Python实现获取sonarqube数据

 更新时间:2023年05月18日 14:58:50   作者:Michaelwubo  
sonarqube是一款代码分析的工具,可以对通过soanrScanner扫描后的数据传递给sonarqube进行分析,本文为大家整理了Python获取sonarqube数据的方法,需要的可以参考下

1.sonarqube是一款代码分析的工具,通过soanrScanner扫描后的数据传递给sonarqube进行分析

2.sonarqube社区版没有对c++类代码的分析,但是可以找到一个开源的包,安装即可,扫描的话可以使用cppcheck来进行扫描

  • 安装python对于sonarqube的api包:python-sonarqube-api
  • 建立sonarqube连接

from sonarqube import SonarQubeClient
sonar = SonarQubeClient(
    sonarqube_url="http://192.168.xx.xx:9000",
    username='admin',
    password='admin'
)

使用:建议大家先参考sonarqube的python-api

Welcome to SonarQube Client with Python’s documentation! — SonarQube Client with Python 1.3.5 documentation

使用示例

# 通过项目名称获取id
# 传递参数:创建分析配置文件时候的项目名称
component = sonar.components.get_project_component_and_ancestors("python_test")
 
# 获取任务
# 参数1:上一步获取的component值
# 参数2:逗号分割的状态值
tasks1 = sonar.ce.search_tasks(
    componentId="AX5v36mo0Npud3J2od3a",
    status="FAILED,CANCELED,PENDING,IN_PROGRESS,SUCCESS"
)
 
# 获取所有项目
projects = list(sonar.projects.search_projects())
 
# 获取这个项目下最近一次分析任务的详情
"""
componentKeys:某一个项目名称
types:参数类型
CODE_SMELL==异常
BUG==bug
VULNERABILITY==漏洞
SECURITY_HOTSPOT==
"""
issues2 = list(sonar.issues.search_issues(componentKeys="python_s", types="CODE_SMELL"))

通过metricKeys参数获取这个项目中需要的值

# 参数1:component 项目名称
# 参数2:metricKeys 想要获取的某个值,逗号分割
component_data = sonar.measures.get_component_with_specified_measures(
    component="python_test",
    metricKeys="functions,classes"
)['component']['measures']
# 目前已收集的值和含义
'''
ncloc==总代码长度
ncloc_language_distribution==其中每种语言的行数
bugs==bug数
vulnerabilities==漏洞数
code_smells==异常数
duplicated_lines_density==重复度百分比
coverage==代码覆盖率百分比
files==文件数量
functions==方法数量
classes==类数量
'''
[root@localhost data]# cat wubo.py 
#!/bin/python3
# encoding: utf-8
from sonarqube import SonarQubeClient
from operator import itemgetter
import json
import csv
import os
import time
import shutil
class SonarQube:
    def __init__(self,url,username="admin",password="123456aA") -> None:
        username = username
        password = password
        sonarqube_url = url
        self.client = SonarQubeClient(username = username,password = password,sonarqube_url = sonarqube_url)
    def getProjects(self):
        """获取项目列表"""
        projects = self.client.projects.search_projects().get("components")
        return projects
    def getIssues(self,jettech_project_name,issues_type):
        """获取项目问题列表"""
        #projects = self.client.issues.search_issues(componentKeys="jettoloader-pressurecommon",types="BUG",s="FILE_LINE",resolved="false",ps=1,organization="default-organization",facets="authors",additionalFields="_all")
        projects = self.client.issues.search_issues(componentKeys=jettech_project_name,types=issues_type,s="FILE_LINE",resolved="false",ps=1,organization="default-organization",facets="authors",additionalFields="_all")
        list_issues = projects["facets"][0]["values"]
        #list_issues_name_count = []
        #for list_issues_item in list_issues:
        #    list_issues_name_count.append(list_issues_item["val"])
        #    list_issues_name_count.append(list_issues_item["count"])
        #print(list_issues)
        #list_issues[0]["val"])
        #list_issues[0]["count"])
        return list_issues
    def getMessages(self,component):
        """ 获取项目各个参数数据"""
        #metricKeys = "alert_status,bugs,,vulnerabilities,security_rating,code_smells,duplicated_lines_density,coverage,ncloc"
        metricKeys = "bugs,,vulnerabilities"
        messages = []
        messages.append(self.client.measures.get_component_with_specified_measures(component, metricKeys))
        return messages[0]
    def getMeasures(self,component,message):
        measures = []
        measures.insert(0,component)
        measures_all = message.get("component").get("measures")
        for measure_item in measures_all:
            measures_type = measure_item.get("metric")
            if "bugs" == measures_type:
                measures.insert(1,measure_item.get("value"))
            if "vulnerabilities" == measures_type:
                measures.insert(2,measure_item.get("value"))
        return measures
 
class CSV:
    def __init__(self,filepath,filename) -> None:
        self.filepath = filepath
        self.filename = filename
    def csv_write(self,project_measure_all):
        #header = ['1project_name','2bugs','3vulnerabilities']
        with open(self.filepath+"/"+self.filename, 'a') as file_obj:
            writer = csv.writer(file_obj)
            #writer.writerow(header)
            for p in project_measure_all:
                writer.writerow(p)
 
    def csv_sort(self):
        datas=[]
        with open(self.filepath+"/"+self.filename, 'r') as f:
            table = []
            for line in f:
                line = line.replace("\n","").replace("\r","")
                col = line.split(',')
                col[0] = str(col[0])
                col[1] = col[1].strip("\n")
                table.append(col)
            table_sorted = sorted(table, key=itemgetter(0), reverse=False)  # 精确的按照第1列排序
            for row in table_sorted:
                datas.append(row)
        f.close()
        with open(self.filepath+"/"+self.filename,"w", newline='') as csvfile:
            writer = csv.writer(csvfile)
            for data in datas:
                writer.writerow(data)
        csvfile.close()
 
    def csv_insert(self):
       header = 'project_name,bugs,vulnerabilities'
       with open(self.filepath+"/"+self.filename, 'r+', encoding='utf-8') as f:
           content = f.read()
           f.seek(0, 0)
           f.write(header + '\n' + content)
       f.close()
 
    def csv_delete(self):
        if (os.path.exists(self.filepath)):
            shutil.rmtree(self.filepath,ignore_errors=True)
   
    def csv_sum(self):
        with open(self.filepath+"/"+self.filename) as fin:
            readline_item=fin.readline()
            total_bug_api = 0
            total_bug_manager = 0
            total_bug_loader = 0
            total_bug_ui = 0
            total_vulnerabilities_api = 0
            total_vulnerabilities_manager = 0
            total_vulnerabilities_loader = 0
            total_vulnerabilities_ui = 0
            for row in csv.reader(fin):
                row_project_name=row[0].split("-")[0]
                if "jettoapi" == row_project_name:
                     total_bug_api += int(row[1])
                     total_vulnerabilities_api += int(row[2])
                if "jettoloader" == row_project_name:
                     total_bug_loader += int(row[1])
                     total_vulnerabilities_loader += int(row[2])
                if "jettomanager" == row_project_name:
                     total_bug_manager += int(row[1])
                     total_vulnerabilities_manager += int(row[2])
                if "jettoui" == row_project_name:
                     total_bug_ui += int(row[1])
                     total_vulnerabilities_ui += int(row[2])
        fin.close()
 
        header_kong = ['','','']
        header_api = ['jettoapi','bug总数',str(total_bug_api),'vulnerabilities总数',str(total_vulnerabilities_api)]
        header_loader = ['jettoloader','bug总数',str(total_bug_loader),'vulnerabilities总数',str(total_vulnerabilities_loader)]
        header_manager = ['jettomanager','bug总数',str(total_bug_manager),'vulnerabilities总数',str(total_vulnerabilities_manager)]
        header_ui = ['jettoui','bug总数',str(total_bug_ui),'vulnerabilities总数',str(total_vulnerabilities_ui)]
        with open(self.filepath+"/"+self.filename, 'a') as file_obj:
            writer = csv.writer(file_obj)
            writer.writerow(header_kong)
            writer.writerow(header_api)
            writer.writerow(header_loader)
            writer.writerow(header_manager)
            writer.writerow(header_ui)
        file_obj.close()
                
 
class SCP: 
    def __init__(self,localdir,remoteip,remotedir) -> None:
        self.localdir = localdir
        self.remoteip = remoteip
        self.remotedir = remotedir
    def scp_operate(self):
        os.system('scp -r "%s" "%s:%s"' % (self.localdir, self.remoteip, self.remotedir))
 
def main():
    sonarQube = SonarQube(url='http://172.16.10.1:9000/')
    all_project_info = sonarQube.getProjects()
    project_measure_all=[]
    project_issues_all=[]
    for project_info in all_project_info:
        component = project_info.get("key")
        message = sonarQube.getMessages(component)
        measure = sonarQube.getMeasures(component,message)
        project_issues=[]
        list_issues_s = sonarQube.getIssues(component,"BUG")
        project_issues.append(component)
        for list_issues_s_item in list_issues_s:
            project_issues.append(list_issues_s_item["val"])
            project_issues.append(list_issues_s_item["count"])
        project_measure_all.extend([tuple(measure)])
        project_issues_all.extend([tuple(project_issues)])
        print([tuple(measure)])
        #print(project_issues_all)
 
    filepath=time.strftime("%Y-%m-%d")
    filename="jettech_sornar_"+filepath+"_projects.csv"
    filename_isuess="jettech_sornar_"+filepath+"_projects_iseuss.csv"
    if not os.path.exists(filepath):
        os.makedirs(filepath)
    if  os.path.exists(filepath+"/"+filename):
        os.remove(filepath+"/"+filename)
    if  os.path.exists(filepath+"/"+filename_isuess):
        os.remove(filepath+"/"+filename_isuess)
    csv_obj = CSV(filepath=filepath,filename=filename)
    csv_obj_isuess = CSV(filepath=filepath,filename=filename_isuess)
    csv_obj.csv_write(project_measure_all)
    csv_obj_isuess.csv_write(project_issues_all)
    csv_obj.csv_sort()
    csv_obj.csv_insert()
    csv_obj.csv_sum()
 
    localdir=filepath
    remoteip="192.168.1.99"
    remotedir="/product/gohttpserver/data/sornar/" 
    scp_obj = SCP(localdir,remoteip,remotedir)
    scp_obj.scp_operate()
    csv_obj.csv_delete()
 
if __name__== "__main__" :
    main()

到此这篇关于Python实现获取sonarqube数据的文章就介绍到这了,更多相关Python获取sonarqube数据内容请搜索脚本之家以前的文章或继续浏览下面的相关文章希望大家以后多多支持脚本之家!

相关文章

  • 详解Python Requests库中的重试策略

    详解Python Requests库中的重试策略

    在Python中,requests库是处理HTTP请求的标准工具之一,这篇文章主要为大家介绍了如何在requests中实现请求的自动重试,感兴趣的小伙伴可以跟随小编一起学习一下
    2023-11-11
  • 解决Python调用df.to_csv()出现中文乱码的问题

    解决Python调用df.to_csv()出现中文乱码的问题

    在Python使用df.to_csv()时,若出现中文乱码,可通过加入参数encoding="utf_8_sig"解决,"utf-8"编码不包含BOM,直接处理文件时会将BOM误读为内容;而"utf_8_sig"会识别并处理BOM,避免乱码,此方法为实践经验,供参考
    2024-09-09
  • python求列表交集的方法汇总

    python求列表交集的方法汇总

    这篇文章主要介绍了python求列表交集的方法,实例汇总了三个常用的方法,具有一定的参考借鉴价值,需要的朋友可以参考下
    2014-11-11
  • 关于文件Permission denied解决方案(pip)

    关于文件Permission denied解决方案(pip)

    这篇文章主要介绍了文件Permission denied解决方案(pip),具有很好的参考价值,希望对大家有所帮助,如有错误或未考虑完全的地方,望不吝赐教
    2023-08-08
  • 使用python编写监听端

    使用python编写监听端

    这篇文章主要为大家详细介绍了使用python编写监听端,具有一定的参考价值,感兴趣的小伙伴们可以参考一下
    2018-04-04
  • PyCharm中New Directory 和 New Python Package的区别解析

    PyCharm中New Directory 和 New Python 

    python package这是一个特殊的目录,因为在创建该python package的时候,系统会自动地生成一个py文件, init.py,这篇文章主要介绍了PyCharm中New Directory 和 New Python Package的区别,需要的朋友可以参考下
    2023-12-12
  • python字典DICT类型合并详解

    python字典DICT类型合并详解

    这篇文章主要为大家详细介绍了python字典DICT类型合并,具有一定的参考价值,感兴趣的小伙伴们可以参考一下
    2017-08-08
  • Python机器学习库scikit-learn入门开发示例

    Python机器学习库scikit-learn入门开发示例

    scikit-learn是一个开源Python语言机器学习工具包,它涵盖了几乎所有主流机器学习算法的实现,并且提供了一致的调用接口。它基于Numpy和scipy等Python数值计算库,提供了高效的算法实现
    2022-07-07
  • pytorch 状态字典:state_dict使用详解

    pytorch 状态字典:state_dict使用详解

    今天小编就为大家分享一篇pytorch 状态字典:state_dict使用详解,具有很好的参考价值,希望对大家有所帮助。一起跟随小编过来看看吧
    2020-01-01
  • Python Requests.post()请求失败时的retry设置方式

    Python Requests.post()请求失败时的retry设置方式

    这篇文章主要介绍了Python Requests.post()请求失败时的retry设置方式,具有很好的参考价值,希望对大家有所帮助,如有错误或未考虑完全的地方,望不吝赐教
    2023-08-08

最新评论