Docker和Dify學習筆記

文章目錄

1 docker學習
- 1.1 基本命令使用
- - 1.1.1 docker ps查看當前正在運行的鏡像
  - 1.1.2 docker stop停止容器
  - 1.1.3 docker compose容器編排
  - 1.1.4 docker網絡
  - - [1] 進入到容器里面敲命令
    - [2] docker network ls
    - [3] brige網絡模式下容器訪問宿主機的方式
2 Dify的安裝和基礎使用
- 2.1 下載dify的工程倉庫
- 2.2 創建.env配置文件
- 2.3 修改Nginx的端口
- 2.4 啟動Dify
- 2.5 添加一個本地模型
- - 2.5.1 添加一個大語言模型
  - 2.5.2 添加一個Embedding模型
3 基于Dify開發功能場景
- 3.1 Echarts繪圖
- 3.2 讓大模型能夠解析Json
4 大模型或者Dify中常見的參數
- 4.1 Temperature
4 Dify使用過程中常見問題
- 4.1 json超長了，超過80000個字符了
參考資料

1 docker學習

1.1 基本命令使用

1.1.1 docker ps查看當前正在運行的鏡像

PS E:\LargeModel\dify\docker> docker ps
CONTAINER ID   IMAGE                                       COMMAND                   CREATED          STATUS                    PORTS                                         NAMES
5b2b8c1636e7   langgenius/dify-web:1.0.0                   "/bin/sh ./entrypoin…"   6 minutes ago    Up 6 minutes              3000/tcp                                      thirsty_diffie
99740eb3c659   langgenius/dify-web:1.0.0                   "/bin/sh ./entrypoin…"   6 minutes ago    Up 6 minutes              3000/tcp                                      upbeat_euclid
f8108bddcff7   nginx:latest                                "sh -c 'cp /docker-e…"   11 minutes ago   Up 11 minutes             0.0.0.0:8001->80/tcp, 0.0.0.0:8443->443/tcp   docker-nginx-1
2794a7f88a69   langgenius/dify-api:1.0.0                   "/bin/bash /entrypoi…"   11 minutes ago   Up 11 minutes             5001/tcp                                      docker-api-1
d51ee39beb2c   langgenius/dify-api:1.0.0                   "/bin/bash /entrypoi…"   11 minutes ago   Up 11 minutes             5001/tcp                                      docker-worker-1
e008dc5e6386   ubuntu/squid:latest                         "sh -c 'cp /docker-e…"   11 minutes ago   Up 11 minutes             3128/tcp                                      docker-ssrf_proxy-1
8c966517876b   langgenius/dify-sandbox:0.2.10              "/main"                   11 minutes ago   Up 11 minutes (healthy)                                                 docker-sandbox-1
a2d137e0f516   langgenius/dify-plugin-daemon:0.0.3-local   "/bin/bash -c /app/e…"   11 minutes ago   Up 11 minutes             0.0.0.0:5003->5003/tcp                        docker-plugin_daemon-1
56f97d78ab4c   langgenius/dify-web:1.0.0                   "/bin/sh ./entrypoin…"   11 minutes ago   Up 11 minutes             3000/tcp                                      docker-web-1
cfe29ccee8df   postgres:15-alpine                          "docker-entrypoint.s…"   11 minutes ago   Up 11 minutes (healthy)   0.0.0.0:5432->5432/tcp                        docker-db-1
edd0a6879ba7   semitechnologies/weaviate:1.19.0            "/bin/weaviate --hos…"   11 minutes ago   Up 11 minutes                                                           docker-weaviate-1
191a080293e4   redis:6-alpine                              "docker-entrypoint.s…"   11 minutes ago   Up 11 minutes (healthy)   6379/tcp                                      docker-redis-1
PS E:\LargeModel\dify\docker>

1.1.2 docker stop停止容器

【停止指定名稱的容器】

PS E:\LargeModel\dify\docker> docker stop thirsty_diffie
thirsty_diffie

【停止所有當前在運行的容器】

PS E:\LargeModel\dify\docker> docker stop $(docker ps -aq)
5b2b8c1636e7
99740eb3c659
f8108bddcff7
2794a7f88a69
d51ee39beb2c
e008dc5e6386
8c966517876b
a2d137e0f516
56f97d78ab4c
cfe29ccee8df
edd0a6879ba7
191a080293e4
fe75be867cd5

1.1.3 docker compose容器編排

1.1.4 docker網絡

[1] 進入到容器里面敲命令

docker exec -it docker-api-1 /bin/bash

案例：安裝ping命令和telnet命令

root@cb7f80d95b40:/app/api# apt-get update
Get:1 http://deb.debian.org/debian bookworm InRelease [151 kB]
Get:2 http://deb.debian.org/debian bookworm-updates InRelease [55.4 kB]
Get:3 http://deb.debian.org/debian-security bookworm-security InRelease [48.0 kB]
Get:4 http://deb.debian.org/debian bookworm/main amd64 Packages [8792 kB]
Get:5 http://deb.debian.org/debian bookworm-updates/main amd64 Packages [13.5 kB]
Get:6 http://deb.debian.org/debian-security bookworm-security/main amd64 Packages [246 kB]
Fetched 9306 kB in 8min 19s (18.6 kB/s)
Reading package lists... Done
root@cb7f80d95b40:/app/api# apt-get install -y iputils-ping
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
The following additional packages will be installed:libcap2-bin libpam-cap
The following NEW packages will be installed:

[2] docker network ls

PS D:\LargeModel> docker network ls
NETWORK ID     NAME                        DRIVER    SCOPE
35353aad5d22   bridge                      bridge    local
6e02ea3b25c6   docker_default              bridge    local
5973270d4a91   docker_ssrf_proxy_network   bridge    local
503869a64910   host                        host      local
a0056c2c396d   none                        null      local

[3] brige網絡模式下容器訪問宿主機的方式

容器內可以使用host.docker.internal來代替主機的ip

2 Dify的安裝和基礎使用

2.1 下載dify的工程倉庫

git clone https://github.com/langgenius/dify.git
# 國內鏡像站
https://gitee.com/dify_ai/dify

2.2 創建.env配置文件

我們進入dify目錄下的docker目錄中,

# 以示例創建一個.env的文件，執行下面命令
cp .\.env.example .env

2.3 修改Nginx的端口

默認占用的是80和443端口，如果你本機已經部署了其他的應用，占了該端口，修改.env文件中的下面兩個變量

EXPOSE_NGINX_PORT=8001
EXPOSE_NGINX_SSL_PORT=8443

2.4 啟動Dify

docker compose up -d

2.5 添加一個本地模型

2.5.1 添加一個大語言模型

在這里插入圖片描述

這里需要注意，我的ollama是直接安裝在宿主機的。但是Dify是通過docker啟動起來的，這里涉及到docker和宿主機之間的通信。如果docker訪問宿主機，可以使用host.docker.internal域名，Docker的DNS可以解析這個域名。

2.5.2 添加一個Embedding模型

（1）安裝bge-m3模型

ollama pull bge-m3

（2）Dify中配置embedding模型
在這里插入圖片描述

3 基于Dify開發功能場景

3.1 Echarts繪圖

思路就是利用Dify的echarts渲染的能力(即使是Dify自帶的Echarts圖表生成工具也是輸出了一串echarts的配置字符串，甚至還沒有直接寫python代碼生成來的直接)。但是只能在工作流和chatflow里面使用。如果將繪圖的工作流集成到Agent里面的話，會導致大模型解析不了json，輸出不了內容了。

import json
import requests
from datetime import datetime, timedelta
import statistics"""這是一個try參數的機制
"""
def getValidResponse(url, headers, params):response = Nonefor resourceFlag in ['sw', 'lk', 'wlw']:params["resource_flag"] = resourceFlagresponse = requests.get(url= url, headers= headers, params= params, timeout= 10)# 檢查請求是否成功if response.status_code == 200:try:json_data = response.json()print(json_data)if (not "data" in json_data) or json_data["data"] is None or len(json_data["data"]) <= 0:continueelse:return responseexcept ValueError:continue        else:continue        """獲取某一個水庫某一段時間內的的水情數據。（主要是水位）參數：startDate: start_dt=2025-03-16+10:00:00endDate: end_dt=2025-03-17+11:00:00
"""
def getWaterStatus(guid: str, stcd: str, startDate: str, endDate: str) -> dict:url = "https://sk.hubeishuiyi.cn/services/1234567890ABCDEFGHIJKLMN/res_z_detail/param"headers = {"Apikey": "F1DBECD719108635189480CF60E6553ADB3109616426BD537F25A430DFC613B491A025C4A51E77FD08C6E5B7CBE05917A461286E7B6D69F1AB1B14F946149D2065B0C675F8FEDF4B9B05C1496881BC5A"}params = {"is_tb": "n","pageNo": 1,"pageSize":10000,"st_cd": stcd,"resource_flag":"sw","flag": 1,"start_dt": startDate,"end_dt": endDate,"res_cd": guid}response = getValidResponse(url, headers, params)# 檢查請求是否成功if response is not None and response.status_code == 200:try:json_data = response.json()# # 轉為雙引號的標準json字符串# json_data = json.dumps(json_data)# json_data = json.loads(json_data)print(json_data)if (not "data" in json_data) or json_data["data"] is None or len(json_data["data"]) <= 0:return {"result" : []} # 以時間作為橫坐標timeList = []# 水位數據waterLevelList = []# 蓄水量waterStorageList = []for dataObject in json_data["data"]:timeStr = dataObject["tm"]timeList.append(timeStr)waterLevel = dataObject["rz"]waterLevelList.append(waterLevel)waterStorage = dataObject["w"]waterStorageList.append(waterStorage)statisticData = {"time": timeList,"waterLevel": waterLevelList,"waterStorage": waterStorageList}return {"result": statisticData}except ValueError:return {"result" : "Failed to parse JSON response"}else:statisticData = {"failMsg": f"Request failed. url = {url}, params = {params}"}return {"result" : statisticData}def calcYAxisMinMax(nums: list):minValue = min(nums) - 8if minValue < 0:minValue = 0maxValue = max(nums) + 8return minValue, maxValue"""參數：data： tuple的列表。
"""
def create_dynamic_table_str(headers: list, data):table = "|"for header in headers:table += f" {header} |"table += "\n|"for _ in headers:table += " ------- |"for row in data:table += "\n|"for item in row:table += f" {item} |"return tabledef constructTableString(timeList: list, waterLevelList: list) -> str:timeList = timeList[-10:]waterLevelList = waterLevelList[-10:]headers = ['時間', '水位(m)']data = list(zip(timeList, waterLevelList))return create_dynamic_table_str(headers, data)def constructStatisticString(waterLevelList: list) -> str:headers = ['統計項', '統計值']data = []minValue = min(waterLevelList)data.append(("最小值", f"{minValue:.2f}"))maxValue = max(waterLevelList)data.append(("最大值", f"{maxValue:.2f}"))meanValue = sum(waterLevelList) / len(waterLevelList)data.append(("平均值", f"{meanValue:.2f}"))# 計算列表的中位數medianValue = statistics.median(waterLevelList)data.append(("中位數", f"{medianValue:.2f}"))# 標準差devValue = statistics.stdev(waterLevelList)data.append(("標準差", f"{devValue:.2f}"))return create_dynamic_table_str(headers, data=data)def drawEcharts(data: dict) -> dict:if data is None or len(data) <= 0:return {"result": ""}if "time" not in data or "waterLevel" not in data or "waterStorage" not in data:return {"result": ""}waterStatusData = data# 提取時間和對應的數據timeList = list(reversed(waterStatusData["time"]))waterLevelList = waterStatusData["waterLevel"]waterStorageList = waterStatusData["waterStorage"]if timeList is None or len(timeList) < 2:return {"result": ""}# 繪圖準備工作UTC_FORMAT = "%Y-%m-%d %H:%M:%S"startTime = datetime.strptime(timeList[0], UTC_FORMAT)endTime = datetime.strptime(timeList[-1], UTC_FORMAT)walterLevelMin, walterLevelMax = calcYAxisMinMax(waterLevelList)# 生成echarts配置echarts_config = {"color": ['#eb6877', '#0f91c4', '#46cbd4'],"title": {"subtext": f"{startTime.strftime('%m')}月{startTime.strftime('%d')}日-{endTime.strftime('%m')}月{endTime.strftime('%d')}日水位情況","left": 20},"tooltip": {"trigger": "axis","axisPointer": {"type": "cross"}},"legend": {# "data": ["水位", "最低氣溫", "降水"],"data": ["水位"],"right": 20},"xAxis": {"data": timeList,"axisLine": {"onZero": False}},"yAxis": [{"type": "value","name": "水位","position": "left","min": walterLevelMin,"max": walterLevelMax,"axisLabel": {"formatter": "{value} m"}},# {#     "type": "value",#     "name": "蓄水量",#     "position": "right",#     "axisLabel": {#         "formatter": "{value} m"#     }# }],"series": [{"name": "水位","type": "line","data": waterLevelList,"yAxisIndex": 0,"itemStyle": {"color": "#eae213"},"markPoint": {"data": [{"type": 'max'},{"type": 'min'},],# 設置為點"symbol": 'circle', # 調整點的大小"symbolSize": 8, "label": {"position": 'right',# 標簽字體加粗"fontWeight": 'bold', # 標簽字體大小"fontSize": 12 }},},# {#     "name": "蓄水量",#     "type": "bar",#     "smooth": True,#     "data": waterStorageList,#     "yAxisIndex": 0,#     "itemStyle": {#         "color": "#4bb2fa"#     }# },# {#     "name": "降水",#     "type": "bar",#     "smooth": True,#     "data": rainfall,#     "yAxisIndex": 1,#     "itemStyle": {#         "color": "#31e84f"#     }# }]}echartString = "```echarts\n" + json.dumps(echarts_config, indent=2, ensure_ascii=False) + "\n```"tableString = constructTableString(timeList, waterLevelList)statisticString = constructStatisticString(waterLevelList)# 生成輸出文件output = echartString + "\n\n"  \+ "最近十條數據展示：\n" + tableString + "\n\n" \+ "常用統計數據展示：\n" + statisticStringreturn {"result":output}def main(guid: str, stcdList: list) -> dict:if guid is None or stcdList is None or len(guid) <= 0 or len(stcdList) <= 0:return {"result" : ""}endDate = datetime.now()endDateStr = endDate.strftime("%Y-%m-%d %H:%M:%S")startDate = endDate - timedelta(days=15)startDateStr = startDate.strftime("%Y-%m-%d %H:%M:%S")result = getWaterStatus(guid=guid, stcd=stcdList[0], startDate=startDateStr, endDate=endDateStr)print(f"main.result = {result['result']}")markdownScript = drawEcharts(result["result"])print(markdownScript)return markdownScriptif __name__ == "__main__":# print(main('42022220001', ['61608180']))# print(main('42130330004', ["61608180"]))# print(main('42122350024', ['90021804']))print(main('42058340006', [ "90006379"]))

3.2 讓大模型能夠解析Json

4 大模型或者Dify中常見的參數

4.1 Temperature

LLM 生成是具有隨機性的，在模型的頂層通過選取不同預測概率的預測結果來生成最后的結果。我們一般可以通過控制 temperature 參數來控制 LLM 生成結果的隨機性與創造性。

Temperature 一般取值在 0~1 之間，當取值較低接近 0 時，預測的隨機性會較低，產生更保守、可預測的文本，不太可能生成意想不到或不尋常的詞。當取值較高接近 1 時，預測的隨機性會較高，所有詞被選擇的可能性更大，會產生更有創意、多樣化的文本，更有可能生成不尋常或意想不到的詞。

4 Dify使用過程中常見問題

4.1 json超長了，超過80000個字符了

這是因為Dify限制了默認的長度。本地部署的情況下可以修改.env配置文件中的相關變量數值。修改之后重啟整個服務。

常見的參數和含義入下圖所示：
在這里插入圖片描述
重啟服務所需要使用的命令：

docker compose down
docker compose up -d

參考資料

[1] https://mp.weixin.qq.com/s/n5GrGZ9hZmdhzt4avs1XSw
[2] https://wiki.eryajf.net/pages/674f53/
[3] https://zhuanlan.zhihu.com/p/20939683190
[4] https://dify.flowus.cn/haojixing/share/943044ea-005e-4e57-9e74-450700df71c2
[5] https://blog.csdn.net/luckcxy/article/details/144900399