bring back prometheus metrics, with new metrics for replication state machine

zfs/prometheus.go

@@ -3,10 +3,10 @@ package zfs
 import "github.com/prometheus/client_golang/prometheus"
 
 var prom struct {
-	ZFSListFilesystemVersionDuration    *prometheus.HistogramVec
-	ZFSDestroyFilesystemVersionDuration *prometheus.HistogramVec
-	ZFSSnapshotDuration                 *prometheus.HistogramVec
-	ZFSBookmarkDuration                 *prometheus.HistogramVec
+	ZFSListFilesystemVersionDuration *prometheus.HistogramVec
+	ZFSSnapshotDuration              *prometheus.HistogramVec
+	ZFSBookmarkDuration              *prometheus.HistogramVec
+	ZFSDestroyDuration               *prometheus.HistogramVec
 }
 
 func init() {
@@ -16,12 +16,6 @@ func init() {
 		Name:      "list_filesystem_versions_duration",
 		Help:      "Seconds it took for listing the versions of a given filesystem",
 	}, []string{"filesystem"})
-	prom.ZFSDestroyFilesystemVersionDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
-		Namespace: "zrepl",
-		Subsystem: "zfs",
-		Name:      "destroy_filesystem_version_duration",
-		Help:      "Seconds it took to destroy a version of a given filesystem",
-	}, []string{"filesystem", "version_type"})
 	prom.ZFSSnapshotDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
 		Namespace: "zrepl",
 		Subsystem: "zfs",
@@ -34,20 +28,26 @@ func init() {
 		Name:      "bookmark_duration",
 		Help:      "Duration it took to bookmark a given snapshot",
 	}, []string{"filesystem"})
+	prom.ZFSDestroyDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
+		Namespace: "zrepl",
+		Subsystem: "zfs",
+		Name:      "destroy_duration",
+		Help:      "Duration it took to destroy a dataset",
+	}, []string{"dataset_type", "filesystem"})
 }
 
 func PrometheusRegister(registry prometheus.Registerer) error {
 	if err := registry.Register(prom.ZFSListFilesystemVersionDuration); err != nil {
 		return err
 	}
-	if err := registry.Register(prom.ZFSDestroyFilesystemVersionDuration); err != nil {
-		return err
-	}
 	if err := registry.Register(prom.ZFSBookmarkDuration); err != nil {
 		return err
 	}
 	if err := registry.Register(prom.ZFSSnapshotDuration); err != nil {
 		return err
 	}
+	if err := registry.Register(prom.ZFSDestroyDuration); err != nil {
+		return err
+	}
 	return nil
 }

zfs/versions.go

@@ -156,9 +156,6 @@ func ZFSListFilesystemVersions(fs *DatasetPath, filter FilesystemVersionFilter)
 
 func ZFSDestroyFilesystemVersion(filesystem *DatasetPath, version *FilesystemVersion) (err error) {
 
-	promTimer := prometheus.NewTimer(prom.ZFSDestroyFilesystemVersionDuration.WithLabelValues(filesystem.ToString(), version.Type.String()))
-	defer promTimer.ObserveDuration()
-
 	datasetPath := version.ToAbsPath(filesystem)
 
 	// Sanity check...

zfs/zfs.go

@@ -566,6 +566,21 @@ func zfsGet(path string, props []string, allowedSources zfsPropertySource) (*ZFS
 
 func ZFSDestroy(dataset string) (err error) {
 
+	var dstype, filesystem string
+	idx := strings.IndexAny(dataset, "@#")
+	if idx == -1 {
+		dstype = "filesystem"
+		filesystem = dataset
+	} else {
+		switch dataset[idx] {
+		case '@': dstype = "snapshot"
+		case '#': dstype = "bookmark"
+		}
+		filesystem = dataset[:idx]
+	}
+
+	defer prometheus.NewTimer(prom.ZFSDestroyDuration.WithLabelValues(dstype, filesystem))
+
 	cmd := exec.Command(ZFS_BINARY, "destroy", dataset)
 
 	stderr := bytes.NewBuffer(make([]byte, 0, 1024))