func MakeIncidentSummary(c conf.RuleConfProvider, s SilenceTester, is *models.IncidentState) (*IncidentSummaryView, error) {
alert := c.GetAlert(is.AlertKey.Name())
if alert == nil {
return nil, fmt.Errorf("alert %v does not exist in the configuration", is.AlertKey.Name())
}
warnNotifications := alert.WarnNotification.Get(c, is.AlertKey.Group())
critNotifications := alert.CritNotification.Get(c, is.AlertKey.Group())
eventSummaries := []EventSummary{}
nonNormalNonUnknownCount := 0
for _, event := range is.Events {
if event.Status > models.StNormal && event.Status < models.StUnknown {
nonNormalNonUnknownCount++
}
if eventSummary, unevaluated := MakeEventSummary(event); !unevaluated {
eventSummaries = append(eventSummaries, eventSummary)
}
}
actions := make([]EpochAction, len(is.Actions))
for i, action := range is.Actions {
actions[i] = MakeEpochAction(action)
}
subject := is.Subject
// There is no rendered subject when the state is unknown and
// there is no other non-normal status in the history.
if subject == "" && nonNormalNonUnknownCount == 0 {
subject = fmt.Sprintf("%s: %v", is.CurrentStatus, is.AlertKey)
}
return &IncidentSummaryView{
Id: is.Id,
Subject: subject,
Start: is.Start.Unix(),
AlertName: is.AlertKey.Name(),
Tags: is.AlertKey.Group(),
TagsString: is.AlertKey.Group().String(),
CurrentStatus: is.CurrentStatus,
WorstStatus: is.WorstStatus,
LastAbnormalStatus: is.LastAbnormalStatus,
LastAbnormalTime: is.LastAbnormalTime,
Unevaluated: is.Unevaluated,
NeedAck: is.NeedAck,
Silenced: s(is.AlertKey) != nil,
Actions: actions,
Events: eventSummaries,
WarnNotificationChains: conf.GetNotificationChains(c, warnNotifications),
CritNotificationChains: conf.GetNotificationChains(c, critNotifications),
}, nil
}
func main() {
flag.Parse()
if *flagVersion {
fmt.Println(version.GetVersionInfo("bosun"))
os.Exit(0)
}
for _, m := range mains {
m()
}
systemConf, err := conf.LoadSystemConfigFile(*flagConf)
if err != nil {
slog.Fatalf("couldn't read system configuration: %v", err)
}
sysProvider, err := systemConf.GetSystemConfProvider()
if err != nil {
slog.Fatal(err)
}
ruleConf, err := rule.ParseFile(sysProvider.GetRuleFilePath(), systemConf.EnabledBackends())
if err != nil {
slog.Fatalf("couldn't read rules: %v", err)
}
if *flagTest {
os.Exit(0)
}
var ruleProvider conf.RuleConfProvider = ruleConf
httpListen := &url.URL{
Scheme: "http",
Host: sysProvider.GetHTTPListen(),
}
if strings.HasPrefix(httpListen.Host, ":") {
httpListen.Host = "localhost" + httpListen.Host
}
if err := metadata.Init(httpListen, false); err != nil {
slog.Fatal(err)
}
if err := sched.Load(sysProvider, ruleProvider, *flagSkipLast, *flagQuiet); err != nil {
slog.Fatal(err)
}
if sysProvider.GetRelayListen() != "" {
go func() {
mux := http.NewServeMux()
mux.Handle("/api/", util.NewSingleHostProxy(httpListen))
s := &http.Server{
Addr: sysProvider.GetRelayListen(),
Handler: mux,
}
slog.Fatal(s.ListenAndServe())
}()
}
if sysProvider.GetTSDBHost() != "" {
if err := collect.Init(httpListen, "bosun"); err != nil {
slog.Fatal(err)
}
tsdbHost := &url.URL{
Scheme: "http",
Host: sysProvider.GetTSDBHost(),
}
if *flagReadonly {
rp := util.NewSingleHostProxy(tsdbHost)
ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.URL.Path == "/api/put" {
w.WriteHeader(204)
return
}
rp.ServeHTTP(w, r)
}))
slog.Infoln("readonly relay at", ts.URL, "to", tsdbHost)
tsdbHost, _ = url.Parse(ts.URL)
sysProvider.SetTSDBHost(tsdbHost.Host)
}
}
if systemConf.GetPing() {
go ping.PingHosts(sched.DefaultSched.Search, systemConf.GetPingDuration())
}
if sysProvider.GetInternetProxy() != "" {
web.InternetProxy, err = url.Parse(sysProvider.GetInternetProxy())
if err != nil {
slog.Fatalf("InternetProxy error: %s", err)
}
}
var cmdHook conf.SaveHook
if hookPath := sysProvider.GetCommandHookPath(); hookPath != "" {
cmdHook, err = conf.MakeSaveCommandHook(hookPath)
if err != nil {
slog.Fatal(err)
}
ruleProvider.SetSaveHook(cmdHook)
}
var reload func() error
reloading := make(chan bool, 1) // a lock that we can give up acquiring
reload = func() error {
select {
case reloading <- true:
// Got lock
default:
return fmt.Errorf("not reloading, reload in progress")
}
defer func() {
<-reloading
}()
newConf, err := rule.ParseFile(sysProvider.GetRuleFilePath(), sysProvider.EnabledBackends())
if err != nil {
return err
}
newConf.SetSaveHook(cmdHook)
newConf.SetReload(reload)
oldSched := sched.DefaultSched
oldDA := oldSched.DataAccess
oldSearch := oldSched.Search
sched.Close(true)
sched.Reset()
newSched := sched.DefaultSched
newSched.Search = oldSearch
newSched.DataAccess = oldDA
slog.Infoln("schedule shutdown, loading new schedule")
// Load does not set the DataAccess or Search if it is already set
if err := sched.Load(sysProvider, newConf, *flagSkipLast, *flagQuiet); err != nil {
slog.Fatal(err)
}
web.ResetSchedule() // Signal web to point to the new DefaultSchedule
go func() {
slog.Infoln("running new schedule")
if !*flagNoChecks {
sched.Run()
}
}()
slog.Infoln("config reload complete")
return nil
}
ruleProvider.SetReload(reload)
go func() {
slog.Fatal(web.Listen(sysProvider.GetHTTPListen(), *flagDev, sysProvider.GetTSDBHost(), reload))
}()
go func() {
if !*flagNoChecks {
sched.Run()
}
}()
go func() {
sc := make(chan os.Signal, 1)
signal.Notify(sc, os.Interrupt, syscall.SIGTERM)
killing := false
for range sc {
if killing {
slog.Infoln("Second interrupt: exiting")
os.Exit(1)
}
killing = true
go func() {
slog.Infoln("Interrupt: closing down...")
sched.Close(false)
slog.Infoln("done")
os.Exit(1)
}()
}
}()
if *flagWatch {
watch(".", "*.go", quit)
watch(filepath.Join("web", "static", "templates"), "*.html", web.RunEsc)
base := filepath.Join("web", "static", "js")
watch(base, "*.ts", web.RunTsc)
}
select {}
}