//暂时没有实现基于enc的此方法
func (this *SafeMonitoredHosts) Init() {
var hostMap map[int]*model.Host
if g.Config().ExternalNodes == "" {
m, err := db.QueryMonitoredHosts()
if err != nil {
return
}
hostMap = m
} else {
m, err := enc.QueryMonitoredHosts()
if err != nil {
return
}
hostMap = m
}
this.Lock()
defer this.Unlock()
this.M = hostMap
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG][CACHE] SafeMonitoredHosts.init : %v", this.M)
}
}
func (this *SafeHostTemplateIds) Init() {
var hostTempMap map[int][]int
if g.Config().ExternalNodes == "" {
m, err := db.QueryHostTemplateIds()
if err != nil {
return
}
hostTempMap = m
} else {
m, err := enc.QueryHostTemplateIds()
if err != nil {
return
}
hostTempMap = m
}
this.Lock()
defer this.Unlock()
this.M = hostTempMap
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG][CACHE] SafeHostTemplateIds.init : %v", this.M)
}
}
func Init() {
var err error
DB, err = sql.Open("mysql", g.Config().Database)
if err != nil {
log.Fatalln("open db fail:", err)
}
DB.SetMaxIdleConns(g.Config().MaxIdle)
err = DB.Ping()
if err != nil {
log.Fatalln("ping db fail:", err)
}
}
// agent按照server端的配置，按需采集的metric，比如net.port.listen port=22 或者 proc.num name=zabbix_agentd
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()
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG] TrustableIps args is %v, reply is %v", args, reply)
}
return nil
}
func Start() {
addr := g.Config().Listen
server := rpc.NewServer()
// server.Register(new(filter.Filter))
server.Register(new(Agent))
server.Register(new(Hbs))
l, e := net.Listen("tcp", addr)
if e != nil {
log.Fatalln("listen error:", e)
} else {
log.Println("listening", addr)
}
for {
conn, err := l.Accept()
if err != nil {
log.Println("listener accept fail:", err)
time.Sleep(time.Duration(100) * time.Millisecond)
continue
}
go server.ServeCodec(jsonrpc.NewServerCodec(conn))
}
}
func Start() {
if !g.Config().Http.Enabled {
return
}
addr := g.Config().Http.Listen
if addr == "" {
return
}
s := &http.Server{
Addr: addr,
MaxHeaderBytes: 1 << 30,
}
log.Println("http listening", addr)
log.Fatalln(s.ListenAndServe())
}
func UpdateAgent(agentInfo *model.AgentUpdateInfo) {
sql := ""
if g.Config().Hosts == "" {
sql = fmt.Sprintf(
"insert into host(hostname, ip, agent_version, plugin_version) values ('%s', '%s', '%s', '%s') on duplicate key update ip='%s', agent_version='%s', plugin_version='%s'",
agentInfo.ReportRequest.Hostname,
agentInfo.ReportRequest.IP,
agentInfo.ReportRequest.AgentVersion,
agentInfo.ReportRequest.PluginVersion,
agentInfo.ReportRequest.IP,
agentInfo.ReportRequest.AgentVersion,
agentInfo.ReportRequest.PluginVersion,
)
} else {
// sync, just update
sql = fmt.Sprintf(
"update host set ip='%s', agent_version='%s', plugin_version='%s' where hostname='%s'",
agentInfo.ReportRequest.IP,
agentInfo.ReportRequest.AgentVersion,
agentInfo.ReportRequest.PluginVersion,
agentInfo.ReportRequest.Hostname,
)
}
_, err := DB.Exec(sql)
if err != nil {
log.Println("exec", sql, "fail", err)
}
}
func (t *Hbs) GetExpressions(req model.NullRpcRequest, reply *model.ExpressionResponse) error {
reply.Expressions = cache.ExpressionCache.Get()
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG] GetExpressions args is %v, reply is %v", req, reply)
}
return nil
}
func (this *SafeHostGroupsMap) GetGroupIds(hid int) ([]int, bool) {
this.RLock()
defer this.RUnlock()
gids, exists := this.M[hid]
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG][CACHE] group.GetGroupIds :hid is %v, gids is %v, exists is %v", hid, gids, exists)
}
return gids, exists
}
func (this *SafeAgents) Put(req *model.AgentReportRequest) {
val := &model.AgentUpdateInfo{
LastUpdate: time.Now().Unix(),
ReportRequest: req,
}
if g.Config().ExternalNodes == "" {
db.UpdateAgent(val)
}
this.Lock()
defer this.Unlock()
this.M[req.Hostname] = val
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG][CACHE] agent.put : %v", this.M)
}
}
func (this *SafeAgents) Delete(hostname string) {
this.Lock()
defer this.Unlock()
delete(this.M, hostname)
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG][CACHE] agent.delete : %v", this.M)
}
}
func (this *SafeAgents) Get(hostname string) (*model.AgentUpdateInfo, bool) {
this.RLock()
defer this.RUnlock()
val, exists := this.M[hostname]
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG][CACHE] agent.get : hostname:%v, val is %v, exists is %v", hostname, val, exists)
}
return val, exists
}
func (this *SafeHostMap) GetID(hostname string) (int, bool) {
this.RLock()
defer this.RUnlock()
id, exists := this.M[hostname]
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG][CACHE] host.getid : hostname is %v,id is %v, exists is %v", hostname, id, exists)
}
return id, exists
}
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()
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG] MinePlugins args is %v, reply is %v", args, reply)
}
return nil
}
func (t *Agent) ReportStatus(args *model.AgentReportRequest, reply *model.SimpleRpcResponse) error {
if args.Hostname == "" {
reply.Code = 1
return nil
}
cache.Agents.Put(args)
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG] ReportStatus args is %v, reply is %v", args, reply)
}
return nil
}
func (this *SafeAgents) Keys() []string {
this.RLock()
defer this.RUnlock()
count := len(this.M)
keys := make([]string, count)
i := 0
for hostname := range this.M {
keys[i] = hostname
i++
}
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG][CACHE] agent.keys : %v", keys)
}
return keys
}
func configCommonRoutes() {
http.HandleFunc("/health", func(w http.ResponseWriter, r *http.Request) {
w.Write([]byte("ok"))
})
http.HandleFunc("/version", func(w http.ResponseWriter, r *http.Request) {
w.Write([]byte(g.VERSION))
})
http.HandleFunc("/workdir", func(w http.ResponseWriter, r *http.Request) {
RenderDataJson(w, file.SelfDir())
})
http.HandleFunc("/config/reload", func(w http.ResponseWriter, r *http.Request) {
if strings.HasPrefix(r.RemoteAddr, "127.0.0.1") {
g.ParseConfig(g.ConfigFile)
RenderDataJson(w, g.Config())
} else {
w.Write([]byte("no privilege"))
}
})
}
func Init() {
encShell = g.Config().ExternalNodes
debug = g.Config().Debug
}
// 需要checksum一下来减少网络开销？其实白名单通常只会有一个或者没有，无需checksum
func (t *Agent) TrustableIps(args *model.NullRpcRequest, ips *string) error {
*ips = strings.Join(g.Config().Trustable, ",")
return nil
}
func (t *Hbs) GetStrategies(req model.NullRpcRequest, reply *model.StrategiesResponse) error {
reply.HostStrategies = []*model.HostStrategy{}
// 一个机器ID对应多个模板ID
hidTids := cache.HostTemplateIds.GetMap()
sz := len(hidTids)
if sz == 0 {
return nil
}
// Judge需要的是hostname，此处要把HostId转换为hostname
// 查出的hosts，是不处于维护时间内的
hosts := cache.MonitoredHosts.Get()
if len(hosts) == 0 {
// 所有机器都处于维护状态，汗
return nil
}
tpls := cache.TemplateCache.GetMap()
if len(tpls) == 0 {
return nil
}
strategies := cache.Strategies.GetMap()
if len(strategies) == 0 {
return nil
}
// 做个索引，给一个tplId，可以很方便的找到对应了哪些Strategy
tpl2Strategies := Tpl2Strategies(strategies)
hostStrategies := make([]*model.HostStrategy, 0, sz)
for hostId, tplIds := range hidTids {
h, exists := hosts[hostId]
if !exists {
continue
}
// 计算当前host配置了哪些监控策略
ss := CalcInheritStrategies(tpls, tplIds, tpl2Strategies)
if len(ss) <= 0 {
continue
}
hs := model.HostStrategy{
Hostname: h.Name,
Strategies: ss,
}
hostStrategies = append(hostStrategies, &hs)
}
reply.HostStrategies = hostStrategies
debug := g.Config().Debug
if debug {
log.Printf("[DEBUG] GetStrategies args is %v, reply is %v", req, reply)
}
return nil
}
