Add basic Prometheus metrics for binlog-collector

The following metrics have been added:
- pxc_binlog_collector_success_total
- pxc_binlog_collector_failure_total
- pxc_binlog_collector_last_processing_timestamp
- pxc_binlog_collector_last_upload_timestamp
- pxc_binlog_collector_gap_detected_total

Additionally, a simple /health endpoint has been added

cmd/pitr/collector/collector.go

@@ -16,11 +16,53 @@ import (
 
 	"github.com/go-sql-driver/mysql"
 	"github.com/pkg/errors"
+	"github.com/prometheus/client_golang/prometheus"
 
 	"github.com/percona/percona-xtradb-cluster-operator/cmd/pitr/pxc"
 	"github.com/percona/percona-xtradb-cluster-operator/pkg/pxc/backup/storage"
 )
 
+var (
+	pxcBinlogCollectorBackupSuccess = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Name: "pxc_binlog_collector_success_total",
+			Help: "Total number of successful binlog backups",
+		},
+	)
+	pxcBinlogCollectorBackupFailure = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Name: "pxc_binlog_collector_failure_total",
+			Help: "Total number of failed binlog backups",
+		},
+	)
+	pxcBinlogCollectorLastProcessingTime = prometheus.NewGauge(
+		prometheus.GaugeOpts{
+			Name: "pxc_binlog_collector_last_processing_timestamp",
+			Help: "Timestamp of the last successful binlog processing",
+		},
+	)
+	pxcBinlogCollectorLastUploadTime = prometheus.NewGauge(
+		prometheus.GaugeOpts{
+			Name: "pxc_binlog_collector_last_upload_timestamp",
+			Help: "Timestamp of the last successful binlog upload",
+		},
+	)
+	pxcBinlogCollectorGapDetected = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Name: "pxc_binlog_collector_gap_detected_total",
+			Help: "Total number of detected gaps in binlog uploads",
+		},
+	)
+)
+
+func init() {
+	prometheus.MustRegister(pxcBinlogCollectorBackupSuccess)
+	prometheus.MustRegister(pxcBinlogCollectorBackupFailure)
+	prometheus.MustRegister(pxcBinlogCollectorLastProcessingTime)
+	prometheus.MustRegister(pxcBinlogCollectorLastUploadTime)
+	prometheus.MustRegister(pxcBinlogCollectorGapDetected)
+}
+
 type Collector struct {
 	db              *pxc.PXC
 	storage         storage.Storage
@@ -103,6 +145,7 @@ func New(ctx context.Context, c Config) (*Collector, error) {
 func (c *Collector) Run(ctx context.Context) error {
 	err := c.newDB(ctx)
 	if err != nil {
+		pxcBinlogCollectorBackupFailure.Inc()
 		return errors.Wrap(err, "new db connection")
 	}
 	defer c.close()
@@ -113,9 +156,11 @@ func (c *Collector) Run(ctx context.Context) error {
 
 	err = c.CollectBinLogs(ctx)
 	if err != nil {
+		pxcBinlogCollectorBackupFailure.Inc()
 		return errors.Wrap(err, "collect binlog files")
 	}
 
+	pxcBinlogCollectorBackupSuccess.Inc()
 	return nil
 }
 
@@ -369,6 +414,7 @@ func (c *Collector) CollectBinLogs(ctx context.Context) error {
 		if lastUploadedBinlogName == "" {
 			log.Println("ERROR: Couldn't find the binlog that contains GTID set:", c.lastUploadedSet.Raw())
 			log.Println("ERROR: Gap detected in the binary logs. Binary logs will be uploaded anyway, but full backup needed for consistent recovery.")
+			pxcBinlogCollectorGapDetected.Inc()
 			if err := createGapFile(c.lastUploadedSet); err != nil {
 				return errors.Wrap(err, "create gap file")
 			}
@@ -382,6 +428,7 @@ func (c *Collector) CollectBinLogs(ctx context.Context) error {
 
 	if len(list) == 0 {
 		log.Println("No binlogs to upload")
+		pxcBinlogCollectorLastProcessingTime.SetToCurrentTime()
 		return nil
 	}
 
@@ -411,6 +458,9 @@ func (c *Collector) CollectBinLogs(ctx context.Context) error {
 			return errors.Wrap(err, "update timeline file")
 		}
 	}
+
+	pxcBinlogCollectorLastUploadTime.SetToCurrentTime()
+	pxcBinlogCollectorLastProcessingTime.SetToCurrentTime()
 	return nil
 }
 

cmd/pitr/main.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"fmt"
 	"log"
+	"net/http"
 	"os"
 	"os/signal"
 	"syscall"
@@ -14,6 +15,7 @@ import (
 	"github.com/percona/percona-xtradb-cluster-operator/cmd/pitr/recoverer"
 
 	"github.com/caarlos0/env"
+	"github.com/prometheus/client_golang/prometheus/promhttp"
 )
 
 func main() {
@@ -23,6 +25,13 @@ func main() {
 	}
 	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGTERM, os.Interrupt)
 	defer stop()
+
+	go func() {
+		http.Handle("/metrics", promhttp.Handler())
+		http.HandleFunc("/health", healthHandler)
+		log.Fatal(http.ListenAndServe(":8080", nil))
+	}()
+
 	switch command {
 	case "collect":
 		runCollector(ctx)
@@ -34,6 +43,13 @@ func main() {
 	}
 }
 
+func healthHandler(w http.ResponseWriter, r *http.Request) {
+	w.WriteHeader(http.StatusOK)
+	if _, err := w.Write([]byte("ok")); err != nil {
+		log.Println("ERROR: writing health response:", err)
+	}
+}
+
 func runCollector(ctx context.Context) {
 	config, err := getCollectorConfig()
 	if err != nil {

pkg/pxc/app/deployment/binlog-collector.go

@@ -91,6 +91,16 @@ func GetBinlogCollectorDeployment(cr *api.PerconaXtraDBCluster, initImage string
 			},
 		},
 	}
+
+	if cr.CompareVersionWith("1.16.0") >= 0 {
+		container.Ports = []corev1.ContainerPort{
+			{
+				ContainerPort: 8080,
+				Name:          "metrics",
+			},
+		}
+	}
+
 	replicas := int32(1)
 
 	var initContainers []corev1.Container