对于系统服务的监控,tcp层通常是用heartbeat来进行,最简单的比如ping-pong。对于http层,来说springboot的actuator内置了/health的endpoint,很方便地规范了每个服务的健康状况的api,而且HealthIndicator可以自己去扩展,增加相关依赖服务的健康状态,非常灵活方便而且可扩展。

/health实例

{
  "status": "UP",
  "custom": {
    "status": "UNKNOWN",
    "custom": {
      "status": "UNKNOWN",
      "msg": "mock down to test aggregator"
    }
  },
  "diskSpace": {
    "status": "UP",
    "total": 249779191808,
    "free": 57925111808,
    "threshold": 10485760
  }
}

status枚举

    /**
     * Convenient constant value representing unknown state.
     */
    public static final Status UNKNOWN = new Status("UNKNOWN");

    /**
     * Convenient constant value representing up state.
     */
    public static final Status UP = new Status("UP");

    /**
     * Convenient constant value representing down state.
     */
    public static final Status DOWN = new Status("DOWN");

    /**
     * Convenient constant value representing out-of-service state.
     */
    public static final Status OUT_OF_SERVICE = new Status("OUT_OF_SERVICE");

/health的两个贴心点

对于多个HealthIndicator的status,spring boot默认对其进行aggregrate,然后计算最顶层的status字段的值,而且对于status是DOWN或者是OUT_OF_SERVICE的,返回的http的状态码是503,这对于应用监控系统来说真是大大的贴心啊,再总结一下:

  • 自动聚合多个HealthIndicator的status

  • 对于status是DOWN或者是OUT_OF_SERVICE的,返回503

这样应用监控系统一来就无需去解析返回结果,直接根据http的状态码就可以判断了,非常方便,太省心了有没有。

HealthAggregator

/**
 * Base {@link HealthAggregator} implementation to allow subclasses to focus on
 * aggregating the {@link Status} instances and not deal with contextual details etc.
 *
 * @author Christian Dupuis
 * @author Vedran Pavic
 * @since 1.1.0
 */
public abstract class AbstractHealthAggregator implements HealthAggregator {

    @Override
    public final Health aggregate(Map<String, Health> healths) {
        List<Status> statusCandidates = new ArrayList<Status>();
        for (Map.Entry<String, Health> entry : healths.entrySet()) {
            statusCandidates.add(entry.getValue().getStatus());
        }
        Status status = aggregateStatus(statusCandidates);
        Map<String, Object> details = aggregateDetails(healths);
        return new Health.Builder(status, details).build();
    }

    /**
     * Return the single 'aggregate' status that should be used from the specified
     * candidates.
     * @param candidates the candidates
     * @return a single status
     */
    protected abstract Status aggregateStatus(List<Status> candidates);

    /**
     * Return the map of 'aggregate' details that should be used from the specified
     * healths.
     * @param healths the health instances to aggregate
     * @return a map of details
     * @since 1.3.1
     */
    protected Map<String, Object> aggregateDetails(Map<String, Health> healths) {
        return new LinkedHashMap<String, Object>(healths);
    }

}

OrderedHealthAggregator#aggregateStatus

    @Override
    protected Status aggregateStatus(List<Status> candidates) {
        // Only sort those status instances that we know about
        List<Status> filteredCandidates = new ArrayList<Status>();
        for (Status candidate : candidates) {
            if (this.statusOrder.contains(candidate.getCode())) {
                filteredCandidates.add(candidate);
            }
        }
        // If no status is given return UNKNOWN
        if (filteredCandidates.isEmpty()) {
            return Status.UNKNOWN;
        }
        // Sort given Status instances by configured order
        Collections.sort(filteredCandidates, new StatusComparator(this.statusOrder));
        return filteredCandidates.get(0);
    }

可以看出是对status进行排序,然后取第一个的状态,其中statusOrder如下:

private List<String> statusOrder;

    /**
     * Create a new {@link OrderedHealthAggregator} instance.
     */
    public OrderedHealthAggregator() {
        setStatusOrder(Status.DOWN, Status.OUT_OF_SERVICE, Status.UP, Status.UNKNOWN);
    }

排序方法

/**
     * {@link Comparator} used to order {@link Status}.
     */
    private class StatusComparator implements Comparator<Status> {

        private final List<String> statusOrder;

        StatusComparator(List<String> statusOrder) {
            this.statusOrder = statusOrder;
        }

        @Override
        public int compare(Status s1, Status s2) {
            int i1 = this.statusOrder.indexOf(s1.getCode());
            int i2 = this.statusOrder.indexOf(s2.getCode());
            return (i1 < i2 ? -1 : (i1 == i2 ? s1.getCode().compareTo(s2.getCode()) : 1));
        }

    }

即Status.DOWN, Status.OUT_OF_SERVICE, Status.UP, Status.UNKNOWN优先级依次递减。status中一旦有出现DOWN的情况,整体的status就是DOWN,依次类推。

doc


codecraft
11.9k 声望2k 粉丝

当一个代码的工匠回首往事时,不因虚度年华而悔恨,也不因碌碌无为而羞愧,这样,当他老的时候,可以很自豪告诉世人,我曾经将代码注入生命去打造互联网的浪潮之巅,那是个很疯狂的时代,我在一波波的浪潮上留下...