
second = asyncio.create_task(bot.run(
|
||||
"second",
|
||||
session_key="sdk:second",
|
||||
model="model:second",
|
||||
))
|
||||
await asyncio.wait_for(both_entered.wait(), timeout=1)
|
||||
assert not first.done()
|
||||
assert not second.done()
|
||||
|
||||
release_first.set()
|
||||
@@ -550,21 +564,21 @@ async def test_run_model_overrides_are_serialized_before_snapshot_build(tmp_path
|
||||
|
||||
assert first_result.content == "first"
|
||||
assert second_result.content == "second"
|
||||
assert active_models == ["model:first", "model:second"]
|
||||
assert snapshot_base_models == [original_model, original_model]
|
||||
assert bot._loop.model == original_model
|
||||
assert set(active_runs) == {
|
||||
("sdk:first", "model:first"),
|
||||
("sdk:second", "model:second"),
|
||||
}
|
||||
assert bot._loop.runtime_resolver.runtime is original_runtime
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_run_model_override_is_per_run_and_restores_default(tmp_path):
|
||||
async def test_run_model_override_is_per_run_without_default_mutation(tmp_path):
|
||||
from nanobot.bus.events import OutboundMessage
|
||||
from nanobot.providers.factory import ProviderSnapshot
|
||||
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
original_provider = bot._loop.provider
|
||||
original_model = bot._loop.model
|
||||
original_signature = bot._loop._provider_signature
|
||||
original_runtime = bot._loop.runtime_resolver.runtime
|
||||
override_provider = _fake_provider("override-provider", max_tokens=2048)
|
||||
override = ProviderSnapshot(
|
||||
provider=override_provider,
|
||||
@@ -572,13 +586,17 @@ async def test_run_model_override_is_per_run_and_restores_default(tmp_path):
|
||||
context_window_tokens=4096,
|
||||
signature=("sdk", "override"),
|
||||
)
|
||||
bot._runtime_overrides.model_override_snapshot = MagicMock(return_value=override)
|
||||
override_runtime = runtime_from_provider_snapshot(override)
|
||||
bot._loop.runtime_resolver.resolve_override = MagicMock(
|
||||
return_value=override_runtime
|
||||
)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, hooks):
|
||||
assert bot._loop.provider is override_provider
|
||||
async def fake_process_direct(message, *, session_key, hooks, runtime):
|
||||
assert runtime is override_runtime
|
||||
assert not hasattr(bot._loop.runner, "provider")
|
||||
assert bot._loop.model == "openai/gpt-4.1-mini"
|
||||
assert bot._loop.context_window_tokens == 4096
|
||||
assert runtime.model == "openai/gpt-4.1-mini"
|
||||
assert runtime.context_window_tokens == 4096
|
||||
assert bot._loop.runtime_resolver.runtime is original_runtime
|
||||
return OutboundMessage(channel="cli", chat_id="direct", content="ok")
|
||||
|
||||
bot._loop.process_direct = fake_process_direct
|
||||
@@ -586,14 +604,13 @@ async def test_run_model_override_is_per_run_and_restores_default(tmp_path):
|
||||
result = await bot.run("hi", model="openai/gpt-4.1-mini")
|
||||
|
||||
assert result.content == "ok"
|
||||
bot._runtime_overrides.model_override_snapshot.assert_called_once_with(
|
||||
bot._loop.runtime_resolver.resolve_override.assert_called_once_with(
|
||||
model="openai/gpt-4.1-mini",
|
||||
model_preset=None,
|
||||
config=bot._config,
|
||||
)
|
||||
assert bot._loop.provider is original_provider
|
||||
assert not hasattr(bot._loop.runner, "provider")
|
||||
assert bot._loop.model == original_model
|
||||
assert bot._loop._provider_signature == original_signature
|
||||
assert bot._loop.runtime_resolver.runtime is original_runtime
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
@@ -603,7 +620,7 @@ async def test_run_model_preset_override_is_per_run(tmp_path):
|
||||
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
original_model = bot._loop.model
|
||||
original_runtime = bot._loop.runtime_resolver.runtime
|
||||
override_provider = _fake_provider("preset-provider", max_tokens=1024)
|
||||
override = ProviderSnapshot(
|
||||
provider=override_provider,
|
||||
@@ -611,19 +628,25 @@ async def test_run_model_preset_override_is_per_run(tmp_path):
|
||||
context_window_tokens=2048,
|
||||
signature=("preset", "fast"),
|
||||
)
|
||||
bot._loop._build_model_preset_snapshot = MagicMock(return_value=override)
|
||||
override_runtime = runtime_from_provider_snapshot(override, model_preset="fast")
|
||||
bot._loop.runtime_resolver.resolve_override = MagicMock(
|
||||
return_value=override_runtime
|
||||
)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, hooks):
|
||||
assert bot._loop.provider is override_provider
|
||||
assert bot._loop.model == "openai/gpt-4.1-mini"
|
||||
async def fake_process_direct(message, *, session_key, hooks, runtime):
|
||||
assert runtime is override_runtime
|
||||
return OutboundMessage(channel="cli", chat_id="direct", content="ok")
|
||||
|
||||
bot._loop.process_direct = fake_process_direct
|
||||
|
||||
await bot.run("hi", model_preset="fast")
|
||||
|
||||
bot._loop._build_model_preset_snapshot.assert_called_once_with("fast")
|
||||
assert bot._loop.model == original_model
|
||||
bot._loop.runtime_resolver.resolve_override.assert_called_once_with(
|
||||
model=None,
|
||||
model_preset="fast",
|
||||
config=bot._config,
|
||||
)
|
||||
assert bot._loop.runtime_resolver.runtime is original_runtime
|
||||
assert bot._loop.model_preset is None
|
||||
|
||||
|
||||
@@ -745,7 +768,9 @@ async def test_stream_yields_text_events_in_order(tmp_path):
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, on_stream, on_stream_end, hooks):
|
||||
async def fake_process_direct(
|
||||
message, *, session_key, on_stream, on_stream_end, hooks, runtime
|
||||
):
|
||||
assert message == "hi"
|
||||
assert session_key == "sdk:default"
|
||||
await on_stream("Hel")
|
||||
@@ -780,7 +805,9 @@ async def test_run_streamed_wait_returns_full_result_without_consuming_events(tm
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, on_stream, on_stream_end, hooks):
|
||||
async def fake_process_direct(
|
||||
message, *, session_key, on_stream, on_stream_end, hooks, runtime
|
||||
):
|
||||
await on_stream("done")
|
||||
await on_stream_end(resuming=False)
|
||||
ctx = AgentRunHookContext(
|
||||
@@ -822,7 +849,9 @@ async def test_run_streamed_cancel_releases_full_queue_without_consuming(tmp_pat
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, on_stream, on_stream_end, hooks):
|
||||
async def fake_process_direct(
|
||||
message, *, session_key, on_stream, on_stream_end, hooks, runtime
|
||||
):
|
||||
for i in range(400):
|
||||
await on_stream(str(i))
|
||||
await on_stream_end(resuming=False)
|
||||
@@ -886,16 +915,17 @@ async def test_run_streamed_forwards_runtime_options(tmp_path):
|
||||
assert callable(kwargs["on_stream"])
|
||||
assert callable(kwargs["on_stream_end"])
|
||||
assert kwargs["hooks"]
|
||||
assert kwargs["runtime"] is bot._loop.runtime_resolver.runtime
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_run_streamed_model_override_reports_model_and_restores(tmp_path):
|
||||
async def test_run_streamed_model_override_reports_admitted_runtime(tmp_path):
|
||||
from nanobot.bus.events import OutboundMessage
|
||||
from nanobot.providers.factory import ProviderSnapshot
|
||||
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
original_model = bot._loop.model
|
||||
original_runtime = bot._loop.runtime_resolver.runtime
|
||||
override_provider = _fake_provider("stream-provider", max_tokens=2048)
|
||||
override = ProviderSnapshot(
|
||||
provider=override_provider,
|
||||
@@ -903,11 +933,22 @@ async def test_run_streamed_model_override_reports_model_and_restores(tmp_path):
|
||||
context_window_tokens=4096,
|
||||
signature=("sdk", "stream"),
|
||||
)
|
||||
bot._runtime_overrides.model_override_snapshot = MagicMock(return_value=override)
|
||||
override_runtime = runtime_from_provider_snapshot(override)
|
||||
bot._loop.runtime_resolver.resolve_override = MagicMock(
|
||||
return_value=override_runtime
|
||||
)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, on_stream, on_stream_end, hooks):
|
||||
assert bot._loop.provider is override_provider
|
||||
assert bot._loop.model == "openai/gpt-4.1-mini"
|
||||
async def fake_process_direct(
|
||||
message,
|
||||
*,
|
||||
session_key,
|
||||
on_stream,
|
||||
on_stream_end,
|
||||
hooks,
|
||||
runtime,
|
||||
):
|
||||
assert runtime is override_runtime
|
||||
assert bot._loop.runtime_resolver.runtime is original_runtime
|
||||
await on_stream("ok")
|
||||
await on_stream_end(resuming=False)
|
||||
return OutboundMessage(channel="cli", chat_id="direct", content="ok")
|
||||
@@ -922,7 +963,7 @@ async def test_run_streamed_model_override_reports_model_and_restores(tmp_path):
|
||||
assert events[0].type == "run.started"
|
||||
assert events[0].metadata["model"] == "openai/gpt-4.1-mini"
|
||||
assert events[0].metadata["model_preset"] is None
|
||||
assert bot._loop.model == original_model
|
||||
assert bot._loop.runtime_resolver.runtime is original_runtime
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
@@ -947,7 +988,9 @@ async def test_run_streamed_emits_tool_events(tmp_path):
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, on_stream, on_stream_end, hooks):
|
||||
async def fake_process_direct(
|
||||
message, *, session_key, on_stream, on_stream_end, hooks, runtime
|
||||
):
|
||||
calls = [
|
||||
ToolCallRequest(id="call_ok", name="read_file", arguments={"path": "README.md"}),
|
||||
ToolCallRequest(id="call_bad", name="exec", arguments={"cmd": "false"}),
|
||||
@@ -993,7 +1036,9 @@ async def test_run_streamed_emits_reasoning_events(tmp_path):
|
||||
config_path = _write_config(tmp_path)
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
|
||||
async def fake_process_direct(message, *, session_key, on_stream, on_stream_end, hooks):
|
||||
async def fake_process_direct(
|
||||
message, *, session_key, on_stream, on_stream_end, hooks, runtime
|
||||
):
|
||||
for hook in hooks:
|
||||
await hook.emit_reasoning("thinking")
|
||||
await hook.emit_reasoning_end()
|
||||
@@ -1020,7 +1065,9 @@ async def test_stream_generator_break_cancels_underlying_run(tmp_path):
|
||||
bot = Nanobot.from_config(config_path, workspace=tmp_path)
|
||||
cancelled = asyncio.Event()
|
||||
|
||||
async def fake_process_direct(message, *, session_key, on_stream, on_stream_end, hooks):
|
||||
async def fake_process_direct(
|
||||
message, *, session_key, on_stream, on_stream_end, hooks, runtime
|
||||
):
|
||||
try:
|
||||
await on_stream("first")
|
||||
await asyncio.sleep(10)
|
||||
|
||||
Reference in New Issue
Block a user