Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
Python agent-contextagent-memoryagent-pluginsagentic-ragcontext-databasedsh-pluginknowledge-basememorymemory-managementself-evolvingvector-database
fix(cli): probe embedding provider in doctor (#2333)
Mingjian Que2d7d93a0a879
diff --git a/openviking_cli/doctor.py b/openviking_cli/doctor.pyindex 849fffb1c..457ce67ac 100644--- a/openviking_cli/doctor.py+++ b/openviking_cli/doctor.py@@ -9,6 +9,7 @@ native vector engine, AGFS, embedding provider, VLM provider, and disk space.from __future__ import annotations+import asyncioimport importlibimport jsonimport os@@ -16,7 +17,7 @@ import platformimport shutilimport sysfrom pathlib import Path-from typing import Optional+from typing import Any, Optionalfrom openviking_cli.utils.config.config_loader import resolve_config_pathfrom openviking_cli.utils.config.consts import OPENVIKING_CONFIG_ENV@@ -143,6 +144,90 @@ def check_agfs() -> tuple[bool, str, Optional[str]]:)+def _embedding_probe_label(provider: str, model: str, dense: dict[str, Any]) -> str:+ api_base = dense.get("api_base")+ dimension = dense.get("dimension")+ parts = [f"{provider}/{model}"]+ if api_base:+ parts.append(f"api_base={api_base}")+ if dimension:+ parts.append(f"dimension={dimension}")+ return " ".join(parts)+++def _probe_embedding_provider(+ embedding: dict[str, Any],+ dense: dict[str, Any],+) -> tuple[bool, str, Optional[str]]:+ """Create the configured embedder, make one request, and validate vector shape."""+ from openviking.models.embedder.base import embed_compat+ from openviking_cli.utils.config.embedding_config import EmbeddingConfig, EmbeddingModelConfig++ provider = dense.get("provider", "local")+ model = dense.get("model", "bge-small-zh-v1.5-f16")+ label = _embedding_probe_label(provider, model, dense)++ try:+ model_config = EmbeddingModelConfig(**dense)+ config_kwargs = {+ key: embedding[key]+ for key in ("max_concurrent", "max_retries", "text_source", "max_input_tokens")+ if key in embedding+ }+ embedding_config = EmbeddingConfig(dense=model_config, **config_kwargs)+ embedder = embedding_config.get_embedder()+ expected_dimension = (+ model_config.dimension+ if model_config.dimension is not None+ else embedder.get_dimension()+ )+ except Exception as exc:+ return (+ False,+ f"{label} (invalid embedding config: {exc})",+ "Fix embedding.dense provider/model/api_base/dimension in ov.conf",+ )++ async def _run_probe():+ return await asyncio.wait_for(+ embed_compat(embedder, "OpenViking doctor embedding probe", is_query=True),+ timeout=10.0,+ )++ try:+ result = asyncio.run(_run_probe())+ except TimeoutError:+ return (+ False,+ f"{label} (probe timed out)",+ "Check embedding.dense.api_base and make sure the embedding provider is reachable",+ )+ except Exception as exc:+ return (+ False,+ f"{label} (probe failed: {exc})",+ "Check the embedding provider service, model name, API key, and api_base",+ )++ vector = getattr(result, "dense_vector", None)+ if not vector:+ return (+ False,+ f"{label} (probe returned no dense vector)",+ "Use a dense embedding model or configure embedding.dense correctly",+ )++ actual_dimension = len(vector)+ if actual_dimension != expected_dimension:+ return (+ False,+ f"{label} (dimension mismatch: expected {expected_dimension}, got {actual_dimension})",+ "Set embedding.dense.dimension to match the provider output, or use a matching model/index",+ )++ return True, f"{label} probe ok (dimension={actual_dimension})", None++def check_embedding() -> tuple[bool, str, Optional[str]]:"""Load embedding config and verify provider connectivity."""config_path = _find_config()@@ -202,19 +287,16 @@ def check_embedding() -> tuple[bool, str, Optional[str]]:None,)- # Ollama doesn't need an API key- if provider == "ollama":- return True, f"{provider}/{model}", None-api_key = dense.get("api_key", "")- if not api_key or api_key.startswith("{"):+ api_base = dense.get("api_base", "")+ if provider != "ollama" and not api_base and (not api_key or api_key.startswith("{")):return (False,f"{provider}/{model} (no API key)","Set embedding.dense.api_key in ov.conf",)- return True, f"{provider}/{model}", None+ return _probe_embedding_provider(embedding, dense)def check_vlm() -> tuple[bool, str, Optional[str]]:diff --git a/tests/cli/test_doctor.py b/tests/cli/test_doctor.pyindex 43e30f642..fee08cdd0 100644--- a/tests/cli/test_doctor.py+++ b/tests/cli/test_doctor.py@@ -4,6 +4,7 @@from __future__ import annotations+import importlibimport jsonimport osimport sys@@ -11,6 +12,7 @@ from pathlib import Pathfrom unittest.mock import patchfrom openviking_cli.doctor import (+ _probe_embedding_provider,check_agfs,check_config,check_disk,@@ -151,10 +153,17 @@ class TestCheckEmbedding:config = tmp_path / "ov.conf"config.write_text(json.dumps({}))+ real_import_module = importlib.import_module++ def import_module_side_effect(name: str, *args, **kwargs):+ if name == "llama_cpp":+ raise ImportError("No module named 'llama_cpp'")+ return real_import_module(name, *args, **kwargs)+with patch("openviking_cli.doctor._find_config", return_value=config):with patch("openviking_cli.doctor.importlib.import_module",- side_effect=ImportError("No module named 'llama_cpp'"),+ side_effect=import_module_side_effect,):ok, detail, fix = check_embedding()@@ -243,9 +252,41 @@ class TestCheckEmbedding:))with patch("openviking_cli.doctor._find_config", return_value=config):- ok, detail, fix = check_embedding()+ with patch(+ "openviking_cli.doctor._probe_embedding_provider",+ return_value=(True, "openai/text-embedding-3-small probe ok", None),+ ):+ ok, detail, fix = check_embedding()assert okassert "openai" in detail+ assert fix is None++ def test_pass_ollama_after_provider_probe(self, tmp_path: Path):+ config = tmp_path / "ov.conf"+ config.write_text(+ json.dumps(+ {+ "embedding": {+ "dense": {+ "provider": "ollama",+ "model": "bge-m3",+ "api_base": "http://localhost:11434/v1",+ "dimension": 1024,+ }+ }+ }+ )+ )+ with patch("openviking_cli.doctor._find_config", return_value=config):+ with patch(+ "openviking_cli.doctor._probe_embedding_provider",+ return_value=(True, "ollama/bge-m3 probe ok (dimension=1024)", None),+ ) as probe:+ ok, detail, fix = check_embedding()+ probe.assert_called_once()+ assert ok+ assert "ollama/bge-m3" in detail+ assert fix is Nonedef test_pass_with_api_key_from_environment_variable(self, tmp_path: Path):config = tmp_path / "ov.conf"@@ -264,7 +305,11 @@ class TestCheckEmbedding:)with patch("openviking_cli.doctor._find_config", return_value=config):with patch.dict(os.environ, {"OPENAI_API_KEY": "sk-env-123"}, clear=False):- ok, detail, fix = check_embedding()+ with patch(+ "openviking_cli.doctor._probe_embedding_provider",+ return_value=(True, "openai/text-embedding-3-small probe ok", None),+ ):+ ok, detail, fix = check_embedding()assert okassert "openai" in detail@@ -298,6 +343,149 @@ class TestCheckEmbedding:assert not okassert "unreadable" in detail+ def test_embedding_probe_fails_on_provider_error(self):+ with patch(+ "openviking_cli.utils.config.embedding_config.EmbeddingConfig.get_embedder",+ return_value=object(),+ ):+ with patch(+ "openviking.models.embedder.base.embed_compat",+ side_effect=RuntimeError("model not found"),+ ):+ ok, detail, fix = _probe_embedding_provider(+ {"max_retries": 0},+ {+ "provider": "ollama",+ "model": "missing-model",+ "api_base": "http://localhost:11434/v1",+ "dimension": 1024,+ },+ )++ assert not ok+ assert "probe failed" in detail+ assert "model not found" in detail+ assert "model name" in fix++ def test_embedding_probe_fails_on_empty_dense_vector(self):+ class FakeEmbedder:+ pass++ async def fake_embed_compat(*args, **kwargs):+ return type("Result", (), {"dense_vector": []})()++ with patch(+ "openviking_cli.utils.config.embedding_config.EmbeddingConfig.get_embedder",+ return_value=FakeEmbedder(),+ ):+ with patch(+ "openviking.models.embedder.base.embed_compat",+ side_effect=fake_embed_compat,+ ):+ ok, detail, fix = _probe_embedding_provider(+ {},+ {+ "provider": "openai",+ "model": "custom-embed",+ "api_base": "http://localhost:11434/v1",+ "dimension": 3,+ },+ )++ assert not ok+ assert "no dense vector" in detail+ assert "embedding.dense" in fix++ def test_embedding_probe_fails_on_dimension_mismatch(self):+ class FakeEmbedder:+ def get_dimension(self):+ return 2++ async def fake_embed_compat(*args, **kwargs):+ return type("Result", (), {"dense_vector": [0.1, 0.2]})()++ with patch(+ "openviking_cli.utils.config.embedding_config.EmbeddingConfig.get_embedder",+ return_value=FakeEmbedder(),+ ):+ with patch(+ "openviking.models.embedder.base.embed_compat",+ side_effect=fake_embed_compat,+ ):+ ok, detail, fix = _probe_embedding_provider(+ {},+ {+ "provider": "openai",+ "model": "custom-embed",+ "api_base": "http://localhost:11434/v1",+ "dimension": 3,+ },+ )++ assert not ok+ assert "dimension mismatch" in detail+ assert "expected 3, got 2" in detail+ assert "embedding.dense.dimension" in fix++ def test_embedding_probe_uses_embedder_dimension_when_config_omits_dimension(self):+ class FakeEmbedder:+ def get_dimension(self):+ return 2++ async def fake_embed_compat(*args, **kwargs):+ return type("Result", (), {"dense_vector": [0.1, 0.2]})()++ with patch(+ "openviking_cli.utils.config.embedding_config.EmbeddingConfig.get_embedder",+ return_value=FakeEmbedder(),+ ):+ with patch(+ "openviking.models.embedder.base.embed_compat",+ side_effect=fake_embed_compat,+ ):+ ok, detail, fix = _probe_embedding_provider(+ {},+ {+ "provider": "openai",+ "model": "custom-embed",+ "api_base": "http://localhost:11434/v1",+ },+ )++ assert ok+ assert "dimension=2" in detail+ assert fix is None++ def test_embedding_probe_fails_when_inferred_dimension_mismatches_probe(self):+ class FakeEmbedder:+ def get_dimension(self):+ return 3++ async def fake_embed_compat(*args, **kwargs):+ return type("Result", (), {"dense_vector": [0.1, 0.2]})()++ with patch(+ "openviking_cli.utils.config.embedding_config.EmbeddingConfig.get_embedder",+ return_value=FakeEmbedder(),+ ):+ with patch(+ "openviking.models.embedder.base.embed_compat",+ side_effect=fake_embed_compat,+ ):+ ok, detail, fix = _probe_embedding_provider(+ {},+ {+ "provider": "openai",+ "model": "custom-embed",+ "api_base": "http://localhost:11434/v1",+ },+ )++ assert not ok+ assert "dimension mismatch" in detail+ assert "expected 3, got 2" in detail+ assert "embedding.dense.dimension" in fix+class TestCheckVlm:def test_pass_with_config(self, tmp_path: Path):@@ -506,7 +694,11 @@ class TestRunDoctor:))with patch("openviking_cli.doctor._find_config", return_value=config):- code = run_doctor()+ with patch(+ "openviking_cli.doctor._probe_embedding_provider",+ return_value=(True, "openai/m probe ok", None),+ ):+ code = run_doctor()captured = capsys.readouterr()assert "OpenViking Doctor" in captured.out# May not be 0 if native engine is missing, but the function should complete