// Query the SHOW variables from the query map
// TODO: make this more functional
func queryShowVariables(ch chan<- prometheus.Metric, db *sql.DB, variableMap map[string]MetricMapNamespace) []error {
log.Debugln("Querying SHOW variables")
nonFatalErrors := []error{}
for _, mapping := range variableMap {
for columnName, columnMapping := range mapping.columnMappings {
// Check for a discard request on this value
if columnMapping.discard {
continue
}
// Use SHOW to get the value
row := db.QueryRow(fmt.Sprintf("SHOW %s;", columnName))
var val interface{}
err := row.Scan(&val)
if err != nil {
nonFatalErrors = append(nonFatalErrors, errors.New(fmt.Sprintln("Error scanning runtime variable:", columnName, err)))
continue
}
fval, ok := columnMapping.conversion(val)
if !ok {
nonFatalErrors = append(nonFatalErrors, errors.New(fmt.Sprintln("Unexpected error parsing column: ", namespace, columnName, val)))
continue
}
ch <- prometheus.MustNewConstMetric(columnMapping.desc, columnMapping.vtype, fval)
}
}
return nonFatalErrors
}
// ScrapeClientStat collects from `information_schema.client_statistics`.
func ScrapeClientStat(db *sql.DB, ch chan<- prometheus.Metric) error {
var varName, varVal string
err := db.QueryRow(userstatCheckQuery).Scan(&varName, &varVal)
if err != nil {
log.Debugln("Detailed client stats are not available.")
return nil
}
if varVal == "OFF" {
log.Debugf("MySQL @@%s is OFF.", varName)
return nil
}
informationSchemaClientStatisticsRows, err := db.Query(clientStatQuery)
if err != nil {
return err
}
defer informationSchemaClientStatisticsRows.Close()
// The client column is assumed to be column[0], while all other data is assumed to be coerceable to float64.
// Because of the client column, clientStatData[0] maps to columnNames[1] when reading off the metrics
// (because clientStatScanArgs is mapped as [ &client, &clientData[0], &clientData[1] ... &clientdata[n] ]
// To map metrics to names therefore we always range over columnNames[1:]
columnNames, err := informationSchemaClientStatisticsRows.Columns()
if err != nil {
return err
}
var (
client string // Holds the client name, which should be in column 0.
clientStatData = make([]float64, len(columnNames)-1) // 1 less because of the client column.
clientStatScanArgs = make([]interface{}, len(columnNames))
)
clientStatScanArgs[0] = &client
for i := range clientStatData {
clientStatScanArgs[i+1] = &clientStatData[i]
}
for informationSchemaClientStatisticsRows.Next() {
if err := informationSchemaClientStatisticsRows.Scan(clientStatScanArgs...); err != nil {
return err
}
// Loop over column names, and match to scan data. Unknown columns
// will be filled with an untyped metric number. We assume other then
// cient, that we'll only get numbers.
for idx, columnName := range columnNames[1:] {
if metricType, ok := informationSchemaClientStatisticsTypes[columnName]; ok {
ch <- prometheus.MustNewConstMetric(metricType.desc, metricType.vtype, float64(clientStatData[idx]), client)
} else {
// Unknown metric. Report as untyped.
desc := prometheus.NewDesc(prometheus.BuildFQName(namespace, informationSchema, fmt.Sprintf("client_statistics_%s", strings.ToLower(columnName))), fmt.Sprintf("Unsupported metric from column %s", columnName), []string{"client"}, nil)
ch <- prometheus.MustNewConstMetric(desc, prometheus.UntypedValue, float64(clientStatData[idx]), client)
}
}
}
return nil
}
// Collect implements prometheus.Collector.
func (e *Exporter) Collect(ch chan<- prometheus.Metric) {
log.Debugln("Collecting metrics")
e.scrape(ch)
ch <- e.duration
ch <- e.totalScrapes
ch <- e.totalErrors
ch <- e.scrapeError
ch <- e.up
}
// ScrapeTableStat collects from `information_schema.table_statistics`.
func ScrapeTableStat(db *sql.DB, ch chan<- prometheus.Metric) error {
var varName, varVal string
err := db.QueryRow(userstatCheckQuery).Scan(&varName, &varVal)
if err != nil {
log.Debugln("Detailed table stats are not available.")
return nil
}
if varVal == "OFF" {
log.Debugf("MySQL @@%s is OFF.", varName)
return nil
}
informationSchemaTableStatisticsRows, err := db.Query(tableStatQuery)
if err != nil {
return err
}
defer informationSchemaTableStatisticsRows.Close()
var (
tableSchema string
tableName string
rowsRead uint64
rowsChanged uint64
rowsChangedXIndexes uint64
)
for informationSchemaTableStatisticsRows.Next() {
err = informationSchemaTableStatisticsRows.Scan(
&tableSchema,
&tableName,
&rowsRead,
&rowsChanged,
&rowsChangedXIndexes,
)
if err != nil {
return err
}
ch <- prometheus.MustNewConstMetric(
infoSchemaTableStatsRowsReadDesc, prometheus.CounterValue, float64(rowsRead),
tableSchema, tableName,
)
ch <- prometheus.MustNewConstMetric(
infoSchemaTableStatsRowsChangedDesc, prometheus.CounterValue, float64(rowsChanged),
tableSchema, tableName,
)
ch <- prometheus.MustNewConstMetric(
infoSchemaTableStatsRowsChangedXIndexesDesc, prometheus.CounterValue, float64(rowsChangedXIndexes),
tableSchema, tableName,
)
}
return nil
}
// ScrapeQueryResponseTime collects from `information_schema.query_response_time`.
func ScrapeQueryResponseTime(db *sql.DB, ch chan<- prometheus.Metric) error {
var queryStats uint8
err := db.QueryRow(queryResponseCheckQuery).Scan(&queryStats)
if err != nil {
log.Debugln("Query response time distribution is not present.")
return nil
}
if queryStats == 0 {
log.Debugln("query_response_time_stats is OFF.")
return nil
}
for i, query := range queryResponseTimeQueries {
err := processQueryResponseTimeTable(db, ch, query, i)
// The first query should not fail if query_response_time_stats is ON,
// unlike the other two when the read/write tables exist only with Percona Server 5.6/5.7.
if i == 0 && err != nil {
return err
}
}
return nil
}
// Implements the real polling functionality, but returns the connection object
// so other classes can inherit it.
func (s *BasicService) doPoll() net.Conn {
log.Debugln("Dialing basic service", s.Host().Hostname, s.Port(), s.Name())
conn, err := s.dialAndScrape()
if err != nil {
log.Infoln("Error", s.Host().Hostname, s.Port(), s.Name(), err)
s.portOpen = FAILED
} else {
log.Infoln("Success", s.Host().Hostname, s.Port(), s.Name())
s.portOpen = SUCCESS
}
return conn
}
// Stop all background processing.
func (tm *TargetManager) Stop() {
log.Infoln("Stopping target manager...")
tm.mtx.Lock()
// Cancel the base context, this will cause all target providers to shut down
// and all in-flight scrapes to abort immmediately.
// Started inserts will be finished before terminating.
tm.cancel()
tm.mtx.Unlock()
// Wait for all scrape inserts to complete.
tm.wg.Wait()
log.Debugln("Target manager stopped")
}
// Describe implements prometheus.Collector.
func (e *Exporter) Describe(ch chan<- *prometheus.Desc) {
log.Debugln("Describing metrics")
metricCh := make(chan prometheus.Metric)
doneCh := make(chan struct{})
go func() {
for m := range metricCh {
ch <- m.Desc()
}
close(doneCh)
}()
e.Collect(metricCh)
close(metricCh)
<-doneCh
}
// Send a list of notifications to the configured alert manager.
func (n *NotificationHandler) sendNotifications(reqs NotificationReqs) error {
n.mtx.RLock()
defer n.mtx.RUnlock()
alerts := make([]map[string]interface{}, 0, len(reqs))
for _, req := range reqs {
for ln, lv := range n.externalLabels {
if _, ok := req.Labels[ln]; !ok {
req.Labels[ln] = lv
}
}
alerts = append(alerts, map[string]interface{}{
"summary": req.Summary,
"description": req.Description,
"runbook": req.Runbook,
"labels": req.Labels,
"payload": map[string]interface{}{
"value": req.Value,
"activeSince": req.ActiveSince,
"generatorURL": req.GeneratorURL,
"alertingRule": req.RuleString,
},
})
}
buf, err := json.Marshal(alerts)
if err != nil {
return err
}
log.Debugln("Sending notifications to alertmanager:", string(buf))
resp, err := n.httpClient.Post(
n.alertmanagerURL+alertmanagerAPIEventsPath,
contentTypeJSON,
bytes.NewBuffer(buf),
)
if err != nil {
return err
}
defer resp.Body.Close()
_, err = ioutil.ReadAll(resp.Body)
if err != nil {
return err
}
// BUG: Do we need to check the response code?
return nil
}
// Iterate through all the namespace mappings in the exporter and run their
// queries.
func queryNamespaceMappings(ch chan<- prometheus.Metric, db *sql.DB, metricMap map[string]MetricMapNamespace, queryOverrides map[string]string) map[string]error {
// Return a map of namespace -> errors
namespaceErrors := make(map[string]error)
for namespace, mapping := range metricMap {
log.Debugln("Querying namespace: ", namespace)
nonFatalErrors, err := queryNamespaceMapping(ch, db, namespace, mapping, queryOverrides)
// Serious error - a namespace disappeard
if err != nil {
namespaceErrors[namespace] = err
log.Infoln(err)
}
// Non-serious errors - likely version or parsing problems.
if len(nonFatalErrors) > 0 {
for _, err := range nonFatalErrors {
log.Infoln(err.Error())
}
}
}
return namespaceErrors
}
func parseMycnf(config interface{}) (string, error) {
var dsn string
cfg, err := ini.Load(config)
if err != nil {
return dsn, fmt.Errorf("failed reading ini file: %s", err)
}
user := cfg.Section("client").Key("user").String()
password := cfg.Section("client").Key("password").String()
if (user == "") || (password == "") {
return dsn, fmt.Errorf("no user or password specified under [client] in %s", config)
}
host := cfg.Section("client").Key("host").MustString("localhost")
port := cfg.Section("client").Key("port").MustUint(3306)
socket := cfg.Section("client").Key("socket").String()
if socket != "" {
dsn = fmt.Sprintf("%s:%s@unix(%s)/", user, password, socket)
} else {
dsn = fmt.Sprintf("%s:%s@tcp(%s:%d)/", user, password, host, port)
}
log.Debugln(dsn)
return dsn, nil
}
// Check and update the exporters query maps if the version has changed.
func (e *Exporter) checkMapVersions(ch chan<- prometheus.Metric, db *sql.DB) error {
log.Debugln("Querying Postgres Version")
versionRow := db.QueryRow("SELECT version();")
var versionString string
err := versionRow.Scan(&versionString)
if err != nil {
return errors.New(fmt.Sprintln("Error scanning version string:", err))
}
semanticVersion, err := parseVersion(versionString)
// Check if semantic version changed and recalculate maps if needed.
if semanticVersion.NE(e.lastMapVersion) || e.variableMap == nil || e.metricMap == nil {
log.Infoln("Semantic Version Changed:", e.lastMapVersion.String(), "->", semanticVersion.String())
e.mappingMtx.Lock()
e.variableMap = makeDescMap(semanticVersion, variableMaps)
e.metricMap = makeDescMap(semanticVersion, metricMaps)
e.queryOverrides = makeQueryOverrideMap(semanticVersion, queryOverrides)
e.lastMapVersion = semanticVersion
if e.userQueriesPath != "" {
if err := addQueries(e.userQueriesPath, semanticVersion, e.metricMap, e.queryOverrides); err != nil {
log.Errorln("Failed to reload user queries:", e.userQueriesPath, err)
}
}
e.mappingMtx.Unlock()
}
// Output the version as a special metric
versionDesc := prometheus.NewDesc(fmt.Sprintf("%s_%s", namespace, staticLabelName),
"Version string as reported by postgres", []string{"version", "short_version"}, nil)
ch <- prometheus.MustNewConstMetric(versionDesc,
prometheus.UntypedValue, 1, versionString, semanticVersion.String())
return nil
}
func marathonConnect(uri *url.URL) error {
config := marathon.NewDefaultConfig()
config.URL = uri.String()
if uri.User != nil {
if passwd, ok := uri.User.Password(); ok {
config.HTTPBasicPassword = passwd
config.HTTPBasicAuthUser = uri.User.Username()
}
}
config.HTTPClient = &http.Client{
Timeout: 10 * time.Second,
Transport: &http.Transport{
Dial: (&net.Dialer{
Timeout: 10 * time.Second,
}).Dial,
TLSClientConfig: &tls.Config{
InsecureSkipVerify: true,
},
},
}
log.Debugln("Connecting to Marathon")
client, err := marathon.NewClient(config)
if err != nil {
return err
}
info, err := client.Info()
if err != nil {
return err
}
log.Debugf("Connected to Marathon! Name=%s, Version=%s\n", info.Name, info.Version)
return nil
}
// Query within a namespace mapping and emit metrics. Returns fatal errors if
// the scrape fails, and a slice of errors if they were non-fatal.
func queryNamespaceMapping(ch chan<- prometheus.Metric, db *sql.DB, namespace string, mapping MetricMapNamespace, queryOverrides map[string]string) ([]error, error) {
// Check for a query override for this namespace
query, found := queryOverrides[namespace]
if !found {
// No query override - do a simple * search.
query = fmt.Sprintf("SELECT * FROM %s;", namespace)
}
// Was this query disabled (i.e. nothing sensible can be queried on this
// version of PostgreSQL?
if query == "" {
// Return success (no pertinent data)
return []error{}, nil
}
// Don't fail on a bad scrape of one metric
rows, err := db.Query(query)
if err != nil {
return []error{}, errors.New(fmt.Sprintln("Error running query on database: ", namespace, err))
}
defer rows.Close()
var columnNames []string
columnNames, err = rows.Columns()
if err != nil {
return []error{}, errors.New(fmt.Sprintln("Error retrieving column list for: ", namespace, err))
}
// Make a lookup map for the column indices
var columnIdx = make(map[string]int, len(columnNames))
for i, n := range columnNames {
columnIdx[n] = i
}
var columnData = make([]interface{}, len(columnNames))
var scanArgs = make([]interface{}, len(columnNames))
for i := range columnData {
scanArgs[i] = &columnData[i]
}
nonfatalErrors := []error{}
for rows.Next() {
err = rows.Scan(scanArgs...)
if err != nil {
return []error{}, errors.New(fmt.Sprintln("Error retrieving rows:", namespace, err))
}
// Get the label values for this row
var labels = make([]string, len(mapping.labels))
for idx, columnName := range mapping.labels {
labels[idx], _ = dbToString(columnData[columnIdx[columnName]])
}
// Loop over column names, and match to scan data. Unknown columns
// will be filled with an untyped metric number *if* they can be
// converted to float64s. NULLs are allowed and treated as NaN.
for idx, columnName := range columnNames {
if metricMapping, ok := mapping.columnMappings[columnName]; ok {
// Is this a metricy metric?
if metricMapping.discard {
continue
}
value, ok := dbToFloat64(columnData[idx])
if !ok {
nonfatalErrors = append(nonfatalErrors, errors.New(fmt.Sprintln("Unexpected error parsing column: ", namespace, columnName, columnData[idx])))
continue
}
// Generate the metric
ch <- prometheus.MustNewConstMetric(metricMapping.desc, metricMapping.vtype, value, labels...)
} else {
// Unknown metric. Report as untyped if scan to float64 works, else note an error too.
desc := prometheus.NewDesc(fmt.Sprintf("%s_%s", namespace, columnName), fmt.Sprintf("Unknown metric from %s", namespace), nil, nil)
// Its not an error to fail here, since the values are
// unexpected anyway.
value, ok := dbToFloat64(columnData[idx])
if !ok {
nonfatalErrors = append(nonfatalErrors, errors.New(fmt.Sprintln("Unparseable column type - discarding: ", namespace, columnName, err)))
continue
}
log.Debugln(columnName, labels)
ch <- prometheus.MustNewConstMetric(desc, prometheus.UntypedValue, value, labels...)
}
}
}
return nonfatalErrors, nil
}
// Turn the MetricMap column mapping into a prometheus descriptor mapping.
func makeDescMap(pgVersion semver.Version, metricMaps map[string]map[string]ColumnMapping) map[string]MetricMapNamespace {
var metricMap = make(map[string]MetricMapNamespace)
for namespace, mappings := range metricMaps {
thisMap := make(map[string]MetricMap)
// Get the constant labels
var constLabels []string
for columnName, columnMapping := range mappings {
if columnMapping.usage == LABEL {
constLabels = append(constLabels, columnName)
}
}
for columnName, columnMapping := range mappings {
// Check column version compatibility for the current map
// Force to discard if not compatible.
if columnMapping.supportedVersions != nil {
if !columnMapping.supportedVersions(pgVersion) {
// It's very useful to be able to see what columns are being
// rejected.
log.Debugln(columnName, "is being forced to discard due to version incompatibility.")
thisMap[columnName] = MetricMap{
discard: true,
conversion: func(in interface{}) (float64, bool) {
return math.NaN(), true
},
}
continue
}
}
// Determine how to convert the column based on its usage.
switch columnMapping.usage {
case DISCARD, LABEL:
thisMap[columnName] = MetricMap{
discard: true,
conversion: func(in interface{}) (float64, bool) {
return math.NaN(), true
},
}
case COUNTER:
thisMap[columnName] = MetricMap{
vtype: prometheus.CounterValue,
desc: prometheus.NewDesc(fmt.Sprintf("%s_%s", namespace, columnName), columnMapping.description, constLabels, nil),
conversion: func(in interface{}) (float64, bool) {
return dbToFloat64(in)
},
}
case GAUGE:
thisMap[columnName] = MetricMap{
vtype: prometheus.GaugeValue,
desc: prometheus.NewDesc(fmt.Sprintf("%s_%s", namespace, columnName), columnMapping.description, constLabels, nil),
conversion: func(in interface{}) (float64, bool) {
return dbToFloat64(in)
},
}
case MAPPEDMETRIC:
thisMap[columnName] = MetricMap{
vtype: prometheus.GaugeValue,
desc: prometheus.NewDesc(fmt.Sprintf("%s_%s", namespace, columnName), columnMapping.description, constLabels, nil),
conversion: func(in interface{}) (float64, bool) {
text, ok := in.(string)
if !ok {
return math.NaN(), false
}
val, ok := columnMapping.mapping[text]
if !ok {
return math.NaN(), false
}
return val, true
},
}
case DURATION:
thisMap[columnName] = MetricMap{
vtype: prometheus.GaugeValue,
desc: prometheus.NewDesc(fmt.Sprintf("%s_%s_milliseconds", namespace, columnName), columnMapping.description, constLabels, nil),
conversion: func(in interface{}) (float64, bool) {
var durationString string
switch t := in.(type) {
case []byte:
durationString = string(t)
case string:
durationString = t
default:
log.Errorln("DURATION conversion metric was not a string")
return math.NaN(), false
}
if durationString == "-1" {
return math.NaN(), false
}
d, err := time.ParseDuration(durationString)
if err != nil {
log.Errorln("Failed converting result to metric:", columnName, in, err)
return math.NaN(), false
}
return float64(d / time.Millisecond), true
},
}
}
}
metricMap[namespace] = MetricMapNamespace{constLabels, thisMap}
}
return metricMap
}
// Add queries to the metricMaps and queryOverrides maps. Added queries do not
// respect version requirements, because it is assumed that the user knows
// what they are doing with their version of postgres.
//
// This function modifies metricMap and queryOverrideMap to contain the new
// queries.
// TODO: test code for all this.
// TODO: use proper struct type system
// TODO: the YAML this supports is "non-standard" - we should move away from it.
func addQueries(queriesPath string, pgVersion semver.Version, exporterMap map[string]MetricMapNamespace, queryOverrideMap map[string]string) error {
var extra map[string]interface{}
content, err := ioutil.ReadFile(queriesPath)
if err != nil {
return err
}
err = yaml.Unmarshal(content, &extra)
if err != nil {
return err
}
// Stores the loaded map representation
metricMaps := make(map[string]map[string]ColumnMapping)
newQueryOverrides := make(map[string]string)
for metric, specs := range extra {
log.Debugln("New user metric namespace from YAML:", metric)
for key, value := range specs.(map[interface{}]interface{}) {
switch key.(string) {
case "query":
query := value.(string)
newQueryOverrides[metric] = query
case "metrics":
for _, c := range value.([]interface{}) {
column := c.(map[interface{}]interface{})
for n, a := range column {
var columnMapping ColumnMapping
// Fetch the metric map we want to work on.
metricMap, ok := metricMaps[metric]
if !ok {
// Namespace for metric not found - add it.
metricMap = make(map[string]ColumnMapping)
metricMaps[metric] = metricMap
}
// Get name.
name := n.(string)
for attrKey, attrVal := range a.(map[interface{}]interface{}) {
switch attrKey.(string) {
case "usage":
usage, err := stringToColumnUsage(attrVal.(string))
if err != nil {
return err
}
columnMapping.usage = usage
case "description":
columnMapping.description = attrVal.(string)
}
}
// TODO: we should support this
columnMapping.mapping = nil
// Should we support this for users?
columnMapping.supportedVersions = nil
metricMap[name] = columnMapping
}
}
}
}
}
// Convert the loaded metric map into exporter representation
partialExporterMap := makeDescMap(pgVersion, metricMaps)
// Merge the two maps (which are now quite flatteend)
for k, v := range partialExporterMap {
_, found := exporterMap[k]
if found {
log.Debugln("Overriding metric", k, "from user YAML file.")
} else {
log.Debugln("Adding new metric", k, "from user YAML file.")
}
exporterMap[k] = v
}
// Merge the query override map
for k, v := range newQueryOverrides {
_, found := queryOverrideMap[k]
if found {
log.Debugln("Overriding query override", k, "from user YAML file.")
} else {
log.Debugln("Adding new query override", k, "from user YAML file.")
}
queryOverrideMap[k] = v
}
return nil
}
// ScrapeQueryResponseTime collects from `information_schema.query_response_time`.
func ScrapeQueryResponseTime(db *sql.DB, ch chan<- prometheus.Metric) error {
var queryStats uint8
err := db.QueryRow(queryResponseCheckQuery).Scan(&queryStats)
if err != nil {
log.Debugln("Query response time distribution is not present.")
return nil
}
if queryStats == 0 {
log.Debugln("query_response_time_stats is OFF.")
return nil
}
queryDistributionRows, err := db.Query(queryResponseTimeQuery)
if err != nil {
return err
}
defer queryDistributionRows.Close()
var (
length string
count uint64
total string
histogramCnt uint64
histogramSum float64
countBuckets = map[float64]uint64{}
)
for queryDistributionRows.Next() {
err = queryDistributionRows.Scan(
&length,
&count,
&total,
)
if err != nil {
return err
}
length, _ := strconv.ParseFloat(strings.TrimSpace(length), 64)
total, _ := strconv.ParseFloat(strings.TrimSpace(total), 64)
histogramCnt += count
histogramSum += total
// Special case for "TOO LONG" row where we take into account the count field which is the only available
// and do not add it as a part of histogram or metric
if length == 0 {
continue
}
countBuckets[length] = histogramCnt
// No histogram with query total times because they are float
ch <- prometheus.MustNewConstMetric(
infoSchemaQueryResponseTimeTotalDesc, prometheus.CounterValue, histogramSum,
fmt.Sprintf("%v", length),
)
}
ch <- prometheus.MustNewConstMetric(
infoSchemaQueryResponseTimeTotalDesc, prometheus.CounterValue, histogramSum,
"+Inf",
)
// Create histogram with query counts
ch <- prometheus.MustNewConstHistogram(
infoSchemaQueryResponseTimeCountDesc, histogramCnt, histogramSum, countBuckets,
)
return nil
}
func main() {
rand.Seed(time.Now().Unix())
flag.Parse()
// This is only used when we're running in -dev mode with bindata
rootDir, _ = osext.ExecutableFolder()
rootDir = path.Join(rootDir, "web")
// Parse configuration
cfg, err := config.LoadFromFile(*configFile)
if err != nil {
log.Fatalln("Error loading config", err)
}
// Templates
amberTmpl, err := Asset("templates/index.amber")
if err != nil {
log.Fatalln("Could not load index template:", err)
}
tmpl := amber.MustCompile(string(amberTmpl), amber.Options{})
// Setup the web UI
router := httprouter.New()
router.Handler("GET", *metricsPath, prometheus.Handler()) // Prometheus
// Static asset handling
router.GET("/static/*filepath", func(w http.ResponseWriter, r *http.Request, ps httprouter.Params) {
reqpath := ps.ByName("filepath")
realpath := path.Join("static", reqpath)
b, err := Asset(realpath)
if err != nil {
log.Debugln("Could not find asset: ", err)
return
} else {
w.Write(b)
}
})
var monitoredHosts []*pollers.Host
router.GET("/", func(w http.ResponseWriter, r *http.Request, _ httprouter.Params) {
data := struct {
Cfg *config.Config
Hosts *[]*pollers.Host
}{
Cfg: cfg,
Hosts: &monitoredHosts,
}
err := tmpl.Execute(w, &data)
if err != nil {
log.Errorln("Error rendering template", err)
}
})
// Initialize the host pollers
monitoredHosts = make([]*pollers.Host, len(cfg.Hosts))
// We don't allow duplicate hosts, but also don't want to panic just due
// to a typo, so keep track and skip duplicates here.
seenHosts := make(map[string]bool)
realidx := 0
for _, hostCfg := range cfg.Hosts {
log.Debugln("Setting up poller for: ", hostCfg.Hostname)
if *skipPing {
hostCfg.PingDisable = true
}
if _, ok := seenHosts[hostCfg.Hostname]; ok {
log.Warnln("Discarding repeat configuration of same hostname", hostCfg.Hostname)
continue
}
host := pollers.NewHost(hostCfg)
monitoredHosts[realidx] = host
prometheus.MustRegister(host)
seenHosts[hostCfg.Hostname] = true
realidx++
}
// Trim monitoredHosts to the number we actually used
monitoredHosts = monitoredHosts[0:realidx]
// This is the dispatcher. It is responsible for invoking the doPoll method
// of hosts.
connectionLimiter := pollers.NewLimiter(*maxConnections)
hostQueue := make(chan *pollers.Host)
// Start the host dispatcher
go func() {
for host := range hostQueue {
go host.Poll(connectionLimiter, hostQueue)
}
}()
// Do the initial host dispatch
go func() {
for _, host := range monitoredHosts {
log.Debugln("Starting polling for hosts")
hostQueue <- host
}
}()
var handler http.Handler
// If basic auth is requested, enable it for the interface.
if cfg.BasicAuthUsername != "" && cfg.BasicAuthPassword != "" {
basicauth := httpauth.SimpleBasicAuth(cfg.BasicAuthUsername,
cfg.BasicAuthPassword)
handler = basicauth(router)
} else {
handler = router
}
// If TLS certificates are specificed, use TLS
if cfg.TLSCertificatePath != "" && cfg.TLSKeyPath != "" {
log.Infof("Listening on (TLS-enabled) %s", *listenAddress)
err = http.ListenAndServeTLS(*listenAddress,
cfg.TLSCertificatePath, cfg.TLSKeyPath, handler)
} else {
log.Infof("Listening on %s", *listenAddress)
err = http.ListenAndServe(*listenAddress, handler)
}
if err != nil {
log.Fatal(err)
}
}