implement automatic snapshotting feature

cmd/autosnap.go

@@ -0,0 +1,42 @@
+package main
+
+import (
+	"fmt"
+	"github.com/zrepl/zrepl/zfs"
+	"time"
+)
+
+type AutosnapContext struct {
+	Autosnap Autosnap
+}
+
+func doAutosnap(ctx AutosnapContext, log Logger) (err error) {
+
+	snap := ctx.Autosnap
+
+	filesystems, err := zfs.ZFSListMapping(snap.DatasetFilter)
+	if err != nil {
+		return fmt.Errorf("cannot filter datasets: %s", err)
+	}
+
+	suffix := time.Now().In(time.UTC).Format("20060102_150405_000")
+	snapname := fmt.Sprintf("%s%s", snap.Prefix, suffix)
+
+	hadError := false
+
+	for _, fs := range filesystems { // optimization: use recursive snapshots / channel programs here
+		log.Printf("snapshotting filesystem %s@%s", fs, snapname)
+		err := zfs.ZFSSnapshot(fs, snapname, false)
+		if err != nil {
+			log.Printf("error snapshotting %s: %s", fs, err)
+			hadError = true
+		}
+	}
+
+	if hadError {
+		err = fmt.Errorf("errors occurred during autosnap, check logs for details")
+	}
+
+	return
+
+}

cmd/config.go

@@ -67,13 +67,21 @@ type Prune struct {
 	RetentionPolicy *RetentionGrid // TODO abstract interface to support future policies?
 }
 
+type Autosnap struct {
+	Name          string
+	Prefix        string
+	Interval      jobrun.RepeatStrategy
+	DatasetFilter zfs.DatasetMapping
+}
+
 type Config struct {
-	Pools    []Pool
-	Pushs    []Push
-	Pulls    []Pull
-	Sinks    []ClientMapping
-	PullACLs []ClientMapping
-	Prunes   []Prune
+	Pools     []Pool
+	Pushs     []Push
+	Pulls     []Pull
+	Sinks     []ClientMapping
+	PullACLs  []ClientMapping
+	Prunes    []Prune
+	Autosnaps []Autosnap
 }
 
 func ParseConfig(path string) (config Config, err error) {
@@ -122,6 +130,9 @@ func parseMain(root map[string]interface{}) (c Config, err error) {
 	if c.Prunes, err = parsePrunes(root["prune"]); err != nil {
 		return
 	}
+	if c.Autosnaps, err = parseAutosnaps(root["autosnap"]); err != nil {
+		return
+	}
 	return
 }
 
@@ -624,3 +635,63 @@ func parseSnapshotFilter(fm map[string]string) (snapFilter zfs.FilesystemVersion
 	snapFilter = prefixSnapshotFilter{prefix}
 	return
 }
+
+func parseAutosnaps(m interface{}) (snaps []Autosnap, err error) {
+
+	asList := make([]map[string]interface{}, 0)
+	if err = mapstructure.Decode(m, &asList); err != nil {
+		return
+	}
+
+	snaps = make([]Autosnap, len(asList))
+
+	for i, e := range asList {
+		if snaps[i], err = parseAutosnap(e); err != nil {
+			err = fmt.Errorf("cannot parse autonsap job #%d: %s", i+1, err)
+			return
+		}
+	}
+
+	return
+
+}
+
+func parseAutosnap(m map[string]interface{}) (a Autosnap, err error) {
+
+	var i struct {
+		Name          string
+		Prefix        string
+		Interval      string
+		DatasetFilter map[string]string `mapstructure:"dataset_filter"`
+	}
+
+	if err = mapstructure.Decode(m, &i); err != nil {
+		err = fmt.Errorf("structure unfit: %s", err)
+		return
+	}
+
+	a.Name = i.Name
+
+	if len(i.Prefix) < 1 {
+		err = fmt.Errorf("prefix must not be empty")
+		return
+	}
+	a.Prefix = i.Prefix
+
+	var interval time.Duration
+	if interval, err = time.ParseDuration(i.Interval); err != nil {
+		err = fmt.Errorf("cannot parse interval: %s", err)
+		return
+	}
+	a.Interval = &jobrun.PeriodicRepeatStrategy{interval}
+
+	if len(i.DatasetFilter) == 0 {
+		err = fmt.Errorf("dataset_filter not specified")
+		return
+	}
+	if a.DatasetFilter, err = parseComboMapping(i.DatasetFilter); err != nil {
+		err = fmt.Errorf("cannot parse dataset filter: %s", err)
+	}
+
+	return
+}

cmd/main.go

@@ -119,6 +119,13 @@ func main() {
 				cli.BoolFlag{Name: "n", Usage: "simulation (dry run)"},
 			},
 		},
+		{
+			Name:   "autosnap",
+			Action: cmdAutosnap,
+			Flags: []cli.Flag{
+				cli.StringFlag{Name: "job"},
+			},
+		},
 	}
 
 	app.Run(os.Args)
@@ -306,6 +313,7 @@ func cmdPrune(c *cli.Context) error {
 				log.Printf("Prune job failed with error: %s", err)
 			}
 			log.Printf("\n")
+
 		}
 
 	}
@@ -313,6 +321,45 @@ func cmdPrune(c *cli.Context) error {
 	if jobFailed {
 		return cli.NewExitError("At least one job failed with an error. Check log for details.", 1)
 	}
+
+	return nil
+}
+
+func cmdAutosnap(c *cli.Context) error {
+
+	log := defaultLog
+
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		runner.Start()
+	}()
+
+	log.Printf("autosnap...")
+
+	for i := range conf.Autosnaps {
+
+		snap := conf.Autosnaps[i]
+
+		if !c.IsSet("job") || (c.IsSet("job") && c.String("job") == snap.Name) {
+
+			job := jobrun.Job{
+				Name:           fmt.Sprintf("autosnap.%s", snap.Name),
+				RepeatStrategy: snap.Interval,
+				RunFunc: func(log jobrun.Logger) error {
+					log.Printf("doing autosnap: %v", snap)
+					ctx := AutosnapContext{snap}
+					return doAutosnap(ctx, log)
+				},
+			}
+			runner.AddJob(job)
+
+		}
+	}
+
+	wg.Wait()
+
 	return nil
 
 }

cmd/sampleconf/zrepl.yml

@@ -97,3 +97,25 @@ prune:
     snapshot_filter: {
       prefix: zrepl_
     }
+
+  - name: hfbak_prune # cleans up after hfbak autosnap job
+    policy: grid
+    grid: 1x1min(keep=all)
+    dataset_filter: {
+      "pool1*": ok
+    }
+    snapshot_filter: {
+      prefix: zrepl_hfbak_
+    }
+
+autosnap:
+
+  - name: hfbak
+    prefix: zrepl_hfbak_
+    interval: 1s
+    dataset_filter: {
+      "pool1*": ok
+    }
+    # prune: hfbak_prune
+    # future versions may inline the retention policy here, but for now,
+    # pruning has to be triggered manually (it's safe to run autosnap + prune in parallel)

zfs/zfs.go

@@ -204,3 +204,26 @@ func ZFSDestroy(dataset string) (err error) {
 	return
 
 }
+
+func ZFSSnapshot(fs DatasetPath, name string, recursive bool) (err error) {
+
+	snapname := fmt.Sprintf("%s@%s", fs.ToString(), name)
+	cmd := exec.Command(ZFS_BINARY, "snapshot", snapname)
+
+	stderr := bytes.NewBuffer(make([]byte, 0, 1024))
+	cmd.Stderr = stderr
+
+	if err = cmd.Start(); err != nil {
+		return err
+	}
+
+	if err = cmd.Wait(); err != nil {
+		err = ZFSError{
+			Stderr:  stderr.Bytes(),
+			WaitErr: err,
+		}
+	}
+
+	return
+
+}