LLM + RAG + Vanna 综合实战
这篇文章介绍如何将LLM + RAG + Vanna结合起来，在一个项目中通过和LLM对话的方式查询数据库中的记录，并显示结果。而不需要人为的手动的通过执行SQL的方式来查询数据库记录。
OLLAMA
Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
Note: The default location to store models file is: /usr/share/ollama/.ollama/models, to change the location, refer to: https://github.com/ollama/ollama/blob/main/docs/faq.md#where-are-models-stored
Query Libraries:
library
Config proxy
sudo vi /etc/systemd/system/ollama.service
如果需要，增加http_proxy和https_proxy，
[Unit]
Description=Ollama Service
After=network-online.target[Service]
ExecStart=/usr/local/bin/ollama serve
User=ollama
Group=ollama
Restart=always
RestartSec=3
Environment="PATH=/home/sunny/bin:/home/sunny/conda/condabin:/home/sunny/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/home/sunny/yunjiang/tools/apache-maven-3.9.9/bin:/home/sunny/conda/bin:/home/sunny/conda/bin"Environment="http_proxy=http://child-prc.intel.com:913"
Environment="https_proxy=http://child-prc.intel.com:913"[Install]
WantedBy=default.target重启服务，
sudo systemctl daemon-reload
sudo systemctl restart ollama
Pull model
ollama pull llama3.2
List model
ollama list
Run model
ollama run llama3.2
Ask Question via Restful API
curl http://localhost:11434/api/generate -d '{ "model": "llama3.2", "prompt": "How are you today?", "stream": false}'
Install CrhomaDB
Refer to: https://blog.csdn.net/funnyrand/article/details/148132003
Install Vanna
Install Vanna
pip3 install vanna
For more details of Vanna, refer to:
Vanna.AI Documentation
https://github.com/vanna-ai/vanna?tab=readme-ov-file
Install ClickHouse Connect
pip3 install clickhouse_connect
Install MySQL Connect
pip3 install PyMySQL
Install ClickHouse
Refer to: ClickHouse总结_checkhouse数据库-CSDN博客
注：需要插入真实数据到ClickHouse，下面的部分仅做演示。
Python code to use Vanna
ntr_test1.py
from vanna.ollama import Ollama
from vanna.chromadb import ChromaDB_VectorStoreclass MyVanna(ChromaDB_VectorStore, Ollama):def __init__(self, config=None):ChromaDB_VectorStore.__init__(self, config=config)Ollama.__init__(self, config=config)def train():# The information schema query may need some tweaking depending on your database. This is a good starting point.df_information_schema = vn.run_sql("SELECT * FROM INFORMATION_SCHEMA.COLUMNS WHERE table_schema = 'ntr'")# This will break up the information schema into bite-sized chunks that can be referenced by the LLMplan = vn.get_training_plan_generic(df_information_schema)#print(plan)# If you like the plan, then uncomment this and run it to train#vn.train(plan=plan)# The following are methods for adding training data. Make sure you modify the examples to match your database.# DDL statements are powerful because they specify table names, colume names, types, and potentially relationshipsvn.train(ddl="""CREATE TABLE ntr.metadata (timestamp DateTime64(6),srcMac FixedString(17),destMac FixedString(17),srcIp IPv4,destIp IPv4,srcPort UInt16,destPort UInt16,etherType UInt16,ipProto UInt16,packetSize UInt16,offset UInt64,pcapFile String)ENGINE = MergeTree()PARTITION BY toStartOfHour(timestamp)ORDER BY (timestamp, srcIp, destIp, srcPort, destPort, etherType, ipProto);""")vn.train(ddl="""CREATE TABLE ntr.metadata_stats_all (timestamp DateTime,throughput Float64,packetCount UInt64,packetSizes UInt64)ENGINE = MergeTree()PARTITION BY toYYYYMMDD(timestamp)ORDER BY (timestamp);""")vn.train(ddl="""CREATE TABLE ntr.global_config (key String,value String,initVal String,regexp String) ENGINE = MergeTree()PARTITION BY (key)ORDER BY (key);""")# Sometimes you may want to add documentation about your business terminology or definitions.# vn.train(documentation="Our business defines OTIF score as the percentage of orders that are delivered on time and in full")# You can also add SQL queries to your training data. This is useful if you have some queries already laying around. You can just copy and paste those from your editor to begin generating new SQL.#vn.train(sql="select * from mydb.person where name='sunny';")# At any time you can inspect what training data the package is able to referencetraining_data = vn.get_training_data()training_data# Main
vn = MyVanna(config={'model': 'llama3.2'})vn.connect_to_clickhouse(host='172.17.28.100', dbname='ntr', user="default", password="changeme", port=8123)#train()# You can remove training data if there's obsolete/incorrect information.
# vn.remove_training_data(id='1-ddl')## Asking the AI
# Whenever you ask a new question, it will find the 10 most relevant pieces of training data and use it as part of the LLM prompt to generate the SQL.# vn.ask(question="How many metadata records with src IP address 117.35.136.101 in ntr?", allow_llm_to_see_data=True)
# vn.ask(question="Query metadata which src IP address is 117.35.136.101.")
# rt = vn.ask(question="How many items in table ntr.metadata witch src IP address 117.35.136.101?", visualize=True, allow_llm_to_see_data=True)
rt = vn.ask(question="Query items in ntr.metadata table which src IP address is 117.35.136.101.", visualize=False, allow_llm_to_see_data=True)
#print(rt)运行结果：
sunny@r05u33-nex-wvie-spr-cd:~/work/llm$ /usr/bin/env /bin/python3 /home/sunny/.vscode-server/extensions/ms-python.debugpy-2025.8.0-linux-x64/bundled/libs/debugpy/adapter/../../debugpy/launcher 39887 -- /home/sunny/work/llm/vanna/ntr/ntr_test1.py
<clickhouse_connect.driver.httpclient.HttpClient object at 0x7f1b477f29e0>
SQL Prompt: [{'role': 'system', 'content': "You are a SQL expert. Please help to generate a SQL query to answer the question. Your response should ONLY be based on the given context and follow the response guidelines and format instructions. \n===Tables \n\n CREATE TABLE ntr.metadata (\n timestamp DateTime64(6),\n srcMac FixedString(17),\n destMac FixedString(17),\n srcIp IPv4,\n destIp IPv4,\n srcPort UInt16,\n destPort UInt16,\n etherType UInt16,\n ipProto UInt16,\n packetSize UInt16,\n offset UInt64,\n pcapFile String\n )\n ENGINE = MergeTree()\n PARTITION BY toStartOfHour(timestamp)\n ORDER BY (\n timestamp, \n srcIp, \n destIp, \n srcPort, \n destPort, \n etherType, \n ipProto\n );\n \n\n\n CREATE TABLE ntr.metadata_stats_all (\n timestamp DateTime,\n throughput Float64,\n packetCount UInt64,\n packetSizes UInt64\n )\n ENGINE = MergeTree()\n PARTITION BY toYYYYMMDD(timestamp)\n ORDER BY (timestamp);\n \n\n\n CREATE TABLE ntr.global_config (\n key String,\n value String,\n initVal String,\n regexp String\n ) \n ENGINE = MergeTree()\n PARTITION BY (key)\n ORDER BY (key);\n \n\n===Response Guidelines \n1. If the provided context is sufficient, please generate a valid SQL query without any explanations for the question. \n2. If the provided context is almost sufficient but requires knowledge of a specific string in a particular column, please generate an intermediate SQL query to find the distinct strings in that column. Prepend the query with a comment saying intermediate_sql \n3. If the provided context is insufficient, please explain why it can't be generated. \n4. Please use the most relevant table(s). \n5. If the question has been asked and answered before, please repeat the answer exactly as it was given before. \n6. Ensure that the output SQL is SQL-compliant and executable, and free of syntax errors. \n"}, {'role': 'user', 'content': 'Query items in ntr.metadata table which src IP address is 117.35.136.101.'}, {'role': 'assistant', 'content': "SELECT * \nFROM ntr.metadata\nWHERE srcIp = '117.35.136.101'"}, {'role': 'user', 'content': 'Query metadata which src port is 61446'}, {'role': 'assistant', 'content': 'SELECT * FROM ntr.metadata WHERE srcPort = 61446'}, {'role': 'user', 'content': 'Query metadata which src port is 61446'}, {'role': 'assistant', 'content': 'SELECT DISTINCT srcPort\nFROM ntr.metadata\nWHERE srcPort = 61446'}, {'role': 'user', 'content': 'How many items in table ntr.metadata witch src IP address 117.35.136.101?'}, {'role': 'assistant', 'content': "SELECT COUNT(*) \nFROM ntr.metadata\nWHERE srcIp = '117.35.136.101'"}, {'role': 'user', 'content': 'How many items in table ntr.metadata?'}, {'role': 'assistant', 'content': 'SELECT COUNT(*) \nFROM ntr.metadata'}, {'role': 'user', 'content': 'Query items in ntr.metadata table which src IP address is 117.35.136.101.'}]
Info: Ollama parameters:
model=llama3.2:latest,
options={},
keep_alive=None
Info: Prompt Content:
[{"role": "system", "content": "You are a SQL expert. Please help to generate a SQL query to answer the question. Your response should ONLY be based on the given context and follow the response guidelines and format instructions. \n===Tables \n\n CREATE TABLE ntr.metadata (\n timestamp DateTime64(6),\n srcMac FixedString(17),\n destMac FixedString(17),\n srcIp IPv4,\n destIp IPv4,\n srcPort UInt16,\n destPort UInt16,\n etherType UInt16,\n ipProto UInt16,\n packetSize UInt16,\n offset UInt64,\n pcapFile String\n )\n ENGINE = MergeTree()\n PARTITION BY toStartOfHour(timestamp)\n ORDER BY (\n timestamp, \n srcIp, \n destIp, \n srcPort, \n destPort, \n etherType, \n ipProto\n );\n \n\n\n CREATE TABLE ntr.metadata_stats_all (\n timestamp DateTime,\n throughput Float64,\n packetCount UInt64,\n packetSizes UInt64\n )\n ENGINE = MergeTree()\n PARTITION BY toYYYYMMDD(timestamp)\n ORDER BY (timestamp);\n \n\n\n CREATE TABLE ntr.global_config (\n key String,\n value String,\n initVal String,\n regexp String\n ) \n ENGINE = MergeTree()\n PARTITION BY (key)\n ORDER BY (key);\n \n\n===Response Guidelines \n1. If the provided context is sufficient, please generate a valid SQL query without any explanations for the question. \n2. If the provided context is almost sufficient but requires knowledge of a specific string in a particular column, please generate an intermediate SQL query to find the distinct strings in that column. Prepend the query with a comment saying intermediate_sql \n3. If the provided context is insufficient, please explain why it can't be generated. \n4. Please use the most relevant table(s). \n5. If the question has been asked and answered before, please repeat the answer exactly as it was given before. \n6. Ensure that the output SQL is SQL-compliant and executable, and free of syntax errors. \n"}, {"role": "user", "content": "Query items in ntr.metadata table which src IP address is 117.35.136.101."}, {"role": "assistant", "content": "SELECT * \nFROM ntr.metadata\nWHERE srcIp = '117.35.136.101'"}, {"role": "user", "content": "Query metadata which src port is 61446"}, {"role": "assistant", "content": "SELECT * FROM ntr.metadata WHERE srcPort = 61446"}, {"role": "user", "content": "Query metadata which src port is 61446"}, {"role": "assistant", "content": "SELECT DISTINCT srcPort\nFROM ntr.metadata\nWHERE srcPort = 61446"}, {"role": "user", "content": "How many items in table ntr.metadata witch src IP address 117.35.136.101?"}, {"role": "assistant", "content": "SELECT COUNT(*) \nFROM ntr.metadata\nWHERE srcIp = '117.35.136.101'"}, {"role": "user", "content": "How many items in table ntr.metadata?"}, {"role": "assistant", "content": "SELECT COUNT(*) \nFROM ntr.metadata"}, {"role": "user", "content": "Query items in ntr.metadata table which src IP address is 117.35.136.101."}]
Info: Ollama Response:
model='llama3.2:latest' created_at='2025-05-30T08:21:59.889943887Z' done=True done_reason='stop' total_duration=8834583846 load_duration=6108105509 prompt_eval_count=679 prompt_eval_duration=2053141142 eval_count=22 eval_duration=651663457 message=Message(role='assistant', content="SELECT * \nFROM ntr.metadata\nWHERE srcIp = '117.35.136.101'", images=None, tool_calls=None)
LLM Response: SELECT *
FROM ntr.metadata
WHERE srcIp = '117.35.136.101'
SELECT *
FROM ntr.metadata
WHERE srcIp = '117.35.136.101'timestamp srcMac destMac srcIp ... ipProto packetSize offset pcapFile
0 2025-05-23 14:56:35.953557+08:00 b'e5:19:63:7b:de:b9' b'55:61:5b:5e:74:de' 117.35.136.101 ... 17 1514 56228199546 output_0_4_5_1747983214.pcap[1 rows x 14 columns]可以看到，其正确的输出了查询结果。