Fix the scraper/discover manager coordination on the Prometheus recei…

…ver (#2089)

* Fix the scraper/discover manager coordination on the Prometheus receiver

The receiver contains various unnecessary sections. Rewriting the
receiver's Start for better maintainability.

Related to #1909.

* Use the background context

* Remove dead code

receiver/prometheusreceiver/internal/ocastore.go

@@ -17,7 +17,6 @@ package internal
 import (
 	"context"
 	"io"
-	"sync"
 	"sync/atomic"
 
 	"github.com/prometheus/prometheus/pkg/labels"
@@ -47,16 +46,17 @@ type OcaStore interface {
 
 // OpenCensus Store for prometheus
 type ocaStore struct {
-	running              int32
-	logger               *zap.Logger
+	ctx context.Context
+
+	running              int32 // access atomically
 	sink                 consumer.MetricsConsumer
 	mc                   *mService
-	once                 *sync.Once
-	ctx                  context.Context
 	jobsMap              *JobsMap
 	useStartTimeMetric   bool
 	startTimeMetricRegex string
 	receiverName         string
+
+	logger *zap.Logger
 }
 
 // NewOcaStore returns an ocaStore instance, which can be acted as prometheus' scrape.Appendable
@@ -66,7 +66,6 @@ func NewOcaStore(ctx context.Context, sink consumer.MetricsConsumer, logger *zap
 		ctx:                  ctx,
 		sink:                 sink,
 		logger:               logger,
-		once:                 &sync.Once{},
 		jobsMap:              jobsMap,
 		useStartTimeMetric:   useStartTimeMetric,
 		startTimeMetricRegex: startTimeMetricRegex,

receiver/prometheusreceiver/metrics_receiver.go

@@ -16,7 +16,6 @@ package prometheusreceiver
 
 import (
 	"context"
-	"sync"
 	"time"
 
 	"github.com/prometheus/prometheus/discovery"
@@ -25,18 +24,16 @@ import (
 
 	"go.opentelemetry.io/collector/component"
 	"go.opentelemetry.io/collector/consumer"
-	"go.opentelemetry.io/collector/obsreport"
 	"go.opentelemetry.io/collector/receiver/prometheusreceiver/internal"
 )
 
 // pReceiver is the type that provides Prometheus scraper/receiver functionality.
 type pReceiver struct {
-	startOnce sync.Once
-	stopOnce  sync.Once
-	cfg       *Config
-	consumer  consumer.MetricsConsumer
-	cancel    context.CancelFunc
-	logger    *zap.Logger
+	cfg        *Config
+	consumer   consumer.MetricsConsumer
+	cancelFunc context.CancelFunc
+
+	logger *zap.Logger
 }
 
 // New creates a new prometheus.Receiver reference.
@@ -51,62 +48,49 @@ func newPrometheusReceiver(logger *zap.Logger, cfg *Config, next consumer.Metric
 
 // Start is the method that starts Prometheus scraping and it
 // is controlled by having previously defined a Configuration using perhaps New.
-func (pr *pReceiver) Start(_ context.Context, host component.Host) error {
-	pr.startOnce.Do(func() {
-		ctx := context.Background()
-		c, cancel := context.WithCancel(ctx)
-		pr.cancel = cancel
-		c = obsreport.ReceiverContext(c, pr.cfg.Name(), "http")
-		var jobsMap *internal.JobsMap
-		if !pr.cfg.UseStartTimeMetric {
-			jobsMap = internal.NewJobsMap(2 * time.Minute)
-		}
-		app := internal.NewOcaStore(c, pr.consumer, pr.logger, jobsMap, pr.cfg.UseStartTimeMetric, pr.cfg.StartTimeMetricRegex, pr.cfg.Name())
-		// need to use a logger with the gokitLog interface
-		l := internal.NewZapToGokitLogAdapter(pr.logger)
-		scrapeManager := scrape.NewManager(l, app)
-		app.SetScrapeManager(scrapeManager)
-		discoveryManagerScrape := discovery.NewManager(ctx, l)
-		go func() {
-			if err := discoveryManagerScrape.Run(); err != nil {
-				host.ReportFatalError(err)
-			}
-		}()
-		if err := scrapeManager.ApplyConfig(pr.cfg.PrometheusConfig); err != nil {
-			host.ReportFatalError(err)
-			return
-		}
+func (r *pReceiver) Start(ctx context.Context, host component.Host) error {
+	discoveryCtx, cancel := context.WithCancel(context.Background())
+	r.cancelFunc = cancel
 
-		// Run the scrape manager.
-		syncConfig := make(chan bool)
-		errsChan := make(chan error, 1)
-		go func() {
-			defer close(errsChan)
-			<-time.After(100 * time.Millisecond)
-			close(syncConfig)
-			if err := scrapeManager.Run(discoveryManagerScrape.SyncCh()); err != nil {
-				errsChan <- err
-			}
-		}()
-		<-syncConfig
-		// By this point we've given time to the scrape manager
-		// to start applying its original configuration.
+	logger := internal.NewZapToGokitLogAdapter(r.logger)
 
-		discoveryCfg := make(map[string]discovery.Configs)
-		for _, scrapeConfig := range pr.cfg.PrometheusConfig.ScrapeConfigs {
-			discoveryCfg[scrapeConfig.JobName] = scrapeConfig.ServiceDiscoveryConfigs
+	discoveryManager := discovery.NewManager(discoveryCtx, logger)
+	discoveryCfg := make(map[string]discovery.Configs)
+	for _, scrapeConfig := range r.cfg.PrometheusConfig.ScrapeConfigs {
+		discoveryCfg[scrapeConfig.JobName] = scrapeConfig.ServiceDiscoveryConfigs
+	}
+	if err := discoveryManager.ApplyConfig(discoveryCfg); err != nil {
+		return err
+	}
+	go func() {
+		if err := discoveryManager.Run(); err != nil {
+			r.logger.Error("Discovery manager failed", zap.Error(err))
+			host.ReportFatalError(err)
 		}
+	}()
+
+	var jobsMap *internal.JobsMap
+	if !r.cfg.UseStartTimeMetric {
+		jobsMap = internal.NewJobsMap(2 * time.Minute)
+	}
+	ocaStore := internal.NewOcaStore(ctx, r.consumer, r.logger, jobsMap, r.cfg.UseStartTimeMetric, r.cfg.StartTimeMetricRegex, r.cfg.Name())
 
-		// Now trigger the discovery notification to the scrape manager.
-		if err := discoveryManagerScrape.ApplyConfig(discoveryCfg); err != nil {
-			errsChan <- err
+	scrapeManager := scrape.NewManager(logger, ocaStore)
+	ocaStore.SetScrapeManager(scrapeManager)
+	if err := scrapeManager.ApplyConfig(r.cfg.PrometheusConfig); err != nil {
+		return err
+	}
+	go func() {
+		if err := scrapeManager.Run(discoveryManager.SyncCh()); err != nil {
+			r.logger.Error("Scrape manager failed", zap.Error(err))
+			host.ReportFatalError(err)
 		}
-	})
+	}()
 	return nil
 }
 
 // Shutdown stops and cancels the underlying Prometheus scrapers.
-func (pr *pReceiver) Shutdown(context.Context) error {
-	pr.stopOnce.Do(pr.cancel)
+func (r *pReceiver) Shutdown(context.Context) error {
+	r.cancelFunc()
 	return nil
 }