Optionally delete checkpoints instead of archiving

archive.go

@@ -449,7 +449,7 @@ func findFiles(direntries []fs.DirEntry, t int64, findMoreRecentFiles bool) ([]s
 
 // ZIP all checkpoint files older than `from` together and write them to the `archiveDir`,
 // deleting them from the `checkpointsDir`.
-func ArchiveCheckpoints(checkpointsDir, archiveDir string, from int64) (int, error) {
+func ArchiveCheckpoints(checkpointsDir, archiveDir string, from int64, deleteInstead bool) (int, error) {
 	entries1, err := os.ReadDir(checkpointsDir)
 	if err != nil {
 		return 0, err
@@ -469,7 +469,7 @@ func ArchiveCheckpoints(checkpointsDir, archiveDir string, from int64) (int, err
 		go func() {
 			defer wg.Done()
 			for workItem := range work {
-				m, err := archiveCheckpoints(workItem.cdir, workItem.adir, from)
+				m, err := archiveCheckpoints(workItem.cdir, workItem.adir, from, deleteInstead)
 				if err != nil {
 					log.Printf("error while archiving %s/%s: %s", workItem.cluster, workItem.host, err.Error())
 					atomic.AddInt32(&errs, 1)
@@ -509,7 +509,7 @@ func ArchiveCheckpoints(checkpointsDir, archiveDir string, from int64) (int, err
 }
 
 // Helper function for `ArchiveCheckpoints`.
-func archiveCheckpoints(dir string, archiveDir string, from int64) (int, error) {
+func archiveCheckpoints(dir string, archiveDir string, from int64, deleteInstead bool) (int, error) {
 	entries, err := os.ReadDir(dir)
 	if err != nil {
 		return 0, err
@@ -520,6 +520,18 @@ func archiveCheckpoints(dir string, archiveDir string, from int64) (int, error)
 		return 0, err
 	}
 
+	if deleteInstead {
+		n := 0
+		for _, checkpoint := range files {
+			filename := filepath.Join(dir, checkpoint)
+			if err = os.Remove(filename); err != nil {
+				return n, err
+			}
+			n += 1
+		}
+		return n, nil
+	}
+
 	filename := filepath.Join(archiveDir, fmt.Sprintf("%d.zip", from))
 	f, err := os.OpenFile(filename, os.O_CREATE|os.O_WRONLY, 0644)
 	if err != nil && os.IsNotExist(err) {
@@ -538,15 +550,15 @@ func archiveCheckpoints(dir string, archiveDir string, from int64) (int, error)
 	defer zw.Close()
 
 	n := 0
-	for _, jsonFile := range files {
-		filename := filepath.Join(dir, jsonFile)
+	for _, checkpoint := range files {
+		filename := filepath.Join(dir, checkpoint)
 		r, err := os.Open(filename)
 		if err != nil {
 			return n, err
 		}
 		defer r.Close()
 
-		w, err := zw.Create(jsonFile)
+		w, err := zw.Create(checkpoint)
 		if err != nil {
 			return n, err
 		}

cc-metric-store.go

@@ -99,6 +99,7 @@ type Config struct {
 	Archive struct {
 		Interval      string `json:"interval"`
 		RootDir       string `json:"directory"`
+		DeleteInstead bool   `json:"delete-instead"`
 	} `json:"archive"`
 }
 
@@ -200,7 +201,7 @@ func intervals(wg *sync.WaitGroup, ctx context.Context) {
 			case <-ticks:
 				t := time.Now().Add(-d)
 				log.Printf("start archiving checkpoints (older than %s)...\n", t.Format(time.RFC3339))
-				n, err := ArchiveCheckpoints(conf.Checkpoints.RootDir, conf.Archive.RootDir, t.Unix())
+				n, err := ArchiveCheckpoints(conf.Checkpoints.RootDir, conf.Archive.RootDir, t.Unix(), conf.Archive.DeleteInstead)
 				if err != nil {
 					log.Printf("archiving failed: %s\n", err.Error())
 				} else {
@@ -236,14 +237,18 @@ func main() {
 	restoreFrom := startupTime.Add(-d)
 	log.Printf("Loading checkpoints newer than %s\n", restoreFrom.Format(time.RFC3339))
 	files, err := memoryStore.FromCheckpoint(conf.Checkpoints.RootDir, restoreFrom.Unix())
+	loadedData := memoryStore.SizeInBytes() / 1024 / 1024 // In MB
 	if err != nil {
 		log.Fatalf("Loading checkpoints failed: %s\n", err.Error())
 	} else {
-		log.Printf("Checkpoints loaded (%d files, that took %dms)\n", files, time.Since(startupTime).Milliseconds())
+		log.Printf("Checkpoints loaded (%d files, %d MB, that took %dms)\n", files, loadedData, time.Since(startupTime).Milliseconds())
 	}
 
 	runtime.GC()
+	if loadedData > 1000 {
 		debug.SetGCPercent(20)
+	}
+
 	ctx, shutdown := context.WithCancel(context.Background())
 
 	var wg sync.WaitGroup

debug.go

@@ -44,7 +44,7 @@ func (l *level) debugDump(w *bufio.Writer, m *MemoryStore, indent string) error
 }
 
 func (m *MemoryStore) DebugDump(w *bufio.Writer) error {
-	fmt.Fprintf(w, "MemoryStore:\n")
+	fmt.Fprintf(w, "MemoryStore (%d MB):\n", m.SizeInBytes()/1024/1024)
 	m.root.debugDump(w, m, "  ")
 	return w.Flush()
 }

memstore.go

@@ -3,6 +3,7 @@ package main
 import (
 	"errors"
 	"sync"
+	"unsafe"
 )
 
 // Default buffer capacity.
@@ -231,6 +232,14 @@ func (b *buffer) iterFromTo(from, to int64, callback func(b *buffer) error) erro
 	return nil
 }
 
+func (b *buffer) count() int64 {
+	res := int64(len(b.data))
+	if b.prev != nil {
+		res += b.prev.count()
+	}
+	return res
+}
+
 // Could also be called "node" as this forms a node in a tree structure.
 // Called level because "node" might be confusing here.
 // Can be both a leaf or a inner node. In this tree structue, inner nodes can
@@ -320,6 +329,24 @@ func (l *level) free(t int64) (int, error) {
 	return n, nil
 }
 
+func (l *level) sizeInBytes() int64 {
+	l.lock.RLock()
+	defer l.lock.RUnlock()
+	size := int64(0)
+
+	for _, b := range l.metrics {
+		if b != nil {
+			size += b.count() * int64(unsafe.Sizeof(Float(0)))
+		}
+	}
+
+	for _, child := range l.children {
+		size += child.sizeInBytes()
+	}
+
+	return size
+}
+
 type MemoryStore struct {
 	root    level // root of the tree structure
 	metrics map[string]MetricConfig
@@ -477,6 +504,10 @@ func (m *MemoryStore) FreeAll() error {
 	return nil
 }
 
+func (m *MemoryStore) SizeInBytes() int64 {
+	return m.root.sizeInBytes()
+}
+
 // Given a selector, return a list of all children of the level selected.
 func (m *MemoryStore) ListChildren(selector []string) []string {
 	lvl := &m.root