diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 5f428f8..a2a2e60 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -21,6 +21,7 @@ jobs:
   ci:
     runs-on: ubuntu-latest
     strategy:
+      max-parallel: 1
       matrix:
         python-version: ["3.8", "3.9", "3.10", "3.11"]
 
@@ -41,10 +42,17 @@ jobs:
 
       - name: Install Dev dependencies
         run: |
-          pip install langchain openai pytest ruff mypy
+          pip install langchain openai pytest ruff mypy pymysql pytest-asyncio types-PyMySQL
 
       # TODO(yuanbohan): code coverage with pytest-cov
       - name: Test with pytest
+        env:
+          GREPTIMEAI_HOST: ${{ secrets.GREPTIMEAI_HOST }}
+          GREPTIMEAI_USERNAME: ${{ secrets.GREPTIMEAI_USERNAME }}
+          GREPTIMEAI_PASSWORD: ${{ secrets.GREPTIMEAI_PASSWORD }}
+          GREPTIMEAI_DATABASE: ${{ secrets.GREPTIMEAI_DATABASE }}
+          GREPTIMEAI_TOKEN: ${{ secrets.GREPTIMEAI_TOKEN }}
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
         run: |
           pytest tests/ -v
 
@@ -56,3 +64,7 @@ jobs:
       - name: Static Type Checking with mypy
         run: |
           mypy $(git ls-files '*.py') --check-untyped-defs
+
+concurrency:
+  group: ${{ github.repository }}
+  cancel-in-progress: true
diff --git a/pyproject.toml b/pyproject.toml
index f759bb9..ce990aa 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -37,6 +37,8 @@ dev-dependencies = [
     "langchain>=0.0.27",
     "openai>=1.3.5",
     "mypy>=1.7.1",
+    "pymysql>=1.1.0",
+    "pytest-asyncio>=0.23.2",
 ]
 
 [tool.rye.scripts]
diff --git a/requirements-dev.lock b/requirements-dev.lock
index 09e0d31..d5af55c 100644
--- a/requirements-dev.lock
+++ b/requirements-dev.lock
@@ -22,7 +22,6 @@ distro==1.8.0
 exceptiongroup==1.1.3
 frozenlist==1.4.0
 googleapis-common-protos==1.61.0
-greenlet==3.0.1
 h11==0.14.0
 httpcore==1.0.2
 httpx==0.25.1
@@ -47,16 +46,18 @@ opentelemetry-sdk==1.20.0
 opentelemetry-semantic-conventions==0.41b0
 packaging==23.2
 pluggy==1.3.0
-protobuf==4.24.4
+protobuf==4.25.1
 pydantic==2.4.2
 pydantic-core==2.10.1
+pymysql==1.1.0
 pytest==7.4.3
+pytest-asyncio==0.23.2
 pyyaml==6.0.1
 regex==2023.10.3
 requests==2.31.0
 ruff==0.1.3
 sniffio==1.3.0
-sqlalchemy==2.0.22
+sqlalchemy==2.0.23
 tenacity==8.2.3
 tiktoken==0.5.1
 tomli==2.0.1
diff --git a/src/greptimeai/openai_patcher.py b/src/greptimeai/openai_patcher.py
index 214f68f..19b5086 100644
--- a/src/greptimeai/openai_patcher.py
+++ b/src/greptimeai/openai_patcher.py
@@ -18,6 +18,8 @@
 )
 from greptimeai.patcher.openai_patcher.retry import _RetryPatcher
 
+_collector: Collector = None  # type: ignore
+
 
 def setup(
     host: str = "",
@@ -37,20 +39,21 @@ def setup(
         token: if None or empty string, GREPTIMEAI_TOKEN environment variable will be used.
         client: if None, then openai module-level client will be patched.
     """
-    collector = Collector(
+    global _collector
+    _collector = Collector(
         service_name="openai", host=host, database=database, token=token
     )
     patchers: List[Patcher] = [
-        _AudioPatcher(collector=collector, client=client),
-        _ChatCompletionPatcher(collector=collector, client=client),
-        _CompletionPatcher(collector=collector, client=client),
-        _EmbeddingPatcher(collector=collector, client=client),
-        _FilePatcher(collector=collector, client=client),
-        _FineTuningPatcher(collector=collector, client=client),
-        _ImagePatcher(collector=collector, client=client),
-        _ModelPatcher(collector=collector, client=client),
-        _ModerationPatcher(collector=collector, client=client),
-        _RetryPatcher(collector=collector, client=client),
+        _AudioPatcher(collector=_collector, client=client),
+        _ChatCompletionPatcher(collector=_collector, client=client),
+        _CompletionPatcher(collector=_collector, client=client),
+        _EmbeddingPatcher(collector=_collector, client=client),
+        _FilePatcher(collector=_collector, client=client),
+        _FineTuningPatcher(collector=_collector, client=client),
+        _ImagePatcher(collector=_collector, client=client),
+        _ModelPatcher(collector=_collector, client=client),
+        _ModerationPatcher(collector=_collector, client=client),
+        _RetryPatcher(collector=_collector, client=client),
     ]
 
     for patcher in patchers:
diff --git a/tests/__init__.py b/tests/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/collection/__init__.py b/tests/collection/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/integrations/__init__.py b/tests/integrations/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/integrations/database/__init__.py b/tests/integrations/database/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/integrations/database/db.py b/tests/integrations/database/db.py
new file mode 100644
index 0000000..a0694ba
--- /dev/null
+++ b/tests/integrations/database/db.py
@@ -0,0 +1,58 @@
+import logging
+import os
+from typing import Union, List
+
+import pymysql
+
+from .model import Tables
+
+db = pymysql.connect(
+    host=os.getenv("GREPTIMEAI_HOST"),
+    user=os.getenv("GREPTIMEAI_USERNAME"),
+    passwd=os.getenv("GREPTIMEAI_PASSWORD"),
+    port=4002,
+    db=os.getenv("GREPTIMEAI_DATABASE"),
+)
+cursor = db.cursor()
+
+trace_sql = "SELECT model,prompt_tokens,completion_tokens FROM %s WHERE user_id = '%s'"
+truncate_sql = "TRUNCATE %s"
+
+
+def get_trace_data(user_id: str) -> List[Union[str, int]]:
+    """
+    get trace data for llm trace by user_id
+    :param is_stream:
+    :param user_id:
+    :return: model, prompt_tokens, completion_tokens
+    """
+
+    cursor.execute(trace_sql % (Tables.llm_trace, user_id))
+    trace = cursor.fetchone()
+    if trace is None:
+        raise Exception("trace data is None")
+    return list(trace)
+
+
+def truncate_tables():
+    """
+    truncate all tables
+    :return:
+    """
+    tables = [
+        "llm_completion_tokens",
+        "llm_completion_tokens_cost",
+        "llm_errors",
+        "llm_prompt_tokens",
+        "llm_prompt_tokens_cost",
+        "llm_request_duration_ms_bucket",
+        "llm_request_duration_ms_count",
+        "llm_request_duration_ms_sum",
+        "llm_traces_preview_v01",
+    ]
+    try:
+        cursor.executemany(truncate_sql, tables)
+        db.commit()
+    except Exception as e:
+        logging.error(e)
+        db.rollback()
diff --git a/tests/integrations/database/model.py b/tests/integrations/database/model.py
new file mode 100644
index 0000000..6fcb3f8
--- /dev/null
+++ b/tests/integrations/database/model.py
@@ -0,0 +1,88 @@
+class LlmTrace(object):
+    table_name = "llm_traces_preview_v01"
+
+    trace_id: str
+    span_id: str
+    parent_span_id: str
+    resource_attributes: str
+    scope_name: str
+    scope_version: str
+    scope_attributes: str
+    trace_state: str
+    span_name: str
+    span_kind: str
+    span_status_code: str
+    span_status_message: str
+    span_attributes: str
+    span_events: str
+    span_links: str
+    start: float
+    end: float
+    user_id: str
+    model: str
+    prompt_tokens: int
+    prompt_cost: float
+    completion_tokens: int
+    completion_cost: float
+    greptime_value: str
+    greptime_timestamp: float
+
+
+class LlmPromptToken(object):
+    table_name = "llm_prompt_tokens"
+
+    telemetry_sdk_language: str
+    telemetry_sdk_name: str
+    telemetry_sdk_version: str
+    service_name: str
+    span_name: str
+    model: str
+    greptime_value: str
+    greptime_timestamp: float
+
+
+class LlmPromptTokenCost(object):
+    table_name = "llm_prompt_tokens_cost"
+
+    telemetry_sdk_language: str
+    telemetry_sdk_name: str
+    telemetry_sdk_version: str
+    service_name: str
+    span_name: str
+    model: str
+    greptime_value: str
+    greptime_timestamp: float
+
+
+class LlmCompletionToken(object):
+    table_name = "llm_completion_tokens"
+
+    telemetry_sdk_language: str
+    telemetry_sdk_name: str
+    telemetry_sdk_version: str
+    service_name: str
+    span_name: str
+    model: str
+    greptime_value: str
+    greptime_timestamp: float
+
+
+class LlmCompletionTokenCost(object):
+    table_name = "llm_completion_tokens_cost"
+
+    telemetry_sdk_language: str
+    telemetry_sdk_name: str
+    telemetry_sdk_version: str
+    service_name: str
+    span_name: str
+    model: str
+    greptime_value: str
+    greptime_timestamp: float
+
+
+class Tables(object):
+    llm_trace = "llm_traces_preview_v01"
+    llm_prompt_tokens = "llm_prompt_tokens"
+    llm_prompt_tokens_cost = "llm_prompt_tokens_cost"
+    llm_completion_tokens = "llm_completion_tokens"
+    llm_completion_tokens_cost = "llm_completion_tokens_cost"
diff --git a/tests/integrations/openai_tracker/__init__.py b/tests/integrations/openai_tracker/__init__.py
new file mode 100644
index 0000000..68ecdd1
--- /dev/null
+++ b/tests/integrations/openai_tracker/__init__.py
@@ -0,0 +1,10 @@
+from openai import AsyncOpenAI
+from openai import OpenAI
+
+from greptimeai import openai_patcher  # type: ignore
+
+async_client = AsyncOpenAI()
+openai_patcher.setup(client=async_client)
+
+client = OpenAI()
+openai_patcher.setup(client=client)
diff --git a/tests/integrations/openai_tracker/test_sync.py b/tests/integrations/openai_tracker/test_sync.py
new file mode 100644
index 0000000..c23e7ef
--- /dev/null
+++ b/tests/integrations/openai_tracker/test_sync.py
@@ -0,0 +1,43 @@
+import time
+import uuid
+
+import pytest
+
+from ..database.db import (
+    get_trace_data,
+    truncate_tables,
+)
+from ..openai_tracker import client
+from greptimeai.openai_patcher import _collector  # type: ignore
+
+
+@pytest.fixture
+def _truncate_tables():
+    truncate_tables()
+    yield
+
+
+def test_chat_completion(_truncate_tables):
+    user_id = str(uuid.uuid4())
+    resp = client.chat.completions.create(
+        messages=[
+            {
+                "role": "user",
+                "content": "1+1=",
+            }
+        ],
+        model="gpt-3.5-turbo",
+        user=user_id,
+        seed=1,
+    )
+    assert resp.choices[0].message.content == "2"
+
+    _collector._collector._force_flush()
+    time.sleep(6)
+    trace = get_trace_data(user_id)
+
+    assert resp.model == trace[0]
+
+    if resp.usage:
+        assert resp.usage.prompt_tokens == trace[1]
+        assert resp.usage.completion_tokens == trace[2]
diff --git a/tests/langchain/__init__.py b/tests/langchain/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/utils/__init__.py b/tests/utils/__init__.py
new file mode 100644
index 0000000..e69de29