关于监控工具:Openfalcon-hbs源码解读

64次阅读

共计 3003 个字符,预计需要花费 8 分钟才能阅读完成。

hbs 负责周期性的读取 db 中内容,缓存到本地 cache,而后提供 RPC 接口以供 agent 和 judge 两个组件查问调用。

// modules/hbs/rpc/rpc.go
func Start() {server := rpc.NewServer()
    server.Register(new(Agent))
    server.Register(new(Hbs))
    l, e := net.Listen("tcp", addr)
    for {conn, err := l.Accept()
        ......
        go server.ServeCodec(jsonrpc.NewServerCodec(conn))
    }
}

hbs 对接 agent:

  • 解决 agent heartbeat 申请;
  • 解决 agent plugin 的查问申请;
  • 解决 agent 监控哪些过程、哪些端口的查问申请;

hbs 对接 judge:

  • 查问以后配置的所有告警策略;
  • 查问以后配置的所有告警表达式;

整体流程图:

1. hbs 对接 agent

agent 查问执行哪些 plugin:

// modules/hbs/rpc/agent.go
func (t *Agent) MinePlugins(args model.AgentHeartbeatRequest, reply *model.AgentPluginsResponse) error {
    if args.Hostname == "" {return nil}

    reply.Plugins = cache.GetPlugins(args.Hostname)
    reply.Timestamp = time.Now().Unix()

    return nil
}

agent 查问监控哪些过程、哪些端口:

// modules/hbs/rpc/agent.go
func (t *Agent) BuiltinMetrics(args *model.AgentHeartbeatRequest, reply *model.BuiltinMetricResponse) error {
    if args.Hostname == "" {return nil}
    metrics, err := cache.GetBuiltinMetrics(args.Hostname)
    if err != nil {return nil}
    checksum := ""
    if len(metrics) > 0 {checksum = DigestBuiltinMetrics(metrics)
    }
    if args.Checksum == checksum {reply.Metrics = []*model.BuiltinMetric{}} else {reply.Metrics = metrics}
    reply.Checksum = checksum
    reply.Timestamp = time.Now().Unix()
    return nil
}

能够看到,下面的 rpc 接口操作的都是 cache,hbs 会定期的从 db 中查问数据,而后缓存在本地 cache,以供 agent 查问:

// modules/hbs/cache/cache.go
func LoopInit() {
    for {
        //1min 周期
        time.Sleep(time.Minute)
        GroupPlugins.Init()
        ....
    }
}

从 db 读取数据,而后保留在本地:为了避免并发写 data,这里加了锁

// moduels/hbs/cache/plugins.go
var GroupPlugins = &SafeGroupPlugins{M: make(map[int][]string)}

func (this *SafeGroupPlugins) Init() {m, err := db.QueryPlugins()
    if err != nil {return}

    this.Lock()
    defer this.Unlock()
    this.M = m
}

2. hbs 对接 judge

judge 查问告警表达式:

// modules/hbs/rpc/hbs.go
func (t *Hbs) GetExpressions(req model.NullRpcRequest, reply *model.ExpressionResponse) error {reply.Expressions = cache.ExpressionCache.Get()
    return nil
}

judge 查问告警策略:因为关联多个 db table,这里进行了比较复杂的拼装

// modules/hbs/rpc/hbs.go
func (t *Hbs) GetStrategies(req model.NullRpcRequest, reply *model.StrategiesResponse) error {reply.HostStrategies = []*model.HostStrategy{}
    hidTids := cache.HostTemplateIds.GetMap()
    hosts := cache.MonitoredHosts.Get()
    tpls := cache.TemplateCache.GetMap()
    strategies := cache.Strategies.GetMap()
    tpl2Strategies := Tpl2Strategies(strategies)
    hostStrategies := make([]*model.HostStrategy, 0, sz)
    for hostId, tplIds := range hidTids {ss := CalcInheritStrategies(tpls, tplIds, tpl2Strategies)
        hs := model.HostStrategy{
            Hostname:   h.Name,
            Strategies: ss,
        }
        hostStrategies = append(hostStrategies, &hs)
    }
    reply.HostStrategies = hostStrategies
    return nil
}

同样的,下面的 rpc 接口操作的都是 cache,hbs 会定期的从 db 中查问数据,而后缓存在本地 cache,以供 judge 查问:

// modules/hbs/cache/cache.go
func LoopInit() {
    for {time.Sleep(time.Minute)
        ......
        GroupTemplates.Init()
        HostGroupsMap.Init()
        HostMap.Init()
        TemplateCache.Init()
        Strategies.Init(TemplateCache.GetMap())
        HostTemplateIds.Init()
        ExpressionCache.Init()}
}

cache 的数据保留在 map 中,而后读取 db 中的数据 (定期),笼罩掉 map

// modules/hbs/cache/templates.go
type SafeTemplateCache struct {
    sync.RWMutex
    M map[int]*model.Template
}

var TemplateCache = &SafeTemplateCache{M: make(map[int]*model.Template)}

func (this *SafeTemplateCache) Init() {ts, err := db.QueryTemplates()
    if err != nil {return}
    this.Lock()
    defer this.Unlock()
    this.M = ts
}

正文完
 0