prometheus/retrieval/scrape_test.go

633 lines
16 KiB
Go
Raw Normal View History

2016-02-23 01:58:16 -08:00
// Copyright 2016 The Prometheus 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 retrieval
import (
"bytes"
2016-02-28 00:51:02 -08:00
"fmt"
"io"
"io/ioutil"
"math"
2016-02-28 14:59:03 -08:00
"net/http"
"net/http/httptest"
"net/url"
2016-02-28 00:51:02 -08:00
"reflect"
2016-02-28 14:59:03 -08:00
"strings"
2016-02-28 00:51:02 -08:00
"sync"
2016-02-23 01:58:16 -08:00
"testing"
"time"
"github.com/prometheus/common/model"
"github.com/stretchr/testify/require"
2016-02-23 01:58:16 -08:00
"golang.org/x/net/context"
"github.com/prometheus/prometheus/config"
"github.com/prometheus/prometheus/pkg/labels"
"github.com/prometheus/prometheus/pkg/timestamp"
2016-02-28 00:51:02 -08:00
"github.com/prometheus/prometheus/storage"
2016-02-23 01:58:16 -08:00
)
2016-02-28 00:51:02 -08:00
func TestNewScrapePool(t *testing.T) {
var (
app = &nopAppendable{}
2016-02-28 00:51:02 -08:00
cfg = &config.ScrapeConfig{}
sp = newScrapePool(context.Background(), cfg, app)
2016-02-28 00:51:02 -08:00
)
if a, ok := sp.appendable.(*nopAppendable); !ok || a != app {
2016-02-28 00:51:02 -08:00
t.Fatalf("Wrong sample appender")
}
if sp.config != cfg {
t.Fatalf("Wrong scrape config")
}
if sp.newLoop == nil {
t.Fatalf("newLoop function not initialized")
}
}
type testLoop struct {
startFunc func(interval, timeout time.Duration, errc chan<- error)
stopFunc func()
}
func (l *testLoop) run(interval, timeout time.Duration, errc chan<- error) {
l.startFunc(interval, timeout, errc)
}
func (l *testLoop) stop() {
l.stopFunc()
}
func TestScrapePoolStop(t *testing.T) {
sp := &scrapePool{
targets: map[uint64]*Target{},
loops: map[uint64]loop{},
2016-02-28 00:51:02 -08:00
}
var mtx sync.Mutex
stopped := map[uint64]bool{}
2016-02-28 00:51:02 -08:00
numTargets := 20
// Stopping the scrape pool must call stop() on all scrape loops,
// clean them and the respective targets up. It must wait until each loop's
// stop function returned before returning itself.
for i := 0; i < numTargets; i++ {
t := &Target{
labels: labels.FromStrings(model.AddressLabel, fmt.Sprintf("example.com:%d", i)),
2016-02-28 00:51:02 -08:00
}
l := &testLoop{}
l.stopFunc = func() {
time.Sleep(time.Duration(i*20) * time.Millisecond)
mtx.Lock()
stopped[t.hash()] = true
2016-02-28 00:51:02 -08:00
mtx.Unlock()
}
sp.targets[t.hash()] = t
sp.loops[t.hash()] = l
2016-02-28 00:51:02 -08:00
}
done := make(chan struct{})
stopTime := time.Now()
go func() {
sp.stop()
close(done)
}()
select {
case <-time.After(5 * time.Second):
t.Fatalf("scrapeLoop.stop() did not return as expected")
case <-done:
// This should have taken at least as long as the last target slept.
if time.Since(stopTime) < time.Duration(numTargets*20)*time.Millisecond {
t.Fatalf("scrapeLoop.stop() exited before all targets stopped")
}
}
mtx.Lock()
if len(stopped) != numTargets {
t.Fatalf("Expected 20 stopped loops, got %d", len(stopped))
}
mtx.Unlock()
if len(sp.targets) > 0 {
t.Fatalf("Targets were not cleared on stopping: %d left", len(sp.targets))
}
if len(sp.loops) > 0 {
t.Fatalf("Loops were not cleared on stopping: %d left", len(sp.loops))
}
}
func TestScrapePoolReload(t *testing.T) {
var mtx sync.Mutex
numTargets := 20
stopped := map[uint64]bool{}
2016-02-28 00:51:02 -08:00
reloadCfg := &config.ScrapeConfig{
ScrapeInterval: model.Duration(3 * time.Second),
ScrapeTimeout: model.Duration(2 * time.Second),
}
2016-09-14 20:23:28 -07:00
// On starting to run, new loops created on reload check whether their preceding
2016-02-28 00:51:02 -08:00
// equivalents have been stopped.
2017-01-13 05:48:01 -08:00
newLoop := func(ctx context.Context, s scraper, app, reportApp func() storage.Appender) loop {
2016-02-28 00:51:02 -08:00
l := &testLoop{}
l.startFunc = func(interval, timeout time.Duration, errc chan<- error) {
if interval != 3*time.Second {
t.Errorf("Expected scrape interval %d but got %d", 3*time.Second, interval)
}
if timeout != 2*time.Second {
t.Errorf("Expected scrape timeout %d but got %d", 2*time.Second, timeout)
}
mtx.Lock()
if !stopped[s.(*targetScraper).hash()] {
t.Errorf("Scrape loop for %v not stopped yet", s.(*targetScraper))
2016-02-28 00:51:02 -08:00
}
mtx.Unlock()
}
return l
}
sp := &scrapePool{
appendable: &nopAppendable{},
targets: map[uint64]*Target{},
loops: map[uint64]loop{},
newLoop: newLoop,
2016-02-28 00:51:02 -08:00
}
// Reloading a scrape pool with a new scrape configuration must stop all scrape
2016-09-14 20:23:28 -07:00
// loops and start new ones. A new loop must not be started before the preceding
2016-02-28 00:51:02 -08:00
// one terminated.
for i := 0; i < numTargets; i++ {
t := &Target{
labels: labels.FromStrings(model.AddressLabel, fmt.Sprintf("example.com:%d", i)),
2016-02-28 00:51:02 -08:00
}
l := &testLoop{}
l.stopFunc = func() {
time.Sleep(time.Duration(i*20) * time.Millisecond)
mtx.Lock()
stopped[t.hash()] = true
2016-02-28 00:51:02 -08:00
mtx.Unlock()
}
sp.targets[t.hash()] = t
sp.loops[t.hash()] = l
2016-02-28 00:51:02 -08:00
}
done := make(chan struct{})
beforeTargets := map[uint64]*Target{}
for h, t := range sp.targets {
beforeTargets[h] = t
2016-02-28 00:51:02 -08:00
}
reloadTime := time.Now()
go func() {
sp.reload(reloadCfg)
close(done)
}()
select {
case <-time.After(5 * time.Second):
t.Fatalf("scrapeLoop.reload() did not return as expected")
case <-done:
// This should have taken at least as long as the last target slept.
if time.Since(reloadTime) < time.Duration(numTargets*20)*time.Millisecond {
t.Fatalf("scrapeLoop.stop() exited before all targets stopped")
}
}
mtx.Lock()
if len(stopped) != numTargets {
2016-05-01 14:37:45 -07:00
t.Fatalf("Expected 20 stopped loops, got %d", len(stopped))
2016-02-28 00:51:02 -08:00
}
mtx.Unlock()
if !reflect.DeepEqual(sp.targets, beforeTargets) {
t.Fatalf("Reloading affected target states unexpectedly")
}
if len(sp.loops) != numTargets {
t.Fatalf("Expected %d loops after reload but got %d", numTargets, len(sp.loops))
}
}
func TestScrapePoolReportAppender(t *testing.T) {
cfg := &config.ScrapeConfig{
MetricRelabelConfigs: []*config.RelabelConfig{
{}, {}, {},
},
}
target := newTestTarget("example.com:80", 10*time.Millisecond, nil)
app := &nopAppendable{}
sp := newScrapePool(context.Background(), cfg, app)
cfg.HonorLabels = false
wrapped := sp.reportAppender(target)
rl, ok := wrapped.(ruleLabelsAppender)
if !ok {
t.Fatalf("Expected ruleLabelsAppender but got %T", wrapped)
}
if _, ok := rl.Appender.(nopAppender); !ok {
t.Fatalf("Expected base appender but got %T", rl.Appender)
}
cfg.HonorLabels = true
wrapped = sp.reportAppender(target)
hl, ok := wrapped.(ruleLabelsAppender)
if !ok {
t.Fatalf("Expected ruleLabelsAppender but got %T", wrapped)
}
if _, ok := rl.Appender.(nopAppender); !ok {
t.Fatalf("Expected base appender but got %T", hl.Appender)
}
}
func TestScrapePoolSampleAppender(t *testing.T) {
cfg := &config.ScrapeConfig{
MetricRelabelConfigs: []*config.RelabelConfig{
{}, {}, {},
},
}
target := newTestTarget("example.com:80", 10*time.Millisecond, nil)
app := &nopAppendable{}
sp := newScrapePool(context.Background(), cfg, app)
cfg.HonorLabels = false
wrapped := sp.sampleAppender(target)
rl, ok := wrapped.(ruleLabelsAppender)
if !ok {
t.Fatalf("Expected ruleLabelsAppender but got %T", wrapped)
}
re, ok := rl.Appender.(relabelAppender)
if !ok {
t.Fatalf("Expected relabelAppender but got %T", rl.Appender)
}
if _, ok := re.Appender.(nopAppender); !ok {
t.Fatalf("Expected base appender but got %T", re.Appender)
}
cfg.HonorLabels = true
cfg.SampleLimit = 100
wrapped = sp.sampleAppender(target)
hl, ok := wrapped.(honorLabelsAppender)
if !ok {
t.Fatalf("Expected honorLabelsAppender but got %T", wrapped)
}
re, ok = hl.Appender.(relabelAppender)
if !ok {
t.Fatalf("Expected relabelAppender but got %T", hl.Appender)
}
lm, ok := re.Appender.(*limitAppender)
if !ok {
t.Fatalf("Expected limitAppender but got %T", lm.Appender)
}
if _, ok := lm.Appender.(nopAppender); !ok {
t.Fatalf("Expected base appender but got %T", re.Appender)
}
}
2016-02-28 00:51:02 -08:00
func TestScrapeLoopStop(t *testing.T) {
scraper := &testScraper{}
sl := newScrapeLoop(context.Background(), scraper, nil, nil)
// The scrape pool synchronizes on stopping scrape loops. However, new scrape
// loops are started asynchronously. Thus it's possible, that a loop is stopped
2016-02-28 00:51:02 -08:00
// again before having started properly.
// Stopping not-yet-started loops must block until the run method was called and exited.
// The run method must exit immediately.
stopDone := make(chan struct{})
go func() {
sl.stop()
close(stopDone)
}()
select {
case <-stopDone:
t.Fatalf("Stopping terminated before run exited successfully")
case <-time.After(500 * time.Millisecond):
}
// Running the scrape loop must exit before calling the scraper even once.
scraper.scrapeFunc = func(context.Context, io.Writer) error {
2016-02-28 00:51:02 -08:00
t.Fatalf("scraper was called for terminated scrape loop")
return nil
2016-02-28 00:51:02 -08:00
}
runDone := make(chan struct{})
go func() {
sl.run(1, 0, nil)
2016-02-28 00:51:02 -08:00
close(runDone)
}()
select {
case <-runDone:
case <-time.After(1 * time.Second):
t.Fatalf("Running terminated scrape loop did not exit")
}
select {
case <-stopDone:
case <-time.After(1 * time.Second):
t.Fatalf("Stopping did not terminate after running exited")
}
}
2016-02-23 01:58:16 -08:00
func TestScrapeLoopRun(t *testing.T) {
var (
signal = make(chan struct{})
errc = make(chan error)
scraper = &testScraper{}
2017-01-13 05:48:01 -08:00
app = func() storage.Appender { return &nopAppender{} }
reportApp = func() storage.Appender { return &nopAppender{} }
2016-02-23 01:58:16 -08:00
)
defer close(signal)
ctx, cancel := context.WithCancel(context.Background())
sl := newScrapeLoop(ctx, scraper, app, reportApp)
// The loop must terminate during the initial offset if the context
// is canceled.
scraper.offsetDur = time.Hour
go func() {
sl.run(time.Second, time.Hour, errc)
signal <- struct{}{}
}()
// Wait to make sure we are actually waiting on the offset.
time.Sleep(1 * time.Second)
cancel()
select {
case <-signal:
case <-time.After(5 * time.Second):
t.Fatalf("Cancelation during initial offset failed")
case err := <-errc:
t.Fatalf("Unexpected error: %s", err)
}
// The provided timeout must cause cancelation of the context passed down to the
// scraper. The scraper has to respect the context.
scraper.offsetDur = 0
block := make(chan struct{})
scraper.scrapeFunc = func(ctx context.Context, _ io.Writer) error {
2016-02-23 01:58:16 -08:00
select {
case <-block:
case <-ctx.Done():
return ctx.Err()
2016-02-23 01:58:16 -08:00
}
return nil
2016-02-23 01:58:16 -08:00
}
ctx, cancel = context.WithCancel(context.Background())
sl = newScrapeLoop(ctx, scraper, app, reportApp)
go func() {
sl.run(time.Second, 100*time.Millisecond, errc)
signal <- struct{}{}
}()
select {
case err := <-errc:
if err != context.DeadlineExceeded {
t.Fatalf("Expected timeout error but got: %s", err)
}
case <-time.After(3 * time.Second):
t.Fatalf("Expected timeout error but got none")
}
// We already caught the timeout error and are certainly in the loop.
// Let the scrapes returns immediately to cause no further timeout errors
// and check whether canceling the parent context terminates the loop.
close(block)
cancel()
select {
case <-signal:
// Loop terminated as expected.
case err := <-errc:
t.Fatalf("Unexpected error: %s", err)
case <-time.After(3 * time.Second):
t.Fatalf("Loop did not terminate on context cancelation")
}
}
func TestScrapeLoopAppend(t *testing.T) {
app := &collectResultAppender{}
sl := &scrapeLoop{
appender: func() storage.Appender { return app },
reportAppender: func() storage.Appender { return nopAppender{} },
cache: map[string]uint64{},
}
now := time.Now()
_, _, err := sl.append([]byte("metric_a 1\nmetric_b NaN\n"), now)
if err != nil {
t.Fatalf("Unexpected append error: %s", err)
}
ingestedNaN := math.Float64bits(app.result[1].v)
if ingestedNaN != normalNaN {
t.Fatalf("Appended NaN samples wasn't as expected. Wanted: %x Got: %x", normalNaN, ingestedNaN)
}
// DeepEqual will report NaNs as being different, so replace with a different value.
app.result[1].v = 42
want := []sample{
{
metric: labels.FromStrings(model.MetricNameLabel, "metric_a"),
t: timestamp.FromTime(now),
v: 1,
},
{
metric: labels.FromStrings(model.MetricNameLabel, "metric_b"),
t: timestamp.FromTime(now),
v: 42,
},
}
if !reflect.DeepEqual(want, app.result) {
t.Fatalf("Appended samples not as expected. Wanted: %+v Got: %+v", want, app.result)
}
}
2016-02-28 14:59:03 -08:00
func TestTargetScraperScrapeOK(t *testing.T) {
const (
configTimeout = 1500 * time.Millisecond
expectedTimeout = "1.500000"
)
2016-02-28 14:59:03 -08:00
server := httptest.NewServer(
http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
timeout := r.Header.Get("X-Prometheus-Scrape-Timeout-Seconds")
if timeout != expectedTimeout {
t.Errorf("Scrape timeout did not match expected timeout")
t.Errorf("Expected: %v", expectedTimeout)
t.Fatalf("Got: %v", timeout)
}
2016-02-28 14:59:03 -08:00
w.Header().Set("Content-Type", `text/plain; version=0.0.4`)
w.Write([]byte("metric_a 1\nmetric_b 2\n"))
}),
)
defer server.Close()
serverURL, err := url.Parse(server.URL)
if err != nil {
panic(err)
}
ts := &targetScraper{
Target: &Target{
labels: labels.FromStrings(
model.SchemeLabel, serverURL.Scheme,
model.AddressLabel, serverURL.Host,
),
2016-02-28 14:59:03 -08:00
},
client: http.DefaultClient,
timeout: configTimeout,
2016-02-28 14:59:03 -08:00
}
var buf bytes.Buffer
2016-02-28 14:59:03 -08:00
if err := ts.scrape(context.Background(), &buf); err != nil {
2016-02-28 14:59:03 -08:00
t.Fatalf("Unexpected scrape error: %s", err)
}
require.Equal(t, "metric_a 1\nmetric_b 2\n", buf.String())
2016-02-28 14:59:03 -08:00
}
func TestTargetScrapeScrapeCancel(t *testing.T) {
block := make(chan struct{})
server := httptest.NewServer(
http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
<-block
}),
)
defer server.Close()
serverURL, err := url.Parse(server.URL)
if err != nil {
panic(err)
}
ts := &targetScraper{
Target: &Target{
labels: labels.FromStrings(
model.SchemeLabel, serverURL.Scheme,
model.AddressLabel, serverURL.Host,
),
2016-02-28 14:59:03 -08:00
},
client: http.DefaultClient,
}
ctx, cancel := context.WithCancel(context.Background())
errc := make(chan error)
2016-02-28 14:59:03 -08:00
go func() {
time.Sleep(1 * time.Second)
cancel()
}()
go func() {
if err := ts.scrape(ctx, ioutil.Discard); err != context.Canceled {
errc <- fmt.Errorf("Expected context cancelation error but got: %s", err)
2016-02-28 14:59:03 -08:00
}
close(errc)
2016-02-28 14:59:03 -08:00
}()
select {
case <-time.After(5 * time.Second):
t.Fatalf("Scrape function did not return unexpectedly")
case err := <-errc:
if err != nil {
t.Fatalf(err.Error())
}
2016-02-28 14:59:03 -08:00
}
// If this is closed in a defer above the function the test server
// does not terminate and the test doens't complete.
close(block)
}
func TestTargetScrapeScrapeNotFound(t *testing.T) {
server := httptest.NewServer(
http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.WriteHeader(http.StatusNotFound)
}),
)
defer server.Close()
serverURL, err := url.Parse(server.URL)
if err != nil {
panic(err)
}
ts := &targetScraper{
Target: &Target{
labels: labels.FromStrings(
model.SchemeLabel, serverURL.Scheme,
model.AddressLabel, serverURL.Host,
),
2016-02-28 14:59:03 -08:00
},
client: http.DefaultClient,
}
if err := ts.scrape(context.Background(), ioutil.Discard); !strings.Contains(err.Error(), "404") {
2016-02-28 14:59:03 -08:00
t.Fatalf("Expected \"404 NotFound\" error but got: %s", err)
}
}
2016-02-23 01:58:16 -08:00
// testScraper implements the scraper interface and allows setting values
// returned by its methods. It also allows setting a custom scrape function.
type testScraper struct {
offsetDur time.Duration
lastStart time.Time
lastDuration time.Duration
lastError error
samples samples
2016-02-23 01:58:16 -08:00
scrapeErr error
scrapeFunc func(context.Context, io.Writer) error
2016-02-23 01:58:16 -08:00
}
func (ts *testScraper) offset(interval time.Duration) time.Duration {
return ts.offsetDur
}
func (ts *testScraper) report(start time.Time, duration time.Duration, err error) {
ts.lastStart = start
ts.lastDuration = duration
ts.lastError = err
}
func (ts *testScraper) scrape(ctx context.Context, w io.Writer) error {
2016-02-23 01:58:16 -08:00
if ts.scrapeFunc != nil {
return ts.scrapeFunc(ctx, w)
2016-02-23 01:58:16 -08:00
}
return ts.scrapeErr
2016-02-23 01:58:16 -08:00
}