sentinel控制台监控数据持久化【MySQL】

创建监控数据表 CREATE TABLE `sentinel_metric1` ( `id` INT NOT NULL AUTO_INCREMENT COMMENT ‘id，主键‘, `gmt_create` DATETIME COMMENT ‘创建时间‘, `gmt_modified` DATETIME COMMENT ‘修改时间‘, `app` VARCHAR(100) COMMENT ‘应用名称‘, `timestamp` DATETIME COMMENT ‘统计时间‘, `resource` VARCHAR(500) COMMENT ‘资源名称‘, `pass_qps` INT COMMENT ‘通过qps‘, `success_qps` INT COMMENT ‘成功qps‘, `block_qps` INT COMMENT ‘限流qps‘, `exception_qps` INT COMMENT ‘发送异常的次数‘, `rt` DOUBLE COMMENT ‘所有successQps的rt的和‘, `_count` INT COMMENT ‘本次聚合的总条数‘, `resource_code` INT COMMENT ‘资源的hashCode‘, INDEX app_idx(`app`) USING BTREE, INDEX resource_idx(`resource`) USING BTREE, INDEX timestamp_idx(`timestamp`) USING BTREE, PRIMARY KEY (`id`) ) ENGINE=INNODB DEFAULT CHARSET=utf8;

注：app、resource、timestamp在查询语句的where条件中用到，因此给它们建立索引提高查询速度；

     count是MySQL的关键字，因此加上_前缀。

 

持久层选用Spring Data JPA框架，在pom中引入starter依赖：

<dependency>
  <groupId>org.springframework.boot</groupId>
  <artifactId>spring-boot-starter-data-jpa</artifactId>
  <version>${spring.boot.version}</version>
</dependency>

在datasource.entity包下，新建jpa包，下面新建sentinel_metric表对应的实体类MetricPO：

package com.taobao.csp.sentinel.dashboard.datasource.entity.jpa;

import javax.persistence.*;
import java.io.Serializable;
import java.util.Date;

/**
 * @author cdfive
 * @date 2018-09-14
 */
@Entity
@Table(name = "sentinel_metric")
public class MetricPO implements Serializable {

    private static final long serialVersionUID = 7200023615444172715L;

    /**id，主键*/
    @Id
    @GeneratedValue
    @Column(name = "id")
    private Long id;

    /**创建时间*/
    @Column(name = "gmt_create")
    private Date gmtCreate;

    /**修改时间*/
    @Column(name = "gmt_modified")
    private Date gmtModified;

    /**应用名称*/
    @Column(name = "app")
    private String app;

    /**统计时间*/
    @Column(name = "timestamp")
    private Date timestamp;

    /**资源名称*/
    @Column(name = "resource")
    private String resource;

    /**通过qps*/
    @Column(name = "pass_qps")
    private Long passQps;

    /**成功qps*/
    @Column(name = "success_qps")
    private Long successQps;

    /**限流qps*/
    @Column(name = "block_qps")
    private Long blockQps;

    /**发送异常的次数*/
    @Column(name = "exception_qps")
    private Long exceptionQps;

    /**所有successQps的rt的和*/
    @Column(name = "rt")
    private Double rt;

    /**本次聚合的总条数*/
    @Column(name = "_count")
    private Integer count;

    /**资源的hashCode*/
    @Column(name = "resource_code")
    private Integer resourceCode;

    // getter setter省略
}

该类也是参考MetricEntity创建，加上了JPA的注解，比如@Table指定表名，@Entity标识为实体，@Id、@GeneratedValue设置id字段为自增主键等；

 

在resources目录下的application.properties文件中，增加数据源和JPA(hibernate)的配置：

# datasource
spring.datasource.driver-class-name=com.mysql.jdbc.Driver
spring.datasource.url=${spring.datasource.url}
spring.datasource.username=${spring.datasource.username}
spring.datasource.password=${spring.datasource.password}

# spring data jpa
spring.jpa.hibernate.ddl-auto=none
spring.jpa.hibernate.use-new-id-generator-mappings=false
spring.jpa.database-platform=org.hibernate.dialect.MySQLDialect
spring.jpa.show-sql=false

这里数据库连接(url)、用户名(username)、密码(password)用${xxx}占位符，这样可以通过maven的pom.xml添加profile配置不同环境(开发、测试、生产) 或 从配置中心读取参数。

 

接着在InMemoryMetricsRepository所在的repository.metric包下新建JpaMetricsRepository类，实现MetricsRepository<MetricEntity>接口：

package com.taobao.csp.sentinel.dashboard.repository.metric;

import com.alibaba.csp.sentinel.util.StringUtil;
import com.taobao.csp.sentinel.dashboard.datasource.entity.MetricEntity;
import com.taobao.csp.sentinel.dashboard.datasource.entity.jpa.MetricPO;
import org.springframework.beans.BeanUtils;
import org.springframework.stereotype.Repository;
import org.springframework.transaction.annotation.Transactional;
import org.springframework.util.CollectionUtils;

import javax.persistence.EntityManager;
import javax.persistence.PersistenceContext;
import javax.persistence.Query;
import java.time.Instant;
import java.util.*;
import java.util.stream.Collectors;

/**
 * @author cdfive
 * @date 2018-09-17
 */
@Transactional
@Repository("jpaMetricsRepository")
public class JpaMetricsRepository implements MetricsRepository<MetricEntity> {

    @PersistenceContext
    private EntityManager em;

    @Override
    public void save(MetricEntity metric) {
        if (metric == null || StringUtil.isBlank(metric.getApp())) {
            return;
        }

        MetricPO metricPO = new MetricPO();
        BeanUtils.copyProperties(metric, metricPO);
        em.persist(metricPO);
    }

    @Override
    public void saveAll(Iterable<MetricEntity> metrics) {
        if (metrics == null) {
            return;
        }

        metrics.forEach(this::save);
    }

    @Override
    public List<MetricEntity> queryByAppAndResourceBetween(String app, String resource, long startTime, long endTime) {
        List<MetricEntity> results = new ArrayList<MetricEntity>();
        if (StringUtil.isBlank(app)) {
            return results;
        }

        if (StringUtil.isBlank(resource)) {
            return results;
        }

        StringBuilder hql = new StringBuilder();
        hql.append("FROM MetricPO");
        hql.append(" WHERE app=:app");
        hql.append(" AND resource=:resource");
        hql.append(" AND timestamp>=:startTime");
        hql.append(" AND timestamp<=:endTime");

        Query query = em.createQuery(hql.toString());
        query.setParameter("app", app);
        query.setParameter("resource", resource);
        query.setParameter("startTime", Date.from(Instant.ofEpochMilli(startTime)));
        query.setParameter("endTime", Date.from(Instant.ofEpochMilli(endTime)));

        List<MetricPO> metricPOs = query.getResultList();
        if (CollectionUtils.isEmpty(metricPOs)) {
            return results;
        }

        for (MetricPO metricPO : metricPOs) {
            MetricEntity metricEntity = new MetricEntity();
            BeanUtils.copyProperties(metricPO, metricEntity);
            results.add(metricEntity);
        }

        return results;
    }

    @Override
    public List<String> listResourcesOfApp(String app) {
        List<String> results = new ArrayList<>();
        if (StringUtil.isBlank(app)) {
            return results;
        }

        StringBuilder hql = new StringBuilder();
        hql.append("FROM MetricPO");
        hql.append(" WHERE app=:app");
        hql.append(" AND timestamp>=:startTime");

        long startTime = System.currentTimeMillis() - 1000 * 60;
        Query query = em.createQuery(hql.toString());
        query.setParameter("app", app);
        query.setParameter("startTime", Date.from(Instant.ofEpochMilli(startTime)));

        List<MetricPO> metricPOs = query.getResultList();
        if (CollectionUtils.isEmpty(metricPOs)) {
            return results;
        }

        List<MetricEntity> metricEntities = new ArrayList<MetricEntity>();
        for (MetricPO metricPO : metricPOs) {
            MetricEntity metricEntity = new MetricEntity();
            BeanUtils.copyProperties(metricPO, metricEntity);
            metricEntities.add(metricEntity);
        }

        Map<String, MetricEntity> resourceCount = new HashMap<>(32);

        for (MetricEntity metricEntity : metricEntities) {
            String resource = metricEntity.getResource();
            if (resourceCount.containsKey(resource)) {
                MetricEntity oldEntity = resourceCount.get(resource);
                oldEntity.addPassQps(metricEntity.getPassQps());
                oldEntity.addRtAndSuccessQps(metricEntity.getRt(), metricEntity.getSuccessQps());
                oldEntity.addBlockQps(metricEntity.getBlockQps());
                oldEntity.addExceptionQps(metricEntity.getExceptionQps());
                oldEntity.addCount(1);
            } else {
                resourceCount.put(resource, MetricEntity.copyOf(metricEntity));
            }
        }

        // Order by last minute b_qps DESC.
        return resourceCount.entrySet()
                .stream()
                .sorted((o1, o2) -> {
                    MetricEntity e1 = o1.getValue();
                    MetricEntity e2 = o2.getValue();
                    int t = e2.getBlockQps().compareTo(e1.getBlockQps());
                    if (t != 0) {
                        return t;
                    }
                    return e2.getPassQps().compareTo(e1.getPassQps());
                })
                .map(Map.Entry::getKey)
                .collect(Collectors.toList());
    }
}

参考InMemoryMetricsRepository类来实现，将其中用map存储和查询的部分改为用JPA实现：

save方法，将MetricEntity转换为MetricPO类，调用EntityManager类的persist方法即可；

saveAll方法，循环调用save；

queryByAppAndResourceBetween、listResourcesOfApp编写查询即可。

 

最后一步，在MetricsRepository、MetricController类注入的metricStore属性，加上@Qualifier("jpaMetricsRepository")注解：

@Qualifier("jpaMetricsRepository")
@Autowired
private MetricsRepository<MetricEntity> metricStore;

 

至此，监控数据MySQL持久化就完成了，得益于sentinel良好的Repository接口设计，是不是很简单：）

 

来验证下成果：

设置sentinel-dashboard工程启动参数：-Dserver.port=8080 -Dcsp.sentinel.dashboard.server=localhost:8080 -Dproject.name=sentinel-dashboard

启动工程，打开http://localhost:8080，查看不同的页面均显示正常，执行sql查询sentinel_metric表已有数据。

 -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

思考：

个人感觉sentinel控制台默认的实现类InMemoryMetricsRepository挺赞的，虽然内存存储重启会清空数据，如果没有对历史数据查询的需求应用于生产环境是没问题的；

其中如何用内存存储，包括保存、查询以及排序等代码都值得学习。

 -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

参考：

https://github.com/alibaba/Sentinel/wiki/控制台

https://github.com/alibaba/Sentinel/wiki/在生产环境中使用-Sentinel-控制台

 

sentinel控制台监控数据持久化【MySQL】

标签：time   alt+   static   org   ase   lan   prope   csp   ddc