k3s/vendor/k8s.io/apiserver/pkg/endpoints/metrics/metrics.go

495 lines
16 KiB
Go
Raw Normal View History

2019-01-12 04:58:27 +00:00
/*
Copyright 2015 The Kubernetes Authors.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
package metrics
import (
"bufio"
"net"
"net/http"
"regexp"
"strconv"
"strings"
"sync"
"time"
"github.com/emicklei/go-restful"
"github.com/prometheus/client_golang/prometheus"
2019-04-07 17:07:55 +00:00
"k8s.io/apimachinery/pkg/apis/meta/v1/validation"
utilnet "k8s.io/apimachinery/pkg/util/net"
utilsets "k8s.io/apimachinery/pkg/util/sets"
"k8s.io/apiserver/pkg/endpoints/request"
2019-01-12 04:58:27 +00:00
)
// resettableCollector is the interface implemented by prometheus.MetricVec
// that can be used by Prometheus to collect metrics and reset their values.
type resettableCollector interface {
prometheus.Collector
Reset()
}
2019-04-07 17:07:55 +00:00
const (
APIServerComponent string = "apiserver"
)
2019-01-12 04:58:27 +00:00
var (
// TODO(a-robinson): Add unit tests for the handling of these metrics once
// the upstream library supports it.
requestCounter = prometheus.NewCounterVec(
2019-04-07 17:07:55 +00:00
prometheus.CounterOpts{
Name: "apiserver_request_total",
Help: "Counter of apiserver requests broken out for each verb, dry run value, group, version, resource, scope, component, client, and HTTP response contentType and code.",
},
// The label_name contentType doesn't follow the label_name convention defined here:
// https://github.com/kubernetes/community/blob/master/contributors/devel/sig-instrumentation/instrumentation.md
// But changing it would break backwards compatibility. Future label_names
// should be all lowercase and separated by underscores.
[]string{"verb", "dry_run", "group", "version", "resource", "subresource", "scope", "component", "client", "contentType", "code"},
)
deprecatedRequestCounter = prometheus.NewCounterVec(
2019-01-12 04:58:27 +00:00
prometheus.CounterOpts{
Name: "apiserver_request_count",
2019-04-07 17:07:55 +00:00
Help: "(Deprecated) Counter of apiserver requests broken out for each verb, group, version, resource, scope, component, client, and HTTP response contentType and code.",
2019-01-12 04:58:27 +00:00
},
2019-04-07 17:07:55 +00:00
[]string{"verb", "group", "version", "resource", "subresource", "scope", "component", "client", "contentType", "code"},
2019-01-12 04:58:27 +00:00
)
longRunningRequestGauge = prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Name: "apiserver_longrunning_gauge",
2019-04-07 17:07:55 +00:00
Help: "Gauge of all active long-running apiserver requests broken out by verb, group, version, resource, scope and component. Not all requests are tracked this way.",
2019-01-12 04:58:27 +00:00
},
2019-04-07 17:07:55 +00:00
[]string{"verb", "group", "version", "resource", "subresource", "scope", "component"},
2019-01-12 04:58:27 +00:00
)
requestLatencies = prometheus.NewHistogramVec(
2019-04-07 17:07:55 +00:00
prometheus.HistogramOpts{
Name: "apiserver_request_duration_seconds",
Help: "Response latency distribution in seconds for each verb, dry run value, group, version, resource, subresource, scope and component.",
// This metric is used for verifying api call latencies SLO,
// as well as tracking regressions in this aspects.
// Thus we customize buckets significantly, to empower both usecases.
Buckets: []float64{0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0,
1.25, 1.5, 1.75, 2.0, 2.5, 3.0, 3.5, 4.0, 4.5, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40, 50, 60},
},
[]string{"verb", "dry_run", "group", "version", "resource", "subresource", "scope", "component"},
)
deprecatedRequestLatencies = prometheus.NewHistogramVec(
2019-01-12 04:58:27 +00:00
prometheus.HistogramOpts{
Name: "apiserver_request_latencies",
2019-04-07 17:07:55 +00:00
Help: "(Deprecated) Response latency distribution in microseconds for each verb, group, version, resource, subresource, scope and component.",
2019-01-12 04:58:27 +00:00
// Use buckets ranging from 125 ms to 8 seconds.
Buckets: prometheus.ExponentialBuckets(125000, 2.0, 7),
},
2019-04-07 17:07:55 +00:00
[]string{"verb", "group", "version", "resource", "subresource", "scope", "component"},
2019-01-12 04:58:27 +00:00
)
2019-04-07 17:07:55 +00:00
deprecatedRequestLatenciesSummary = prometheus.NewSummaryVec(
2019-01-12 04:58:27 +00:00
prometheus.SummaryOpts{
Name: "apiserver_request_latencies_summary",
2019-04-07 17:07:55 +00:00
Help: "(Deprecated) Response latency summary in microseconds for each verb, group, version, resource, subresource, scope and component.",
2019-01-12 04:58:27 +00:00
// Make the sliding window of 5h.
// TODO: The value for this should be based on our SLI definition (medium term).
MaxAge: 5 * time.Hour,
},
2019-04-07 17:07:55 +00:00
[]string{"verb", "group", "version", "resource", "subresource", "scope", "component"},
2019-01-12 04:58:27 +00:00
)
responseSizes = prometheus.NewHistogramVec(
prometheus.HistogramOpts{
Name: "apiserver_response_sizes",
2019-04-07 17:07:55 +00:00
Help: "Response size distribution in bytes for each group, version, verb, resource, subresource, scope and component.",
2019-01-12 04:58:27 +00:00
// Use buckets ranging from 1000 bytes (1KB) to 10^9 bytes (1GB).
Buckets: prometheus.ExponentialBuckets(1000, 10.0, 7),
},
2019-04-07 17:07:55 +00:00
[]string{"verb", "group", "version", "resource", "subresource", "scope", "component"},
2019-01-12 04:58:27 +00:00
)
// DroppedRequests is a number of requests dropped with 'Try again later' response"
DroppedRequests = prometheus.NewCounterVec(
prometheus.CounterOpts{
2019-04-07 17:07:55 +00:00
Name: "apiserver_dropped_requests_total",
2019-01-12 04:58:27 +00:00
Help: "Number of requests dropped with 'Try again later' response",
},
[]string{"requestKind"},
)
2019-04-07 17:07:55 +00:00
DeprecatedDroppedRequests = prometheus.NewCounterVec(
prometheus.CounterOpts{
Name: "apiserver_dropped_requests",
Help: "(Deprecated) Number of requests dropped with 'Try again later' response",
},
[]string{"requestKind"},
)
2019-01-12 04:58:27 +00:00
// RegisteredWatchers is a number of currently registered watchers splitted by resource.
RegisteredWatchers = prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Name: "apiserver_registered_watchers",
Help: "Number of currently registered watchers for a given resources",
},
[]string{"group", "version", "kind"},
)
// Because of volatality of the base metric this is pre-aggregated one. Instead of reporing current usage all the time
// it reports maximal usage during the last second.
currentInflightRequests = prometheus.NewGaugeVec(
prometheus.GaugeOpts{
Name: "apiserver_current_inflight_requests",
2019-04-07 17:07:55 +00:00
Help: "Maximal number of currently used inflight request limit of this apiserver per request kind in last second.",
2019-01-12 04:58:27 +00:00
},
[]string{"requestKind"},
)
kubectlExeRegexp = regexp.MustCompile(`^.*((?i:kubectl\.exe))`)
metrics = []resettableCollector{
requestCounter,
2019-04-07 17:07:55 +00:00
deprecatedRequestCounter,
2019-01-12 04:58:27 +00:00
longRunningRequestGauge,
requestLatencies,
2019-04-07 17:07:55 +00:00
deprecatedRequestLatencies,
deprecatedRequestLatenciesSummary,
2019-01-12 04:58:27 +00:00
responseSizes,
DroppedRequests,
2019-04-07 17:07:55 +00:00
DeprecatedDroppedRequests,
2019-01-12 04:58:27 +00:00
RegisteredWatchers,
currentInflightRequests,
}
)
const (
// ReadOnlyKind is a string identifying read only request kind
ReadOnlyKind = "readOnly"
// MutatingKind is a string identifying mutating request kind
MutatingKind = "mutating"
)
var registerMetrics sync.Once
// Register all metrics.
func Register() {
registerMetrics.Do(func() {
for _, metric := range metrics {
prometheus.MustRegister(metric)
}
})
}
// Reset all metrics.
func Reset() {
for _, metric := range metrics {
metric.Reset()
}
}
func UpdateInflightRequestMetrics(nonmutating, mutating int) {
currentInflightRequests.WithLabelValues(ReadOnlyKind).Set(float64(nonmutating))
currentInflightRequests.WithLabelValues(MutatingKind).Set(float64(mutating))
}
// Record records a single request to the standard metrics endpoints. For use by handlers that perform their own
// processing. All API paths should use InstrumentRouteFunc implicitly. Use this instead of MonitorRequest if
// you already have a RequestInfo object.
2019-04-07 17:07:55 +00:00
func Record(req *http.Request, requestInfo *request.RequestInfo, component, contentType string, code int, responseSizeInBytes int, elapsed time.Duration) {
2019-01-12 04:58:27 +00:00
if requestInfo == nil {
requestInfo = &request.RequestInfo{Verb: req.Method, Path: req.URL.Path}
}
scope := CleanScope(requestInfo)
if requestInfo.IsResourceRequest {
2019-04-07 17:07:55 +00:00
MonitorRequest(req, strings.ToUpper(requestInfo.Verb), requestInfo.APIGroup, requestInfo.APIVersion, requestInfo.Resource, requestInfo.Subresource, scope, component, contentType, code, responseSizeInBytes, elapsed)
2019-01-12 04:58:27 +00:00
} else {
2019-04-07 17:07:55 +00:00
MonitorRequest(req, strings.ToUpper(requestInfo.Verb), "", "", "", requestInfo.Path, scope, component, contentType, code, responseSizeInBytes, elapsed)
2019-01-12 04:58:27 +00:00
}
}
// RecordLongRunning tracks the execution of a long running request against the API server. It provides an accurate count
// of the total number of open long running requests. requestInfo may be nil if the caller is not in the normal request flow.
2019-04-07 17:07:55 +00:00
func RecordLongRunning(req *http.Request, requestInfo *request.RequestInfo, component string, fn func()) {
2019-01-12 04:58:27 +00:00
if requestInfo == nil {
requestInfo = &request.RequestInfo{Verb: req.Method, Path: req.URL.Path}
}
var g prometheus.Gauge
scope := CleanScope(requestInfo)
reportedVerb := cleanVerb(strings.ToUpper(requestInfo.Verb), req)
if requestInfo.IsResourceRequest {
2019-04-07 17:07:55 +00:00
g = longRunningRequestGauge.WithLabelValues(reportedVerb, requestInfo.APIGroup, requestInfo.APIVersion, requestInfo.Resource, requestInfo.Subresource, scope, component)
2019-01-12 04:58:27 +00:00
} else {
2019-04-07 17:07:55 +00:00
g = longRunningRequestGauge.WithLabelValues(reportedVerb, "", "", "", requestInfo.Path, scope, component)
2019-01-12 04:58:27 +00:00
}
g.Inc()
defer g.Dec()
fn()
}
// MonitorRequest handles standard transformations for client and the reported verb and then invokes Monitor to record
// a request. verb must be uppercase to be backwards compatible with existing monitoring tooling.
2019-04-07 17:07:55 +00:00
func MonitorRequest(req *http.Request, verb, group, version, resource, subresource, scope, component, contentType string, httpCode, respSize int, elapsed time.Duration) {
2019-01-12 04:58:27 +00:00
reportedVerb := cleanVerb(verb, req)
2019-04-07 17:07:55 +00:00
dryRun := cleanDryRun(req.URL.Query()["dryRun"])
2019-01-12 04:58:27 +00:00
client := cleanUserAgent(utilnet.GetHTTPClient(req))
elapsedMicroseconds := float64(elapsed / time.Microsecond)
2019-04-07 17:07:55 +00:00
elapsedSeconds := elapsed.Seconds()
requestCounter.WithLabelValues(reportedVerb, dryRun, group, version, resource, subresource, scope, component, client, contentType, codeToString(httpCode)).Inc()
deprecatedRequestCounter.WithLabelValues(reportedVerb, group, version, resource, subresource, scope, component, client, contentType, codeToString(httpCode)).Inc()
requestLatencies.WithLabelValues(reportedVerb, dryRun, group, version, resource, subresource, scope, component).Observe(elapsedSeconds)
deprecatedRequestLatencies.WithLabelValues(reportedVerb, group, version, resource, subresource, scope, component).Observe(elapsedMicroseconds)
deprecatedRequestLatenciesSummary.WithLabelValues(reportedVerb, group, version, resource, subresource, scope, component).Observe(elapsedMicroseconds)
2019-01-12 04:58:27 +00:00
// We are only interested in response sizes of read requests.
if verb == "GET" || verb == "LIST" {
2019-04-07 17:07:55 +00:00
responseSizes.WithLabelValues(reportedVerb, group, version, resource, subresource, scope, component).Observe(float64(respSize))
2019-01-12 04:58:27 +00:00
}
}
// InstrumentRouteFunc works like Prometheus' InstrumentHandlerFunc but wraps
// the go-restful RouteFunction instead of a HandlerFunc plus some Kubernetes endpoint specific information.
2019-04-07 17:07:55 +00:00
func InstrumentRouteFunc(verb, group, version, resource, subresource, scope, component string, routeFunc restful.RouteFunction) restful.RouteFunction {
2019-01-12 04:58:27 +00:00
return restful.RouteFunction(func(request *restful.Request, response *restful.Response) {
now := time.Now()
delegate := &ResponseWriterDelegator{ResponseWriter: response.ResponseWriter}
_, cn := response.ResponseWriter.(http.CloseNotifier)
_, fl := response.ResponseWriter.(http.Flusher)
_, hj := response.ResponseWriter.(http.Hijacker)
var rw http.ResponseWriter
if cn && fl && hj {
rw = &fancyResponseWriterDelegator{delegate}
} else {
rw = delegate
}
response.ResponseWriter = rw
routeFunc(request, response)
2019-04-07 17:07:55 +00:00
MonitorRequest(request.Request, verb, group, version, resource, subresource, scope, component, delegate.Header().Get("Content-Type"), delegate.Status(), delegate.ContentLength(), time.Since(now))
2019-01-12 04:58:27 +00:00
})
}
// InstrumentHandlerFunc works like Prometheus' InstrumentHandlerFunc but adds some Kubernetes endpoint specific information.
2019-04-07 17:07:55 +00:00
func InstrumentHandlerFunc(verb, group, version, resource, subresource, scope, component string, handler http.HandlerFunc) http.HandlerFunc {
2019-01-12 04:58:27 +00:00
return func(w http.ResponseWriter, req *http.Request) {
now := time.Now()
delegate := &ResponseWriterDelegator{ResponseWriter: w}
_, cn := w.(http.CloseNotifier)
_, fl := w.(http.Flusher)
_, hj := w.(http.Hijacker)
if cn && fl && hj {
w = &fancyResponseWriterDelegator{delegate}
} else {
w = delegate
}
handler(w, req)
2019-04-07 17:07:55 +00:00
MonitorRequest(req, verb, group, version, resource, subresource, scope, component, delegate.Header().Get("Content-Type"), delegate.Status(), delegate.ContentLength(), time.Since(now))
2019-01-12 04:58:27 +00:00
}
}
// CleanScope returns the scope of the request.
func CleanScope(requestInfo *request.RequestInfo) string {
if requestInfo.Namespace != "" {
return "namespace"
}
if requestInfo.Name != "" {
return "resource"
}
if requestInfo.IsResourceRequest {
return "cluster"
}
// this is the empty scope
return ""
}
func cleanVerb(verb string, request *http.Request) string {
reportedVerb := verb
if verb == "LIST" {
// see apimachinery/pkg/runtime/conversion.go Convert_Slice_string_To_bool
if values := request.URL.Query()["watch"]; len(values) > 0 {
if value := strings.ToLower(values[0]); value != "0" && value != "false" {
reportedVerb = "WATCH"
}
}
}
// normalize the legacy WATCHLIST to WATCH to ensure users aren't surprised by metrics
if verb == "WATCHLIST" {
reportedVerb = "WATCH"
}
return reportedVerb
}
2019-04-07 17:07:55 +00:00
func cleanDryRun(dryRun []string) string {
if errs := validation.ValidateDryRun(nil, dryRun); len(errs) > 0 {
return "invalid"
}
// Since dryRun could be valid with any arbitrarily long length
// we have to dedup and sort the elements before joining them together
return strings.Join(utilsets.NewString(dryRun...).List(), ",")
}
2019-01-12 04:58:27 +00:00
func cleanUserAgent(ua string) string {
// We collapse all "web browser"-type user agents into one "browser" to reduce metric cardinality.
if strings.HasPrefix(ua, "Mozilla/") {
return "Browser"
}
// If an old "kubectl.exe" has passed us its full path, we discard the path portion.
ua = kubectlExeRegexp.ReplaceAllString(ua, "$1")
return ua
}
// ResponseWriterDelegator interface wraps http.ResponseWriter to additionally record content-length, status-code, etc.
type ResponseWriterDelegator struct {
http.ResponseWriter
status int
written int64
wroteHeader bool
}
func (r *ResponseWriterDelegator) WriteHeader(code int) {
r.status = code
r.wroteHeader = true
r.ResponseWriter.WriteHeader(code)
}
func (r *ResponseWriterDelegator) Write(b []byte) (int, error) {
if !r.wroteHeader {
r.WriteHeader(http.StatusOK)
}
n, err := r.ResponseWriter.Write(b)
r.written += int64(n)
return n, err
}
func (r *ResponseWriterDelegator) Status() int {
return r.status
}
func (r *ResponseWriterDelegator) ContentLength() int {
return int(r.written)
}
type fancyResponseWriterDelegator struct {
*ResponseWriterDelegator
}
func (f *fancyResponseWriterDelegator) CloseNotify() <-chan bool {
return f.ResponseWriter.(http.CloseNotifier).CloseNotify()
}
func (f *fancyResponseWriterDelegator) Flush() {
f.ResponseWriter.(http.Flusher).Flush()
}
func (f *fancyResponseWriterDelegator) Hijack() (net.Conn, *bufio.ReadWriter, error) {
return f.ResponseWriter.(http.Hijacker).Hijack()
}
// Small optimization over Itoa
func codeToString(s int) string {
switch s {
case 100:
return "100"
case 101:
return "101"
case 200:
return "200"
case 201:
return "201"
case 202:
return "202"
case 203:
return "203"
case 204:
return "204"
case 205:
return "205"
case 206:
return "206"
case 300:
return "300"
case 301:
return "301"
case 302:
return "302"
case 304:
return "304"
case 305:
return "305"
case 307:
return "307"
case 400:
return "400"
case 401:
return "401"
case 402:
return "402"
case 403:
return "403"
case 404:
return "404"
case 405:
return "405"
case 406:
return "406"
case 407:
return "407"
case 408:
return "408"
case 409:
return "409"
case 410:
return "410"
case 411:
return "411"
case 412:
return "412"
case 413:
return "413"
case 414:
return "414"
case 415:
return "415"
case 416:
return "416"
case 417:
return "417"
case 418:
return "418"
case 500:
return "500"
case 501:
return "501"
case 502:
return "502"
case 503:
return "503"
case 504:
return "504"
case 505:
return "505"
case 428:
return "428"
case 429:
return "429"
case 431:
return "431"
case 511:
return "511"
default:
return strconv.Itoa(s)
}
}