Merge pull request #7913 from prometheus/release-2.21

Merge release 2.21 into master

CHANGELOG.md

+## 2.21.0-rc.1 / 2020-09-08
+
+This release is built with Go 1.15, which deprecates [X.509
+CommonName](https://golang.org/doc/go1.15#commonname) in TLS certificates
+validation.
+
+In the unlikely case that you use the gRPC API v2 (which is limited to TSDB
+admin commands), please note that we will remove this experimental API in the
+next minor release 2.22.
+
+* [CHANGE] Disable HTTP2 because of concerns with the Go HTTP/2 client. #7588 #7701
+* [CHANGE] PromQL: `query_log_file` path is now relative to the config file. #7701
+* [CHANGE] Promtool: Replace the tsdb command line tool by a promtool tsdb subcommand. #6088
+* [CHANGE] Rules: Label `rule_group_iterations` metric with group name. #7823
+* [FEATURE] Eureka SD: New service discovery. #3369
+* [FEATURE] Hetzner SD: New service discovery. #7822
+* [FEATURE] Kubernetes SD: Support Kubernetes EndpointSlices. #6838
+* [FEATURE] Scrape: Add per scrape-config targets limit. #7554
+* [ENHANCEMENT] Support composite durations in PromQL, config and UI, e.g. 1h30m. #7713 #7833
+* [ENHANCEMENT] DNS SD: Add SRV record target and port meta labels. #7678
+* [ENHANCEMENT] Docker Swarm SD: Support tasks and service without published ports. #7686
+* [ENHANCEMENT] PromQL: Reduce the amount of data queried by remote read when a subquery has an offset. #7667
+* [ENHANCEMENT] Promtool: Add `--time` option to query instant command. #7829
+* [ENHANCEMENT] UI: Respect the `--web.page-title` parameter in the React UI. #7607
+* [ENHANCEMENT] UI: Add duration, labels, annotations to alerts page in the React UI. #7605
+* [ENHANCEMENT] UI: Add duration on the React UI rules page, hide annotation and labels if empty. #7606
+* [BUGFIX] API: Deduplicate series in /api/v1/series. #7862
+* [BUGFIX] PromQL: Drop metric name in bool comparison between two instant vectors. #7819
+* [BUGFIX] PromQL: Exit with an error when time parameters can't be parsed. #7505
+* [BUGFIX] Rules: Detect extra fields in rule files. #7767
+* [BUGFIX] Rules: Disallow overwriting the metric name in the `labels` section of recording rules. #7787
+* [BUGFIX] Rules: Keep evaluation timestamp across reloads. #7775
+* [BUGFIX] Scrape: Do not stop scrapes in progress during reload. #7752
+* [BUGFIX] TSDB: Fix `chunks.HeadReadWriter: maxt of the files are not set` error. #7856
+* [BUGFIX] TSDB: Delete blocks atomically to prevent corruption when there is a panic/crash during deletion. #7772
+* [BUGFIX] Triton SD: Fix a panic when triton_sd_config is nil. #7671
+* [BUGFIX] UI: Fix react UI bug with series going on and off. #7804
+* [BUGFIX] Web: Stop CMUX and GRPC servers even with stale connections, preventing the server to stop on SIGTERM. #7810
+
 ## 2.20.1 / 2020-08-05
 
 * [BUGFIX] SD: Reduce the Consul watch timeout to 2m and adjust the request timeout accordingly. #7724

VERSION

-2.20.1
+2.21.0-rc.1

discovery/hetzner/robot.go

@@ -26,7 +26,8 @@ import (
 	"time"
 
 	"github.com/go-kit/kit/log"
-	config_util "github.com/prometheus/common/config"
+	"github.com/pkg/errors"
+	"github.com/prometheus/common/config"
 	"github.com/prometheus/common/model"
 	"github.com/prometheus/prometheus/discovery/refresh"
 	"github.com/prometheus/prometheus/discovery/targetgroup"
@@ -54,7 +55,7 @@ func newRobotDiscovery(conf *SDConfig, logger log.Logger) (*robotDiscovery, erro
 		endpoint: conf.robotEndpoint,
 	}
 
-	rt, err := config_util.NewRoundTripperFromConfig(conf.HTTPClientConfig, "hetzner_sd", false, false)
+	rt, err := config.NewRoundTripperFromConfig(conf.HTTPClientConfig, "hetzner_sd", false, false)
 	if err != nil {
 		return nil, err
 	}
@@ -70,10 +71,16 @@ func (d *robotDiscovery) refresh(ctx context.Context) ([]*targetgroup.Group, err
 	if err != nil {
 		return nil, err
 	}
+
 	defer func() {
 		io.Copy(ioutil.Discard, resp.Body)
 		resp.Body.Close()
 	}()
+
+	if resp.StatusCode/100 != 2 {
+		return nil, errors.Errorf("non 2xx status '%d' response during hetzner service discovery with role robot", resp.StatusCode)
+	}
+
 	var servers serversList
 	err = json.NewDecoder(resp.Body).Decode(&servers)
 	if err != nil {

discovery/hetzner/robot_test.go

@@ -37,7 +37,6 @@ func (s *robotSDTestSuite) SetupTest(t *testing.T) {
 func TestRobotSDRefresh(t *testing.T) {
 	suite := &robotSDTestSuite{}
 	suite.SetupTest(t)
-
 	cfg := DefaultSDConfig
 	cfg.HTTPClientConfig.BasicAuth = &config.BasicAuth{Username: robotTestUsername, Password: robotTestPassword}
 	cfg.robotEndpoint = suite.Mock.Endpoint()
@@ -84,3 +83,19 @@ func TestRobotSDRefresh(t *testing.T) {
 		})
 	}
 }
+
+func TestRobotSDRefreshHandleError(t *testing.T) {
+	suite := &robotSDTestSuite{}
+	suite.SetupTest(t)
+	cfg := DefaultSDConfig
+	cfg.robotEndpoint = suite.Mock.Endpoint()
+
+	d, err := newRobotDiscovery(&cfg, log.NewNopLogger())
+	testutil.Ok(t, err)
+
+	targetGroups, err := d.refresh(context.Background())
+	testutil.NotOk(t, err)
+	testutil.Equals(t, "non 2xx status '401' response during hetzner service discovery with role robot", err.Error())
+
+	testutil.Equals(t, 0, len(targetGroups))
+}

docs/configuration/configuration.md

@@ -936,6 +936,7 @@ changed with relabeling, as demonstrated in [the Prometheus hetzner-sd
 configuration file](/documentation/examples/prometheus-hetzner.yml).
 
 The following meta labels are available on all targets during [relabeling](#relabel_config):
+
 * `__meta_hetzner_server_id`: the ID of the server
 * `__meta_hetzner_server_name`: the name of the server
 * `__meta_hetzner_server_status`: the status of the server
@@ -944,6 +945,7 @@ The following meta labels are available on all targets during [relabeling](#rela
 * `__meta_hetzner_datacenter`: the datacenter of the server
 
 The labels below are only available for targets with `role` set to `hcloud`:
+
 * `__meta_hetzner_hcloud_image_name`: the image name of the server
 * `__meta_hetzner_hcloud_image_description`: the description of the server image
 * `__meta_hetzner_hcloud_image_os_flavor`: the OS flavor of the server image
@@ -960,6 +962,7 @@ The labels below are only available for targets with `role` set to `hcloud`:
 * `__meta_hetzner_hcloud_label_<labelname>`: each label of the server
 
 The labels below are only available for targets with `role` set to `robot`:
+
 * `__meta_hetzner_robot_product`: the product of the server
 * `__meta_hetzner_robot_cancelled`: the server cancellation status
 

storage/merge.go

@@ -16,6 +16,7 @@ package storage
 import (
 	"bytes"
 	"container/heap"
+	"math"
 	"sort"
 	"strings"
 	"sync"
@@ -331,7 +332,7 @@ func (c *genericMergeSeriesSet) Next() bool {
 	// If, for the current label set, all the next series sets come from
 	// failed remote storage sources, we want to keep trying with the next label set.
 	for {
-		// Firstly advance all the current series sets. If any of them have run out
+		// Firstly advance all the current series sets. If any of them have run out,
 		// we can drop them, otherwise they should be inserted back into the heap.
 		for _, set := range c.currentSets {
 			if set.Next() {
@@ -418,8 +419,7 @@ func (h *genericSeriesSetHeap) Pop() interface{} {
 // with "almost" the same data, e.g. from 2 Prometheus HA replicas. This is fine, since from the Prometheus perspective
 // this never happens.
 //
-// NOTE: Use this merge function only when you see potentially overlapping series, as this introduces a small overhead
-// to handle overlaps between series.
+// It's optimized for non-overlap cases as well.
 func ChainedSeriesMerge(series ...Series) Series {
 	if len(series) == 0 {
 		return nil
@@ -438,16 +438,20 @@ func ChainedSeriesMerge(series ...Series) Series {
 
 // chainSampleIterator is responsible to iterate over samples from different iterators of the same time series in timestamps
 // order. If one or more samples overlap, one sample from random overlapped ones is kept and all others with the same
-// timestamp are dropped.
+// timestamp are dropped. It's optimized for non-overlap cases as well.
 type chainSampleIterator struct {
 	iterators []chunkenc.Iterator
 	h         samplesIteratorHeap
+
+	curr  chunkenc.Iterator
+	lastt int64
 }
 
 func newChainSampleIterator(iterators []chunkenc.Iterator) chunkenc.Iterator {
 	return &chainSampleIterator{
 		iterators: iterators,
 		h:         nil,
+		lastt:     math.MinInt64,
 	}
 }
 
@@ -458,47 +462,74 @@ func (c *chainSampleIterator) Seek(t int64) bool {
 			heap.Push(&c.h, iter)
 		}
 	}
-	return len(c.h) > 0
+	if len(c.h) > 0 {
+		c.curr = heap.Pop(&c.h).(chunkenc.Iterator)
+		return true
+	}
+	c.curr = nil
+	return false
 }
 
 func (c *chainSampleIterator) At() (t int64, v float64) {
-	if len(c.h) == 0 {
-		panic("chainSampleIterator.At() called after .Next() returned false.")
+	if c.curr == nil {
+		panic("chainSampleIterator.At() called before first .Next() or after .Next() returned false.")
 	}
-
-	return c.h[0].At()
+	return c.curr.At()
 }
 
 func (c *chainSampleIterator) Next() bool {
 	if c.h == nil {
-		for _, iter := range c.iterators {
+		c.h = samplesIteratorHeap{}
+		// We call c.curr.Next() as the first thing below.
+		// So, we don't call Next() on it here.
+		c.curr = c.iterators[0]
+		for _, iter := range c.iterators[1:] {
 			if iter.Next() {
 				heap.Push(&c.h, iter)
 			}
 		}
-
-		return len(c.h) > 0
 	}
 
-	if len(c.h) == 0 {
+	if c.curr == nil {
 		return false
 	}
 
-	currt, _ := c.At()
-	for len(c.h) > 0 {
-		nextt, _ := c.h[0].At()
-		// All but one of the overlapping samples will be dropped.
-		if nextt != currt {
-			break
+	var currt int64
+	for {
+		if c.curr.Next() {
+			currt, _ = c.curr.At()
+			if currt == c.lastt {
+				// Ignoring sample for the same timestamp.
+				continue
+			}
+			if len(c.h) == 0 {
+				// curr is the only iterator remaining,
+				// no need to check with the heap.
+				break
+			}
+
+			// Check current iterator with the top of the heap.
+			if nextt, _ := c.h[0].At(); currt < nextt {
+				// Current iterator has smaller timestamp than the heap.
+				break
+			}
+			// Current iterator does not hold the smallest timestamp.
+			heap.Push(&c.h, c.curr)
+		} else if len(c.h) == 0 {
+			// No iterator left to iterate.
+			c.curr = nil
+			return false
 		}
 
-		iter := heap.Pop(&c.h).(chunkenc.Iterator)
-		if iter.Next() {
-			heap.Push(&c.h, iter)
+		c.curr = heap.Pop(&c.h).(chunkenc.Iterator)
+		currt, _ = c.curr.At()
+		if currt != c.lastt {
+			break
 		}
 	}
 
-	return len(c.h) > 0
+	c.lastt = currt
+	return true
 }
 
 func (c *chainSampleIterator) Err() error {

storage/merge_test.go

@@ -448,10 +448,10 @@ func TestCompactingChunkSeriesMerger(t *testing.T) {
 			name: "three in chained overlap",
 			input: []ChunkSeries{
 				NewListChunkSeriesFromSamples(labels.FromStrings("bar", "baz"), []tsdbutil.Sample{sample{1, 1}, sample{2, 2}, sample{3, 3}, sample{5, 5}}),
-				NewListChunkSeriesFromSamples(labels.FromStrings("bar", "baz"), []tsdbutil.Sample{sample{4, 4}, sample{6, 6}}),
+				NewListChunkSeriesFromSamples(labels.FromStrings("bar", "baz"), []tsdbutil.Sample{sample{4, 4}, sample{6, 66}}),
 				NewListChunkSeriesFromSamples(labels.FromStrings("bar", "baz"), []tsdbutil.Sample{sample{6, 6}, sample{10, 10}}),
 			},
-			expected: NewListChunkSeriesFromSamples(labels.FromStrings("bar", "baz"), []tsdbutil.Sample{sample{1, 1}, sample{2, 2}, sample{3, 3}, sample{4, 4}, sample{5, 5}, sample{6, 6}, sample{10, 10}}),
+			expected: NewListChunkSeriesFromSamples(labels.FromStrings("bar", "baz"), []tsdbutil.Sample{sample{1, 1}, sample{2, 2}, sample{3, 3}, sample{4, 4}, sample{5, 5}, sample{6, 66}, sample{10, 10}}),
 		},
 		{
 			name: "three in chained overlap complex",

web/api/v1/api.go

@@ -603,7 +603,8 @@ func (api *API) series(r *http.Request) (result apiFuncResult) {
 
 	var sets []storage.SeriesSet
 	for _, mset := range matcherSets {
-		s := q.Select(false, nil, mset...)
+		// We need to sort this select results to merge (deduplicate) the series sets later.
+		s := q.Select(true, nil, mset...)
 		sets = append(sets, s)
 	}
 

web/api/v1/api_test.go

@@ -303,6 +303,11 @@ func TestEndpoints(t *testing.T) {
 			test_metric1{foo="bar"} 0+100x100
 			test_metric1{foo="boo"} 1+0x100
 			test_metric2{foo="boo"} 1+0x100
+			test_metric3{foo="bar", dup="1"} 1+0x100
+			test_metric3{foo="boo", dup="1"} 1+0x100
+			test_metric4{foo="bar", dup="1"} 1+0x100
+			test_metric4{foo="boo", dup="1"} 1+0x100
+			test_metric4{foo="boo"} 1+0x100
 	`)
 	testutil.Ok(t, err)
 	defer suite.Close()
@@ -737,6 +742,18 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 				labels.FromStrings("__name__", "test_metric1", "foo", "boo"),
 			},
 		},
+		// Try to overlap the selected series set as much as possible to test the result de-duplication works well.
+		{
+			endpoint: api.series,
+			query: url.Values{
+				"match[]": []string{`test_metric4{foo=~".+o$"}`, `test_metric4{dup=~"^1"}`},
+			},
+			response: []labels.Labels{
+				labels.FromStrings("__name__", "test_metric4", "dup", "1", "foo", "bar"),
+				labels.FromStrings("__name__", "test_metric4", "dup", "1", "foo", "boo"),
+				labels.FromStrings("__name__", "test_metric4", "foo", "boo"),
+			},
+		},
 		{
 			endpoint: api.series,
 			query: url.Values{
@@ -1449,6 +1466,8 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 				response: []string{
 					"test_metric1",
 					"test_metric2",
+					"test_metric3",
+					"test_metric4",
 				},
 			},
 			{
@@ -1597,7 +1616,7 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 			// Label names.
 			{
 				endpoint: api.labelNames,
-				response: []string{"__name__", "foo"},
+				response: []string{"__name__", "dup", "foo"},
 			},
 			// Start and end before Label names starts.
 			{
@@ -1615,7 +1634,7 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 					"start": []string{"1"},
 					"end":   []string{"100"},
 				},
-				response: []string{"__name__", "foo"},
+				response: []string{"__name__", "dup", "foo"},
 			},
 			// Start before Label names, end within Label names.
 			{
@@ -1624,7 +1643,7 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 					"start": []string{"-1"},
 					"end":   []string{"10"},
 				},
-				response: []string{"__name__", "foo"},
+				response: []string{"__name__", "dup", "foo"},
 			},
 
 			// Start before Label names starts, end after Label names ends.
@@ -1634,7 +1653,7 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 					"start": []string{"-1"},
 					"end":   []string{"100000"},
 				},
-				response: []string{"__name__", "foo"},
+				response: []string{"__name__", "dup", "foo"},
 			},
 			// Start with bad data for Label names, end within Label names.
 			{
@@ -1652,7 +1671,7 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 					"start": []string{"1"},
 					"end":   []string{"1000000006"},
 				},
-				response: []string{"__name__", "foo"},
+				response: []string{"__name__", "dup", "foo"},
 			},
 			// Start and end after Label names ends.
 			{
@@ -1669,7 +1688,7 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 				query: url.Values{
 					"start": []string{"4"},
 				},
-				response: []string{"__name__", "foo"},
+				response: []string{"__name__", "dup", "foo"},
 			},
 			// Only provide End within Label names, don't provide a start time.
 			{
@@ -1677,7 +1696,7 @@ func testEndpoints(t *testing.T, api *API, tr *testTargetRetriever, testLabelAPI
 				query: url.Values{
 					"end": []string{"20"},
 				},
-				response: []string{"__name__", "foo"},
+				response: []string{"__name__", "dup", "foo"},
 			},
 		}...)
 	}

web/ui/react-app/src/pages/targets/TargetLabels.test.tsx

@@ -31,8 +31,10 @@ describe('targetLabels', () => {
   it('wraps each label in a label badge', () => {
     const l: { [key: string]: string } = defaultProps.labels;
     Object.keys(l).forEach((labelName: string): void => {
-      const badge = targetLabels.find(Badge).filterWhere(badge => badge.hasClass(labelName));
-      expect(badge.children().text()).toEqual(`${labelName}="${l[labelName]}"`);
+      const badge = targetLabels
+        .find(Badge)
+        .filterWhere(badge => badge.children().text() === `${labelName}="${l[labelName]}"`);
+      expect(badge).toHaveLength(1);
     });
     expect(targetLabels.find(Badge)).toHaveLength(3);
   });

web/ui/react-app/src/pages/targets/TargetLabels.tsx

@@ -27,7 +27,7 @@ const TargetLabels: FC<TargetLabelsProps> = ({ discoveredLabels, labels, idx, sc
       <div id={id} className="series-labels-container">
         {Object.keys(labels).map(labelName => {
           return (
-            <Badge color="primary" className={`mr-1 ${labelName}`} key={labelName}>
+            <Badge color="primary" className="mr-1" key={labelName}>
               {`${labelName}="${labels[labelName]}"`}
             </Badge>
           );

web/ui/react-app/src/pages/targets/__snapshots__/TargetLabels.test.tsx.snap

@@ -7,7 +7,7 @@ exports[`targetLabels renders discovered labels 1`] = `
     id="series-labels-cortex/node-exporter_group/0-1"
   >
     <Badge
-      className="mr-1 instance"
+      className="mr-1"
       color="primary"
       key="instance"
       pill={false}
@@ -16,7 +16,7 @@ exports[`targetLabels renders discovered labels 1`] = `
       instance="localhost:9100"
     </Badge>
     <Badge
-      className="mr-1 job"
+      className="mr-1"
       color="primary"
       key="job"
       pill={false}
@@ -25,7 +25,7 @@ exports[`targetLabels renders discovered labels 1`] = `
       job="node_exporter"
     </Badge>
     <Badge
-      className="mr-1 foo"
+      className="mr-1"
       color="primary"
       key="foo"
       pill={false}

web/ui/templates/rules.html

@@ -10,8 +10,8 @@
         <thead>
           <tr>
             <td colspan="3"><h2><a href="#{{reReplaceAll "([^a-zA-Z0-9])" "$1" .Name}}" id="{{reReplaceAll "([^a-zA-Z0-9])" "$1" .Name}}">{{.Name}}</a></h2></td>
-            <td><h2>{{if .GetEvaluationTimestamp.IsZero}}Never{{else}}{{since .GetEvaluationTimestamp}} ago{{end}}</h2></td>
-            <td><h2>{{humanizeDuration .GetEvaluationDuration.Seconds}}</h2></td>
+            <td><h2>{{if .GetLastEvaluation.IsZero}}Never{{else}}{{since .GetLastEvaluation}} ago{{end}}</h2></td>
+            <td><h2>{{humanizeDuration .GetEvaluationTime.Seconds}}</h2></td>
           </tr>
         </thead>
         <tbody>