mirror
/
client_golang
mirror of https://github.com/prometheus/client_golang.git
1
0
Fork 1
Code Issues Packages Projects Releases Wiki Activity

Take into account curried labels in promhttp

This commit is contained in:
beorn7 2017-12-23 01:06:17 +01:00
parent 1ba60c7d58
commit 9ac0bad606
3 changed files with 243 additions and 81 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
prometheus/promhttp/instrument_client.go
View File

@ -45,12 +45,11 @@ func InstrumentRoundTripperInFlight(gauge prometheus.Gauge, next http.RoundTripp
// InstrumentRoundTripperCounter is a middleware that wraps the provided
// http.RoundTripper to observe the request result with the provided CounterVec.
// The CounterVec must have zero, one, or two labels. The only allowed label
// names are "code" and "method". The function panics if any other instance
// labels are provided. Partitioning of the CounterVec happens by HTTP status
// code and/or HTTP method if the respective instance label names are present
// in the CounterVec. For unpartitioned counting, use a CounterVec with
// zero labels.
// The CounterVec must have zero, one, or two non-const non-curried labels. For
// those, the only allowed label names are "code" and "method". The function
// panics otherwise. Partitioning of the CounterVec happens by HTTP status code
// and/or HTTP method if the respective instance label names are present in the
// CounterVec. For unpartitioned counting, use a CounterVec with zero labels.
//
// If the wrapped RoundTripper panics or returns a non-nil error, the Counter
// is not incremented.

157
prometheus/promhttp/instrument_server.go
View File

@ -14,6 +14,7 @@
package promhttp
import (
"errors"
"net/http"
"strconv"
"strings"
@ -42,10 +43,10 @@ func InstrumentHandlerInFlight(g prometheus.Gauge, next http.Handler) http.Handl
// InstrumentHandlerDuration is a middleware that wraps the provided
// http.Handler to observe the request duration with the provided ObserverVec.
// The ObserverVec must have zero, one, or two labels. The only allowed label
// names are "code" and "method". The function panics if any other instance
// labels are provided. The Observe method of the Observer in the ObserverVec
// is called with the request duration in seconds. Partitioning happens by HTTP
// The ObserverVec must have zero, one, or two non-const non-curried labels. For
// those, the only allowed label names are "code" and "method". The function
// panics otherwise. The Observe method of the Observer in the ObserverVec is
// called with the request duration in seconds. Partitioning happens by HTTP
// status code and/or HTTP method if the respective instance label names are
// present in the ObserverVec. For unpartitioned observations, use an
// ObserverVec with zero labels. Note that partitioning of Histograms is
@ -77,14 +78,13 @@ func InstrumentHandlerDuration(obs prometheus.ObserverVec, next http.Handler) ht
})
}
// InstrumentHandlerCounter is a middleware that wraps the provided
// http.Handler to observe the request result with the provided CounterVec.
// The CounterVec must have zero, one, or two labels. The only allowed label
// names are "code" and "method". The function panics if any other instance
// labels are provided. Partitioning of the CounterVec happens by HTTP status
// code and/or HTTP method if the respective instance label names are present
// in the CounterVec. For unpartitioned counting, use a CounterVec with
// zero labels.
// InstrumentHandlerCounter is a middleware that wraps the provided http.Handler
// to observe the request result with the provided CounterVec. The CounterVec
// must have zero, one, or two non-const non-curried labels. For those, the only
// allowed label names are "code" and "method". The function panics
// otherwise. Partitioning of the CounterVec happens by HTTP status code and/or
// HTTP method if the respective instance label names are present in the
// CounterVec. For unpartitioned counting, use a CounterVec with zero labels.
//
// If the wrapped Handler does not set a status code, a status code of 200 is assumed.
//
@ -111,14 +111,13 @@ func InstrumentHandlerCounter(counter *prometheus.CounterVec, next http.Handler)
// InstrumentHandlerTimeToWriteHeader is a middleware that wraps the provided
// http.Handler to observe with the provided ObserverVec the request duration
// until the response headers are written. The ObserverVec must have zero, one,
// or two labels. The only allowed label names are "code" and "method". The
// function panics if any other instance labels are provided. The Observe
// method of the Observer in the ObserverVec is called with the request
// duration in seconds. Partitioning happens by HTTP status code and/or HTTP
// method if the respective instance label names are present in the
// ObserverVec. For unpartitioned observations, use an ObserverVec with zero
// labels. Note that partitioning of Histograms is expensive and should be used
// judiciously.
// or two non-const non-curried labels. For those, the only allowed label names
// are "code" and "method". The function panics otherwise. The Observe method of
// the Observer in the ObserverVec is called with the request duration in
// seconds. Partitioning happens by HTTP status code and/or HTTP method if the
// respective instance label names are present in the ObserverVec. For
// unpartitioned observations, use an ObserverVec with zero labels. Note that
// partitioning of Histograms is expensive and should be used judiciously.
//
// If the wrapped Handler panics before calling WriteHeader, no value is
// reported.
@ -140,15 +139,15 @@ func InstrumentHandlerTimeToWriteHeader(obs prometheus.ObserverVec, next http.Ha
}
// InstrumentHandlerRequestSize is a middleware that wraps the provided
// http.Handler to observe the request size with the provided ObserverVec.
// The ObserverVec must have zero, one, or two labels. The only allowed label
// names are "code" and "method". The function panics if any other instance
// labels are provided. The Observe method of the Observer in the ObserverVec
// is called with the request size in bytes. Partitioning happens by HTTP
// status code and/or HTTP method if the respective instance label names are
// present in the ObserverVec. For unpartitioned observations, use an
// ObserverVec with zero labels. Note that partitioning of Histograms is
// expensive and should be used judiciously.
// http.Handler to observe the request size with the provided ObserverVec. The
// ObserverVec must have zero, one, or two non-const non-curried labels. For
// those, the only allowed label names are "code" and "method". The function
// panics otherwise. The Observe method of the Observer in the ObserverVec is
// called with the request size in bytes. Partitioning happens by HTTP status
// code and/or HTTP method if the respective instance label names are present in
// the ObserverVec. For unpartitioned observations, use an ObserverVec with zero
// labels. Note that partitioning of Histograms is expensive and should be used
// judiciously.
//
// If the wrapped Handler does not set a status code, a status code of 200 is assumed.
//
@ -175,15 +174,15 @@ func InstrumentHandlerRequestSize(obs prometheus.ObserverVec, next http.Handler)
}
// InstrumentHandlerResponseSize is a middleware that wraps the provided
// http.Handler to observe the response size with the provided ObserverVec.
// The ObserverVec must have zero, one, or two labels. The only allowed label
// names are "code" and "method". The function panics if any other instance
// labels are provided. The Observe method of the Observer in the ObserverVec
// is called with the response size in bytes. Partitioning happens by HTTP
// status code and/or HTTP method if the respective instance label names are
// present in the ObserverVec. For unpartitioned observations, use an
// ObserverVec with zero labels. Note that partitioning of Histograms is
// expensive and should be used judiciously.
// http.Handler to observe the response size with the provided ObserverVec. The
// ObserverVec must have zero, one, or two non-const non-curried labels. For
// those, the only allowed label names are "code" and "method". The function
// panics otherwise. The Observe method of the Observer in the ObserverVec is
// called with the response size in bytes. Partitioning happens by HTTP status
// code and/or HTTP method if the respective instance label names are present in
// the ObserverVec. For unpartitioned observations, use an ObserverVec with zero
// labels. Note that partitioning of Histograms is expensive and should be used
// judiciously.
//
// If the wrapped Handler does not set a status code, a status code of 200 is assumed.
//
@ -204,9 +203,12 @@ func checkLabels(c prometheus.Collector) (code bool, method bool) {
// once Descriptors can have their dimensionality queried.
var (
desc *prometheus.Desc
m prometheus.Metric
pm dto.Metric
lvs []string
)
// Get the Desc from the Collector.
descc := make(chan *prometheus.Desc, 1)
c.Describe(descc)
@ -223,49 +225,54 @@ func checkLabels(c prometheus.Collector) (code bool, method bool) {
close(descc)
if _, err := prometheus.NewConstMetric(desc, prometheus.UntypedValue, 0); err == nil {
return
// Create a ConstMetric with the Desc. Since we don't know how many
// variable labels there are, try for as long as it needs.
for err := errors.New("dummy"); err != nil; lvs = append(lvs, magicString) {
m, err = prometheus.NewConstMetric(desc, prometheus.UntypedValue, 0, lvs...)
}
if m, err := prometheus.NewConstMetric(desc, prometheus.UntypedValue, 0, magicString); err == nil {
if err := m.Write(&pm); err != nil {
panic("error checking metric for labels")
}
for _, label := range pm.Label {
name, value := label.GetName(), label.GetValue()
if value != magicString {
continue
}
switch name {
case "code":
code = true
case "method":
method = true
default:
panic("metric partitioned with non-supported labels")
}
return
}
panic("previously set label not found – this must never happen")
// Write out the metric into a proto message and look at the labels.
// If the value is not the magicString, it is a constLabel, which doesn't interest us.
// If the label is curried, it doesn't interest us.
// In all other cases, only "code" or "method" is allowed.
if err := m.Write(&pm); err != nil {
panic("error checking metric for labels")
}
if m, err := prometheus.NewConstMetric(desc, prometheus.UntypedValue, 0, magicString, magicString); err == nil {
if err := m.Write(&pm); err != nil {
panic("error checking metric for labels")
for _, label := range pm.Label {
name, value := label.GetName(), label.GetValue()
if value != magicString || isLabelCurried(c, name) {
continue
}
for _, label := range pm.Label {
name, value := label.GetName(), label.GetValue()
if value != magicString {
continue
}
if name == "code" || name == "method" {
continue
}
switch name {
case "code":
code = true
case "method":
method = true
default:
panic("metric partitioned with non-supported labels")
}
code = true
method = true
return
}
panic("metric partitioned with non-supported labels")
return
}
func isLabelCurried(c prometheus.Collector, label string) bool {
// This is even hackier than the label test above.
// We essentially try to curry again and see if it works.
// But for that, we need to type-convert to the two
// types we use here, ObserverVec or *CounterVec.
switch v := c.(type) {
case *prometheus.CounterVec:
if _, err := v.CurryWith(prometheus.Labels{label: "dummy"}); err == nil {
return false
}
case prometheus.ObserverVec:
if _, err := v.CurryWith(prometheus.Labels{label: "dummy"}); err == nil {
return false
}
default:
panic("unsupported metric vec type")
}
return true
}
// emptyLabels is a one-time allocation for non-partitioned metrics to avoid

156
prometheus/promhttp/instrument_server_test.go
View File

@ -23,6 +23,162 @@ import (
"github.com/prometheus/client_golang/prometheus"
)
func TestLabelCheck(t *testing.T) {
scenarios := map[string]struct {
varLabels []string
constLabels []string
curriedLabels []string
ok bool
}{
"empty": {
varLabels: []string{},
constLabels: []string{},
curriedLabels: []string{},
ok: true,
},
"code as single var label": {
varLabels: []string{"code"},
constLabels: []string{},
curriedLabels: []string{},
ok: true,
},
"method as single var label": {
varLabels: []string{"method"},
constLabels: []string{},
curriedLabels: []string{},
ok: true,
},
"cade and method as var labels": {
varLabels: []string{"method", "code"},
constLabels: []string{},
curriedLabels: []string{},
ok: true,
},
"valid case with all labels used": {
varLabels: []string{"code", "method"},
constLabels: []string{"foo", "bar"},
curriedLabels: []string{"dings", "bums"},
ok: true,
},
"unsupported var label": {
varLabels: []string{"foo"},
constLabels: []string{},
curriedLabels: []string{},
ok: false,
},
"mixed var labels": {
varLabels: []string{"method", "foo", "code"},
constLabels: []string{},
curriedLabels: []string{},
ok: false,
},
"unsupported var label but curried": {
varLabels: []string{},
constLabels: []string{},
curriedLabels: []string{"foo"},
ok: true,
},
"mixed var labels but unsupported curried": {
varLabels: []string{"code", "method"},
constLabels: []string{},
curriedLabels: []string{"foo"},
ok: true,
},
"supported label as const and curry": {
varLabels: []string{},
constLabels: []string{"code"},
curriedLabels: []string{"method"},
ok: true,
},
"supported label as const and curry with unsupported as var": {
varLabels: []string{"foo"},
constLabels: []string{"code"},
curriedLabels: []string{"method"},
ok: false,
},
}
for name, sc := range scenarios {
t.Run(name, func(t *testing.T) {
constLabels := prometheus.Labels{}
for _, l := range sc.constLabels {
constLabels[l] = "dummy"
}
c := prometheus.NewCounterVec(
prometheus.CounterOpts{
Name: "c",
Help: "c help",
ConstLabels: constLabels,
},
append(sc.varLabels, sc.curriedLabels...),
)
o := prometheus.ObserverVec(prometheus.NewHistogramVec(
prometheus.HistogramOpts{
Name: "c",
Help: "c help",
ConstLabels: constLabels,
},
append(sc.varLabels, sc.curriedLabels...),
))
for _, l := range sc.curriedLabels {
c = c.MustCurryWith(prometheus.Labels{l: "dummy"})
o = o.MustCurryWith(prometheus.Labels{l: "dummy"})
}
func() {
defer func() {
if err := recover(); err != nil {
if sc.ok {
t.Error("unexpected panic:", err)
}
} else if !sc.ok {
t.Error("expected panic")
}
}()
InstrumentHandlerCounter(c, nil)
}()
func() {
defer func() {
if err := recover(); err != nil {
if sc.ok {
t.Error("unexpected panic:", err)
}
} else if !sc.ok {
t.Error("expected panic")
}
}()
InstrumentHandlerDuration(o, nil)
}()
if sc.ok {
// Test if wantCode and wantMethod were detected correctly.
var wantCode, wantMethod bool
for _, l := range sc.varLabels {
if l == "code" {
wantCode = true
}
if l == "method" {
wantMethod = true
}
}
gotCode, gotMethod := checkLabels(c)
if gotCode != wantCode {
t.Errorf("wanted code=%t for counter, got code=%t", wantCode, gotCode)
}
if gotMethod != wantMethod {
t.Errorf("wanted method=%t for counter, got method=%t", wantMethod, gotMethod)
}
gotCode, gotMethod = checkLabels(o)
if gotCode != wantCode {
t.Errorf("wanted code=%t for observer, got code=%t", wantCode, gotCode)
}
if gotMethod != wantMethod {
t.Errorf("wanted method=%t for observer, got method=%t", wantMethod, gotMethod)
}
}
})
}
}
func TestMiddlewareAPI(t *testing.T) {
reg := prometheus.NewRegistry()