Merge branch 'main' into register_custom_model

tests/unit/providers/inference/test_remote_vllm.py

@@ -26,7 +26,12 @@ from openai.types.chat.chat_completion_chunk import (
 )
 from openai.types.model import Model as OpenAIModel
 
-from llama_stack.apis.inference import ToolChoice, ToolConfig
+from llama_stack.apis.inference import (
+    ChatCompletionRequest,
+    ToolChoice,
+    ToolConfig,
+    UserMessage,
+)
 from llama_stack.apis.models import Model
 from llama_stack.models.llama.datatypes import StopReason
 from llama_stack.providers.remote.inference.vllm.config import VLLMInferenceAdapterConfig
@@ -232,3 +237,14 @@ def test_chat_completion_doesnt_block_event_loop(caplog):
     # above.
     asyncio_warnings = [record.message for record in caplog.records if record.name == "asyncio"]
     assert not asyncio_warnings
+
+
+@pytest.mark.asyncio
+async def test_get_params_empty_tools(vllm_inference_adapter):
+    request = ChatCompletionRequest(
+        tools=[],
+        model="test_model",
+        messages=[UserMessage(content="test")],
+    )
+    params = await vllm_inference_adapter._get_params(request)
+    assert "tools" not in params

tests/unit/providers/utils/test_scheduler.py

@@ -0,0 +1,120 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import asyncio
+
+import pytest
+
+from llama_stack.providers.utils.scheduler import JobStatus, Scheduler
+
+
+@pytest.mark.asyncio
+async def test_scheduler_unknown_backend():
+    with pytest.raises(ValueError):
+        Scheduler(backend="unknown")
+
+
+@pytest.mark.asyncio
+async def test_scheduler_naive():
+    sched = Scheduler()
+
+    # make sure the scheduler starts empty
+    with pytest.raises(ValueError):
+        sched.get_job("unknown")
+    assert sched.get_jobs() == []
+
+    called = False
+
+    # schedule a job that will exercise the handlers
+    async def job_handler(on_log, on_status, on_artifact):
+        nonlocal called
+        called = True
+        # exercise the handlers
+        on_log("test log1")
+        on_log("test log2")
+        on_artifact({"type": "type1", "path": "path1"})
+        on_artifact({"type": "type2", "path": "path2"})
+        on_status(JobStatus.completed)
+
+    job_id = "test_job_id"
+    job_type = "test_job_type"
+    sched.schedule(job_type, job_id, job_handler)
+
+    # make sure the job was properly registered
+    with pytest.raises(ValueError):
+        sched.get_job("unknown")
+    assert sched.get_job(job_id) is not None
+    assert sched.get_jobs() == [sched.get_job(job_id)]
+
+    assert sched.get_jobs("unknown") == []
+    assert sched.get_jobs(job_type) == [sched.get_job(job_id)]
+
+    # now shut the scheduler down and make sure the job ran
+    await sched.shutdown()
+
+    assert called
+
+    job = sched.get_job(job_id)
+    assert job is not None
+
+    assert job.status == JobStatus.completed
+
+    assert job.scheduled_at is not None
+    assert job.started_at is not None
+    assert job.completed_at is not None
+    assert job.scheduled_at < job.started_at < job.completed_at
+
+    assert job.artifacts == [
+        {"type": "type1", "path": "path1"},
+        {"type": "type2", "path": "path2"},
+    ]
+    assert [msg[1] for msg in job.logs] == ["test log1", "test log2"]
+    assert job.logs[0][0] < job.logs[1][0]
+
+
+@pytest.mark.asyncio
+async def test_scheduler_naive_handler_raises():
+    sched = Scheduler()
+
+    async def failing_job_handler(on_log, on_status, on_artifact):
+        on_status(JobStatus.running)
+        raise ValueError("test error")
+
+    job_id = "test_job_id1"
+    job_type = "test_job_type"
+    sched.schedule(job_type, job_id, failing_job_handler)
+
+    job = sched.get_job(job_id)
+    assert job is not None
+
+    # confirm the exception made the job transition to failed state, even
+    # though it was set to `running` before the error
+    for _ in range(10):
+        if job.status == JobStatus.failed:
+            break
+        await asyncio.sleep(0.1)
+    assert job.status == JobStatus.failed
+
+    # confirm that the raised error got registered in log
+    assert job.logs[0][1] == "test error"
+
+    # even after failed job, we can schedule another one
+    called = False
+
+    async def successful_job_handler(on_log, on_status, on_artifact):
+        nonlocal called
+        called = True
+        on_status(JobStatus.completed)
+
+    job_id = "test_job_id2"
+    sched.schedule(job_type, job_id, successful_job_handler)
+
+    await sched.shutdown()
+
+    assert called
+    job = sched.get_job(job_id)
+    assert job is not None
+    assert job.status == JobStatus.completed