docker实现locust+prometheus+grafana性能测试监控

1. Locust + Prometheus + Grafana

简单总结起来就是：实现一个Locust的prometheus的exporter，将数据导入prometheus，然后使用grafana进行数据展示。

2.安装python和docker，整个项目框架如下图所示

commmon可以放置常用的方法，比如获取数据，mysql操作等方法

data放置数据

prometheus_exporter采集locust的性能结果， locust结果的采集在boomer 项目下有一个年久失修的 exporter 实现——prometheus_exporter.py

import six
from itertools import chain
from flask import request, Response
from locust import stats as locust_stats, runners as locust_runners
from locust import User, task, events
from prometheus_client import Metric, REGISTRY, exposition


# This locustfile adds an external web endpoint to the locust master, and makes it serve as a prometheus exporter.
# Runs it as a normal locustfile, then points prometheus to it.
# locust -f prometheus_exporter.py --master

# Lots of code taken from [mbolek's locust_exporter](https://github.com/mbolek/locust_exporter), thx mbolek!


class LocustCollector(object):
    registry = REGISTRY

    def __init__(self, environment, runner):
        self.environment = environment
        self.runner = runner

    def collect(self):
        # collect metrics only when locust runner is hatching or running.
        runner = self.runner

        if runner and runner.state in (locust_runners.STATE_SPAWNING, locust_runners.STATE_RUNNING):
            stats = []
            for s in chain(locust_stats.sort_stats(runner.stats.entries), [runner.stats.total]):
                stats.append({
                    "method": s.method,
                    "name": s.name,
                    "num_requests": s.num_requests,
                    "num_failures": s.num_failures,
                    "avg_response_time": s.avg_response_time,
                    "min_response_time": s.min_response_time or 0,
                    "max_response_time": s.max_response_time,
                    "current_rps": s.current_rps,
                    "median_response_time": s.median_response_time,
                    "ninetieth_response_time": s.get_response_time_percentile(0.9),
                    # only total stats can use current_response_time, so sad.
                    # "current_response_time_percentile_95": s.get_current_response_time_percentile(0.95),
                    "avg_content_length": s.avg_content_length,
                    "current_fail_per_sec": s.current_fail_per_sec
                })

            # perhaps StatsError.parse_error in e.to_dict only works in python slave, take notices!
            errors = [e.to_dict() for e in six.itervalues(runner.stats.errors)]

            metric = Metric('locust_user_count', 'Swarmed users', 'gauge')
            metric.add_sample('locust_user_count', value=runner.user_count, labels={})
            yield metric

            metric = Metric('locust_errors', 'Locust requests errors', 'gauge')
            for err in errors:
                metric.add_sample('locust_errors', value=err['occurrences'],
                                  labels={'path': err['name'], 'method': err['method'],
                                          'error': err['error']})
            yield metric

            is_distributed = isinstance(runner, locust_runners.MasterRunner)
            if is_distributed:
                metric = Metric('locust_slave_count', 'Locust number of slaves', 'gauge')
                metric.add_sample('locust_slave_count', value=len(runner.clients.values()), labels={})
                yield metric

            metric = Metric('locust_fail_ratio', 'Locust failure ratio', 'gauge')
            metric.add_sample('locust_fail_ratio', value=runner.stats.total.fail_ratio, labels={})
            yield metric

            metric = Metric('locust_state', 'State of the locust swarm', 'gauge')
            metric.add_sample('locust_state', value=1, labels={'state': runner.state})
            yield metric

            stats_metrics = ['avg_content_length', 'avg_response_time', 'current_rps', 'current_fail_per_sec',
                             'max_response_time', 'ninetieth_response_time', 'median_response_time',
                             'min_response_time',
                             'num_failures', 'num_requests']

            for mtr in stats_metrics:
                mtype = 'gauge'
                if mtr in ['num_requests', 'num_failures']:
                    mtype = 'counter'
                metric = Metric('locust_stats_' + mtr, 'Locust stats ' + mtr, mtype)
                for stat in stats:
                    # Aggregated stat's method label is None, so name it as Aggregated
                    # locust has changed name Total to Aggregated since 0.12.1
                    if 'Aggregated' != stat['name']:
                        metric.add_sample('locust_stats_' + mtr, value=stat[mtr],
                                          labels={'path': stat['name'], 'method': stat['method']})
                    else:
                        metric.add_sample('locust_stats_' + mtr, value=stat[mtr],
                                          labels={'path': stat['name'], 'method': 'Aggregated'})
                yield metric


@events.init.add_listener
def locust_init(environment, runner, **kwargs):
    print("locust init event received")
    if environment.web_ui and runner:
        @environment.web_ui.app.route("/export/prometheus")
        def prometheus_exporter():
            registry = REGISTRY
            encoder, content_type = exposition.choose_encoder(request.headers.get('Accept'))
            if 'name[]' in request.args:
                registry = REGISTRY.restricted_registry(request.args.get('name[]'))
            body = encoder(registry)
            return Response(body, content_type=content_type)

        REGISTRY.register(LocustCollector(environment, runner))


class Dummy(User):
    @task(2)
    def hello(self):
        pass

docker-compose 或者swarm管理locust master和work，这里是单机主从配置

locustfile.py为接口测试设计，下为网上的一个demo

from locust import HttpLocust, TaskSet, task, between

class NoSlowQTaskSet(TaskSet):

    def on_start(self):
        # 登录
        data = {"username": "admin", "password": "admin"}
        self.client.post("/user/login", json=data)

    @task(50)
    def getTables(self):
        r = self.client.get("/newsql/api/getTablesByAppId?appId=1")

    @task(50)
    def get_apps(self):
        r = self.client.get("/user/api/getApps")


class MyLocust(HttpUser):
    task_set = NoSlowQTaskSet
    host = "http://localhost:9528"

prometheus.yml文件 prometheus配置文件

global:
  scrape_interval: 15s
  evaluation_interval: 15s
 
scrape_configs:
  - job_name: prometheus
    static_configs:
      - targets: ['localhost:9090']
        labels:
          instance: prometheus
          
  - job_name: locust
    metrics_path: '/export/prometheus'
    static_configs:
      - targets: ['192.168.20.8:8089']
        labels:
          instance: locust

 3.用docker built构建locust镜像，我这里镜像名为locust:v1，然后执行docker init ，docker stack deploy -c ./docker-compose.yml  test (或者执行docker-compose up -d)启动locust matser和work，此时就可以访问locust页面，http://ip:8089，可以进行性能测试

4.访问http://masterip:8089/export/prometheus是否正常，也可以执行性能测试，查看该页面收集数据是否正常

5.安装prometheus和grafana镜像

docker pull prom/prometheus

docker pull grafana/grafana

并用下面命令启动两个服务

docker run -d --name prometheus -p 9090:9090 -v $PWD/prometheus.yml:/etc/prometheus/prometheus.yml prom/prometheus

docker run -d --name grafana -p 3000:3000 grafana/grafana

启动之后可通过http://ip:9090/查看prometheus是否启动正常

6.网页端访问 localhost:3000 验证部署成功

 选择添加 prometheus 数据源

然后再配置模板，导入模板有几种方式，选择一种方式将dashboard模板导入

最后效果展示