lib/logger: prevent from blocking when log output isn't consumed in timely manner

Drop log messages instead of blocking and increment `vm_log_messages_dropped_total` metric.

README.md

@@ -22,12 +22,13 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
 
 * [Wix.com](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#wixcom)
 * [Wedos.com](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#wedoscom)
+* [Dreamteam](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#dreamteam)
 
 
 ## Prominent features
 
 * Supports [Prometheus querying API](https://prometheus.io/docs/prometheus/latest/querying/api/), so it can be used as Prometheus drop-in replacement in Grafana.
-  Additionally, VictoriaMetrics extends PromQL with opt-in [useful features](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL).
+  VictoriaMetrics implements [MetricsQL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL) query language, which is inspired by PromQL.
 * Supports global query view. Multiple Prometheus instances may write data into VictoriaMetrics. Later this data may be used in a single query.
 * High performance and good scalability for both [inserts](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b)
   and [selects](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4).
@@ -404,6 +405,18 @@ VictoriaMetrics supports the following handlers from [Prometheus querying API](h
 
 These handlers can be queried from Prometheus-compatible clients such as Grafana or curl.
 
+VictoriaMetrics accepts additional args for `/api/v1/labels` and `/api/v1/label/.../values` handlers.
+See [this feature request](https://github.com/prometheus/prometheus/issues/6178) for details:
+
+* Any number [time series selectors](https://prometheus.io/docs/prometheus/latest/querying/basics/#time-series-selectors) via `match[]` query arg.
+* Optional `start` and `end` query args for limiting the time range for the selected labels or label values.
+
+Additionally VictoriaMetrics provides the following handlers:
+
+* `/api/v1/series/count` - it returns the total number of time series in the database. Note that this handler scans all the inverted index,
+  so it can be slow if the database contains tens of millions of time series.
+* `/api/v1/labels/count` - it returns a list of `label: values_count` entries. It can be used for determining labels with the maximum number of values.
+
 
 ### How to build from sources
 
@@ -519,7 +532,8 @@ before actually deleting the metrics.
 
 Send a request to `http://<victoriametrics-addr>:8428/api/v1/export?match[]=<timeseries_selector_for_export>`,
 where `<timeseries_selector_for_export>` may contain any [time series selector](https://prometheus.io/docs/prometheus/latest/querying/basics/#time-series-selectors)
-for metrics to export. The response would contain all the data for the selected time series in [JSON streaming format](https://en.wikipedia.org/wiki/JSON_streaming#Line-delimited_JSON).
+for metrics to export. Use `{__name__!=""}` selector for fetching all the time series.
+The response would contain all the data for the selected time series in [JSON streaming format](https://en.wikipedia.org/wiki/JSON_streaming#Line-delimited_JSON).
 Each JSON line would contain data for a single time series. An example output:
 
 ```
@@ -720,7 +734,7 @@ mkfs.ext4 ... -O 64bit,huge_file,extent -T huge
 
 VictoriaMetrics exports internal metrics in Prometheus format on the `/metrics` page.
 Add this page to Prometheus' scrape config in order to collect VictoriaMetrics metrics.
-There is [an official Grafana dashboard for single-node VictoriaMetrics](https://grafana.com/dashboards/10229).
+There are officials Grafana dashboards for [single-node VictoriaMetrics](https://grafana.com/dashboards/10229) and [clustered VictoriaMetrics](https://grafana.com/grafana/dashboards/11176).
 
 The most interesting metrics are:
 

app/victoria-metrics/main.go

@@ -21,7 +21,7 @@ func main() {
 	flag.Parse()
 	buildinfo.Init()
 	logger.Init()
-	logger.Infof("starting VictoraMetrics at %q...", *httpListenAddr)
+	logger.Infof("starting VictoriaMetrics at %q...", *httpListenAddr)
 	startTime := time.Now()
 	vmstorage.Init()
 	vmselect.Init()

app/vmbackup/README.md

@@ -6,7 +6,7 @@ Supported storage systems for backups:
 
 * [GCS](https://cloud.google.com/storage/). Example: `gcs://<bucket>/<path/to/backup>`
 * [S3](https://aws.amazon.com/s3/). Example: `s3://<bucket>/<path/to/backup>`
-* Any S3-compatible storage such as [MinIO](https://github.com/minio/minio). See `-customS3Endpoint` command-line flag.
+* Any S3-compatible storage such as [MinIO](https://github.com/minio/minio), [Ceph](https://docs.ceph.com/docs/mimic/radosgw/s3/) or [Swift](https://www.swiftstack.com/docs/admin/middleware/s3_middleware.html). See `-customS3Endpoint` command-line flag.
 * Local filesystem. Example: `fs://</absolute/path/to/backup>`
 
 Incremental backups and full backups are supported. Incremental backups are created automatically if the destination path already contains data from the previous backup.

app/vminsert/graphite/server.go

@@ -21,36 +21,62 @@ var (
 	writeErrorsUDP   = metrics.NewCounter(`vm_graphite_request_errors_total{name="write", net="udp"}`)
 )
 
-// Serve starts graphite server on the given addr.
-func Serve(addr string) {
+// Server accepts Graphite plaintext lines over TCP and UDP.
+type Server struct {
+	addr  string
+	lnTCP net.Listener
+	lnUDP net.PacketConn
+	wg    sync.WaitGroup
+}
+
+// MustStart starts graphite server on the given addr.
+//
+// MustStop must be called on the returned server when it is no longer needed.
+func MustStart(addr string) *Server {
 	logger.Infof("starting TCP Graphite server at %q", addr)
 	lnTCP, err := netutil.NewTCPListener("graphite", addr)
 	if err != nil {
 		logger.Fatalf("cannot start TCP Graphite server at %q: %s", addr, err)
 	}
-	listenerTCP = lnTCP
 
 	logger.Infof("starting UDP Graphite server at %q", addr)
 	lnUDP, err := net.ListenPacket("udp4", addr)
 	if err != nil {
 		logger.Fatalf("cannot start UDP Graphite server at %q: %s", addr, err)
 	}
-	listenerUDP = lnUDP
 
-	var wg sync.WaitGroup
-	wg.Add(1)
+	s := &Server{
+		addr:  addr,
+		lnTCP: lnTCP,
+		lnUDP: lnUDP,
+	}
+	s.wg.Add(1)
 	go func() {
-		defer wg.Done()
-		serveTCP(listenerTCP)
+		defer s.wg.Done()
+		serveTCP(lnTCP)
 		logger.Infof("stopped TCP Graphite server at %q", addr)
 	}()
-	wg.Add(1)
+	s.wg.Add(1)
 	go func() {
-		defer wg.Done()
-		serveUDP(listenerUDP)
+		defer s.wg.Done()
+		serveUDP(lnUDP)
 		logger.Infof("stopped UDP Graphite server at %q", addr)
 	}()
-	wg.Wait()
+	return s
+}
+
+// MustStop stops the server.
+func (s *Server) MustStop() {
+	logger.Infof("stopping TCP Graphite server at %q...", s.addr)
+	if err := s.lnTCP.Close(); err != nil {
+		logger.Errorf("cannot close TCP Graphite server: %s", err)
+	}
+	logger.Infof("stopping UDP Graphite server at %q...", s.addr)
+	if err := s.lnUDP.Close(); err != nil {
+		logger.Errorf("cannot close UDP Graphite server: %s", err)
+	}
+	s.wg.Wait()
+	logger.Infof("TCP and UDP Graphite servers at %q have been stopped", s.addr)
 }
 
 func serveTCP(ln net.Listener) {
@@ -59,6 +85,7 @@ func serveTCP(ln net.Listener) {
 		if err != nil {
 			if ne, ok := err.(net.Error); ok {
 				if ne.Temporary() {
+					logger.Errorf("graphite: temporary error when listening for TCP addr %q: %s", ln.Addr(), err)
 					time.Sleep(time.Second)
 					continue
 				}
@@ -97,6 +124,7 @@ func serveUDP(ln net.PacketConn) {
 					writeErrorsUDP.Inc()
 					if ne, ok := err.(net.Error); ok {
 						if ne.Temporary() {
+							logger.Errorf("graphite: temporary error when listening for UDP addr %q: %s", ln.LocalAddr(), err)
 							time.Sleep(time.Second)
 							continue
 						}
@@ -119,20 +147,3 @@ func serveUDP(ln net.PacketConn) {
 	}
 	wg.Wait()
 }
-
-var (
-	listenerTCP net.Listener
-	listenerUDP net.PacketConn
-)
-
-// Stop stops the server.
-func Stop() {
-	logger.Infof("stopping TCP Graphite server at %q...", listenerTCP.Addr())
-	if err := listenerTCP.Close(); err != nil {
-		logger.Errorf("cannot close TCP Graphite server: %s", err)
-	}
-	logger.Infof("stopping UDP Graphite server at %q...", listenerUDP.LocalAddr())
-	if err := listenerUDP.Close(); err != nil {
-		logger.Errorf("cannot close UDP Graphite server: %s", err)
-	}
-}

app/vminsert/main.go

@@ -19,39 +19,47 @@ import (
 )
 
 var (
-	graphiteListenAddr     = flag.String("graphiteListenAddr", "", "TCP and UDP address to listen for Graphite plaintext data. Usually :2003 must be set. Doesn't work if empty")
-	opentsdbListenAddr     = flag.String("opentsdbListenAddr", "", "TCP and UDP address to listen for OpentTSDB put messages. Usually :4242 must be set. Doesn't work if empty")
+	graphiteListenAddr = flag.String("graphiteListenAddr", "", "TCP and UDP address to listen for Graphite plaintext data. Usually :2003 must be set. Doesn't work if empty")
+	opentsdbListenAddr = flag.String("opentsdbListenAddr", "", "TCP and UDP address to listen for OpentTSDB metrics. "+
+		"Telnet put messages and HTTP /api/put messages are simultaneously served on TCP port. "+
+		"Usually :4242 must be set. Doesn't work if empty")
 	opentsdbHTTPListenAddr = flag.String("opentsdbHTTPListenAddr", "", "TCP address to listen for OpentTSDB HTTP put requests. Usually :4242 must be set. Doesn't work if empty")
 	maxInsertRequestSize   = flag.Int("maxInsertRequestSize", 32*1024*1024, "The maximum size of a single insert request in bytes")
 	maxLabelsPerTimeseries = flag.Int("maxLabelsPerTimeseries", 30, "The maximum number of labels accepted per time series. Superflouos labels are dropped")
 )
 
+var (
+	graphiteServer     *graphite.Server
+	opentsdbServer     *opentsdb.Server
+	opentsdbhttpServer *opentsdbhttp.Server
+)
+
 // Init initializes vminsert.
 func Init() {
 	storage.SetMaxLabelsPerTimeseries(*maxLabelsPerTimeseries)
 
 	concurrencylimiter.Init()
 	if len(*graphiteListenAddr) > 0 {
-		go graphite.Serve(*graphiteListenAddr)
+		graphiteServer = graphite.MustStart(*graphiteListenAddr)
 	}
 	if len(*opentsdbListenAddr) > 0 {
-		go opentsdb.Serve(*opentsdbListenAddr)
+		opentsdbServer = opentsdb.MustStart(*opentsdbListenAddr, int64(*maxInsertRequestSize))
 	}
 	if len(*opentsdbHTTPListenAddr) > 0 {
-		go opentsdbhttp.Serve(*opentsdbHTTPListenAddr, int64(*maxInsertRequestSize))
+		opentsdbhttpServer = opentsdbhttp.MustStart(*opentsdbHTTPListenAddr, int64(*maxInsertRequestSize))
 	}
 }
 
 // Stop stops vminsert.
 func Stop() {
 	if len(*graphiteListenAddr) > 0 {
-		graphite.Stop()
+		graphiteServer.MustStop()
 	}
 	if len(*opentsdbListenAddr) > 0 {
-		opentsdb.Stop()
+		opentsdbServer.MustStop()
 	}
 	if len(*opentsdbHTTPListenAddr) > 0 {
-		opentsdbhttp.Stop()
+		opentsdbhttpServer.MustStop()
 	}
 }
 
@@ -75,6 +83,7 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 			httpserver.Errorf(w, "error in %q: %s", r.URL.Path, err)
 			return true
 		}
+		w.WriteHeader(http.StatusNoContent)
 		return true
 	case "/write", "/api/v2/write":
 		influxWriteRequests.Inc()

app/vminsert/opentsdb/listener_switch.go

@@ -0,0 +1,159 @@
+package opentsdb
+
+import (
+	"io"
+	"net"
+	"sync"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+// listenerSwitch listens for incoming connections and multiplexes them to OpenTSDB http or telnet listeners
+// depending on the first byte in the accepted connection.
+//
+// It is expected that both listeners - http and telnet consume incoming connections as soon as possible.
+type listenerSwitch struct {
+	ln net.Listener
+	wg sync.WaitGroup
+
+	telnetConnsCh chan net.Conn
+	httpConnsCh   chan net.Conn
+
+	closeLock sync.Mutex
+	closed    bool
+	acceptErr error
+	closeErr  error
+}
+
+func newListenerSwitch(ln net.Listener) *listenerSwitch {
+	ls := &listenerSwitch{
+		ln: ln,
+	}
+	ls.telnetConnsCh = make(chan net.Conn)
+	ls.httpConnsCh = make(chan net.Conn)
+	ls.wg.Add(1)
+	go func() {
+		ls.worker()
+		close(ls.telnetConnsCh)
+		close(ls.httpConnsCh)
+		ls.wg.Done()
+	}()
+	return ls
+}
+
+func (ls *listenerSwitch) stop() error {
+	var err error
+	ls.closeLock.Lock()
+	if !ls.closed {
+		err = ls.ln.Close()
+		ls.closeErr = err
+		ls.closed = true
+	}
+	ls.closeLock.Unlock()
+
+	if err == nil {
+		// Wait until worker detects the closed ls.ln and exits.
+		ls.wg.Wait()
+	}
+	return err
+}
+
+func (ls *listenerSwitch) worker() {
+	var buf [1]byte
+	for {
+		c, err := ls.ln.Accept()
+		if err != nil {
+			if ne, ok := err.(net.Error); ok && ne.Temporary() {
+				logger.Infof("listenerSwitch: temporary error at %q: %s; sleeping for a second...", ls.ln.Addr(), err)
+				time.Sleep(time.Second)
+				continue
+			}
+			ls.closeLock.Lock()
+			ls.acceptErr = err
+			ls.closeLock.Unlock()
+			return
+		}
+		if _, err := io.ReadFull(c, buf[:]); err != nil {
+			logger.Errorf("listenerSwitch: cannot read one byte from the underlying connection for %q: %s", ls.ln.Addr(), err)
+			_ = c.Close()
+			continue
+		}
+
+		// It is expected that both listeners - http and telnet consume incoming connections as soon as possible,
+		// so the below code shouldn't block for extended periods of time.
+		pc := &peekedConn{
+			Conn:      c,
+			firstChar: buf[0],
+		}
+		if buf[0] == 'p' {
+			// Assume the request starts with `put`.
+			ls.telnetConnsCh <- pc
+		} else {
+			// Assume the request starts with `POST`.
+			ls.httpConnsCh <- pc
+		}
+	}
+}
+
+type peekedConn struct {
+	net.Conn
+	firstChar     byte
+	firstCharRead bool
+}
+
+func (pc *peekedConn) Read(p []byte) (int, error) {
+	// It is assumed that the pc cannot be read from concurrent goroutines.
+	if pc.firstCharRead {
+		// Fast path - first char already read.
+		return pc.Conn.Read(p)
+	}
+
+	// Slow path - read the first char.
+	if len(p) == 0 {
+		return 0, nil
+	}
+	p[0] = pc.firstChar
+	pc.firstCharRead = true
+	n, err := pc.Conn.Read(p[1:])
+	return n + 1, err
+}
+
+func (ls *listenerSwitch) newTelnetListener() *chanListener {
+	return &chanListener{
+		ls: ls,
+		ch: ls.telnetConnsCh,
+	}
+}
+
+func (ls *listenerSwitch) newHTTPListener() *chanListener {
+	return &chanListener{
+		ls: ls,
+		ch: ls.httpConnsCh,
+	}
+}
+
+type chanListener struct {
+	ls *listenerSwitch
+	ch chan net.Conn
+}
+
+func (cl *chanListener) Accept() (net.Conn, error) {
+	c, ok := <-cl.ch
+	if ok {
+		return c, nil
+	}
+
+	cl.ls.closeLock.Lock()
+	err := cl.ls.acceptErr
+	cl.ls.closeLock.Unlock()
+	return nil, err
+}
+
+func (cl *chanListener) Close() error {
+	return cl.ls.stop()
+}
+
+func (cl *chanListener) Addr() net.Addr {
+	return cl.ls.ln.Addr()
+}

app/vminsert/opentsdb/server.go

@@ -7,6 +7,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vminsert/opentsdbhttp"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/netutil"
@@ -21,44 +22,91 @@ var (
 	writeErrorsUDP   = metrics.NewCounter(`vm_opentsdb_request_errors_total{name="write", net="udp"}`)
 )
 
-// Serve starts OpenTSDB collector on the given addr.
-func Serve(addr string) {
+// Server is a server for collecting OpenTSDB TCP and UDP metrics.
+//
+// It accepts simultaneously Telnet put requests and HTTP put requests over TCP.
+type Server struct {
+	addr       string
+	ls         *listenerSwitch
+	httpServer *opentsdbhttp.Server
+	lnUDP      net.PacketConn
+	wg         sync.WaitGroup
+}
+
+// MustStart starts OpenTSDB collector on the given addr.
+//
+// MustStop must be called on the returned server when it is no longer needed.
+func MustStart(addr string, maxRequestSize int64) *Server {
 	logger.Infof("starting TCP OpenTSDB collector at %q", addr)
 	lnTCP, err := netutil.NewTCPListener("opentsdb", addr)
 	if err != nil {
 		logger.Fatalf("cannot start TCP OpenTSDB collector at %q: %s", addr, err)
 	}
-	listenerTCP = lnTCP
+	ls := newListenerSwitch(lnTCP)
+	lnHTTP := ls.newHTTPListener()
+	lnTelnet := ls.newTelnetListener()
+	httpServer := opentsdbhttp.MustServe(lnHTTP, maxRequestSize)
 
 	logger.Infof("starting UDP OpenTSDB collector at %q", addr)
 	lnUDP, err := net.ListenPacket("udp4", addr)
 	if err != nil {
 		logger.Fatalf("cannot start UDP OpenTSDB collector at %q: %s", addr, err)
 	}
-	listenerUDP = lnUDP
 
-	var wg sync.WaitGroup
-	wg.Add(1)
+	s := &Server{
+		addr:       addr,
+		ls:         ls,
+		httpServer: httpServer,
+		lnUDP:      lnUDP,
+	}
+	s.wg.Add(1)
 	go func() {
-		defer wg.Done()
-		serveTCP(listenerTCP)
-		logger.Infof("stopped TCP OpenTSDB collector at %q", addr)
+		defer s.wg.Done()
+		serveTelnet(lnTelnet)
+		logger.Infof("stopped TCP telnet OpenTSDB server at %q", addr)
 	}()
-	wg.Add(1)
+	s.wg.Add(1)
 	go func() {
-		defer wg.Done()
-		serveUDP(listenerUDP)
-		logger.Infof("stopped UDP OpenTSDB collector at %q", addr)
+		defer s.wg.Done()
+		httpServer.Wait()
+		// Do not log when httpServer is stopped, since this is logged by the server itself.
 	}()
-	wg.Wait()
+	s.wg.Add(1)
+	go func() {
+		defer s.wg.Done()
+		serveUDP(lnUDP)
+		logger.Infof("stopped UDP OpenTSDB server at %q", addr)
+	}()
+	return s
 }
 
-func serveTCP(ln net.Listener) {
+// MustStop stops the server.
+func (s *Server) MustStop() {
+	// Stop HTTP server. Do not emit log message, since it is emitted by the httpServer.
+	s.httpServer.MustStop()
+
+	logger.Infof("stopping TCP telnet OpenTSDB server at %q...", s.addr)
+	if err := s.ls.stop(); err != nil {
+		logger.Errorf("cannot stop TCP telnet OpenTSDB server: %s", err)
+	}
+
+	logger.Infof("stopping UDP OpenTSDB server at %q...", s.addr)
+	if err := s.lnUDP.Close(); err != nil {
+		logger.Errorf("cannot stop UDP OpenTSDB server: %s", err)
+	}
+
+	// Wait until all the servers are stopped.
+	s.wg.Wait()
+	logger.Infof("TCP and UDP OpenTSDB servers at %q have been stopped", s.addr)
+}
+
+func serveTelnet(ln net.Listener) {
 	for {
 		c, err := ln.Accept()
 		if err != nil {
 			if ne, ok := err.(net.Error); ok {
 				if ne.Temporary() {
+					logger.Errorf("opentsdb: temporary error when listening for TCP addr %q: %s", ln.Addr(), err)
 					time.Sleep(time.Second)
 					continue
 				}
@@ -97,6 +145,7 @@ func serveUDP(ln net.PacketConn) {
 					writeErrorsUDP.Inc()
 					if ne, ok := err.(net.Error); ok {
 						if ne.Temporary() {
+							logger.Errorf("opentsdb: temporary error when listening for UDP addr %q: %s", ln.LocalAddr(), err)
 							time.Sleep(time.Second)
 							continue
 						}
@@ -119,20 +168,3 @@ func serveUDP(ln net.PacketConn) {
 	}
 	wg.Wait()
 }
-
-var (
-	listenerTCP net.Listener
-	listenerUDP net.PacketConn
-)
-
-// Stop stops the server.
-func Stop() {
-	logger.Infof("stopping TCP OpenTSDB server at %q...", listenerTCP.Addr())
-	if err := listenerTCP.Close(); err != nil {
-		logger.Errorf("cannot close TCP OpenTSDB server: %s", err)
-	}
-	logger.Infof("stopping UDP OpenTSDB server at %q...", listenerUDP.LocalAddr())
-	if err := listenerUDP.Close(); err != nil {
-		logger.Errorf("cannot close UDP OpenTSDB server: %s", err)
-	}
-}

app/vminsert/opentsdbhttp/server.go

@@ -2,11 +2,14 @@ package opentsdbhttp
 
 import (
 	"context"
+	"net"
 	"net/http"
+	"sync"
 	"time"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/httpserver"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/netutil"
 	"github.com/VictoriaMetrics/metrics"
 )
 
@@ -15,56 +18,84 @@ var (
 	writeErrors   = metrics.NewCounter(`vm_http_request_errors_total{path="/api/put", protocol="opentsdb-http"}`)
 )
 
-var (
-	httpServer     *http.Server
-	httpAddr       string
-	maxRequestSize int64
-)
+// Server represents HTTP OpenTSDB server.
+type Server struct {
+	s  *http.Server
+	ln net.Listener
+	wg sync.WaitGroup
+}
 
-// Serve starts HTTP OpenTSDB server on the given addr.
-func Serve(addr string, maxReqSize int64) {
+// MustStart starts HTTP OpenTSDB server on the given addr.
+//
+// MustStop must be called on the returned server when it is no longer needed.
+func MustStart(addr string, maxRequestSize int64) *Server {
 	logger.Infof("starting HTTP OpenTSDB server at %q", addr)
-	httpAddr = addr
-	maxRequestSize = maxReqSize
-	httpServer = &http.Server{
-		Addr:         addr,
-		Handler:      http.HandlerFunc(requestHandler),
+	lnTCP, err := netutil.NewTCPListener("opentsdbhttp", addr)
+	if err != nil {
+		logger.Fatalf("cannot start HTTP OpenTSDB collector at %q: %s", addr, err)
+	}
+	return MustServe(lnTCP, maxRequestSize)
+}
+
+// MustServe serves OpenTSDB HTTP put requests from ln with up to maxRequestSize size.
+//
+// MustStop must be called on the returned server when it is no longer needed.
+func MustServe(ln net.Listener, maxRequestSize int64) *Server {
+	h := newRequestHandler(maxRequestSize)
+	hs := &http.Server{
+		Handler:      h,
 		ReadTimeout:  30 * time.Second,
 		WriteTimeout: 10 * time.Second,
 	}
+	s := &Server{
+		s:  hs,
+		ln: ln,
+	}
+	s.wg.Add(1)
 	go func() {
-		err := httpServer.ListenAndServe()
+		defer s.wg.Done()
+		err := s.s.Serve(s.ln)
 		if err == http.ErrServerClosed {
 			return
 		}
 		if err != nil {
-			logger.Fatalf("error serving HTTP OpenTSDB: %s", err)
+			logger.Fatalf("error serving HTTP OpenTSDB at %q: %s", s.ln.Addr(), err)
 		}
 	}()
+	return s
 }
 
-// requestHandler handles HTTP OpenTSDB insert request.
-func requestHandler(w http.ResponseWriter, r *http.Request) {
-	switch r.URL.Path {
-	case "/api/put":
-		writeRequests.Inc()
-		if err := insertHandler(r, maxRequestSize); err != nil {
-			writeErrors.Inc()
-			httpserver.Errorf(w, "error in %q: %s", r.URL.Path, err)
-			return
-		}
-		w.WriteHeader(http.StatusNoContent)
-	default:
-		httpserver.Errorf(w, "unexpected path requested on HTTP OpenTSDB server: %q", r.URL.Path)
-	}
+// Wait waits until the server is stopped with MustStop.
+func (s *Server) Wait() {
+	s.wg.Wait()
 }
 
-// Stop stops HTTP OpenTSDB server.
-func Stop() {
-	logger.Infof("stopping HTTP OpenTSDB server at %q...", httpAddr)
+// MustStop stops HTTP OpenTSDB server.
+func (s *Server) MustStop() {
+	logger.Infof("stopping HTTP OpenTSDB server at %q...", s.ln.Addr())
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancel()
-	if err := httpServer.Shutdown(ctx); err != nil {
-		logger.Fatalf("cannot close HTTP OpenTSDB server: %s", err)
+	if err := s.s.Shutdown(ctx); err != nil {
+		logger.Fatalf("cannot close HTTP OpenTSDB server at %q: %s", s.ln.Addr(), err)
 	}
+	s.wg.Wait()
+	logger.Infof("OpenTSDB HTTP server at %q has been stopped", s.ln.Addr())
+}
+
+func newRequestHandler(maxRequestSize int64) http.Handler {
+	rh := func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/api/put":
+			writeRequests.Inc()
+			if err := insertHandler(r, maxRequestSize); err != nil {
+				writeErrors.Inc()
+				httpserver.Errorf(w, "error in %q: %s", r.URL.Path, err)
+				return
+			}
+			w.WriteHeader(http.StatusNoContent)
+		default:
+			httpserver.Errorf(w, "unexpected path requested on HTTP OpenTSDB server: %q", r.URL.Path)
+		}
+	}
+	return http.HandlerFunc(rh)
 }

app/vmselect/netstorage/netstorage.go

@@ -432,13 +432,10 @@ func GetLabelEntries(deadline Deadline) ([]storage.TagEntry, error) {
 	// Sort labelEntries by the number of label values in each entry.
 	sort.Slice(labelEntries, func(i, j int) bool {
 		a, b := labelEntries[i].Values, labelEntries[j].Values
-		if len(a) < len(b) {
-			return true
+		if len(a) != len(b) {
+			return len(a) > len(b)
 		}
-		if len(a) > len(b) {
-			return false
-		}
-		return labelEntries[i].Key < labelEntries[j].Key
+		return labelEntries[i].Key > labelEntries[j].Key
 	})
 
 	return labelEntries, nil

app/vmselect/prometheus/prometheus.go

@@ -23,9 +23,10 @@ import (
 var (
 	latencyOffset = flag.Duration("search.latencyOffset", time.Second*30, "The time when data points become visible in query results after the colection. "+
 		"Too small value can result in incomplete last points for query results")
-	maxQueryDuration = flag.Duration("search.maxQueryDuration", time.Second*30, "The maximum time for search query execution")
-	maxQueryLen      = flag.Int("search.maxQueryLen", 16*1024, "The maximum search query length in bytes")
-	maxLookback      = flag.Duration("search.maxLookback", 0, "Synonim to `-search.lookback-delta` from Prometheus. "+
+	maxExportDuration = flag.Duration("search.maxExportDuration", 10*time.Minute, "The maximum duration for `/api/v1/export` call")
+	maxQueryDuration  = flag.Duration("search.maxQueryDuration", time.Second*30, "The maximum duration for search query execution")
+	maxQueryLen       = flag.Int("search.maxQueryLen", 16*1024, "The maximum search query length in bytes")
+	maxLookback       = flag.Duration("search.maxLookback", 0, "Synonim to `-search.lookback-delta` from Prometheus. "+
 		"The value is dynamically detected from interval between time series datapoints if not set. It can be overridden on per-query basis via `max_lookback` arg")
 )
 
@@ -58,7 +59,7 @@ func FederateHandler(w http.ResponseWriter, r *http.Request) error {
 	if err != nil {
 		return err
 	}
-	deadline := getDeadline(r)
+	deadline := getDeadlineForQuery(r)
 	if start >= end {
 		start = end - defaultStep
 	}
@@ -129,7 +130,7 @@ func ExportHandler(w http.ResponseWriter, r *http.Request) error {
 		return err
 	}
 	format := r.FormValue("format")
-	deadline := getDeadline(r)
+	deadline := getDeadlineForExport(r)
 	if start >= end {
 		end = start + defaultStep
 	}
@@ -235,7 +236,7 @@ var deleteDuration = metrics.NewSummary(`vm_request_duration_seconds{path="/api/
 // See https://prometheus.io/docs/prometheus/latest/querying/api/#querying-label-values
 func LabelValuesHandler(labelName string, w http.ResponseWriter, r *http.Request) error {
 	startTime := time.Now()
-	deadline := getDeadline(r)
+	deadline := getDeadlineForQuery(r)
 
 	if err := r.ParseForm(); err != nil {
 		return fmt.Errorf("cannot parse form values: %s", err)
@@ -285,6 +286,13 @@ func labelValuesWithMatches(labelName string, matches []string, start, end int64
 	if err != nil {
 		return nil, err
 	}
+	for i, tfs := range tagFilterss {
+		// Add `labelName!=''` tag filter in order to filter out series without the labelName.
+		tagFilterss[i] = append(tfs, storage.TagFilter{
+			Key:        []byte(labelName),
+			IsNegative: true,
+		})
+	}
 	if start >= end {
 		end = start + defaultStep
 	}
@@ -326,7 +334,7 @@ var labelValuesDuration = metrics.NewSummary(`vm_request_duration_seconds{path="
 // LabelsCountHandler processes /api/v1/labels/count request.
 func LabelsCountHandler(w http.ResponseWriter, r *http.Request) error {
 	startTime := time.Now()
-	deadline := getDeadline(r)
+	deadline := getDeadlineForQuery(r)
 	labelEntries, err := netstorage.GetLabelEntries(deadline)
 	if err != nil {
 		return fmt.Errorf(`cannot obtain label entries: %s`, err)
@@ -345,10 +353,38 @@ var labelsCountDuration = metrics.NewSummary(`vm_request_duration_seconds{path="
 // See https://prometheus.io/docs/prometheus/latest/querying/api/#getting-label-names
 func LabelsHandler(w http.ResponseWriter, r *http.Request) error {
 	startTime := time.Now()
-	deadline := getDeadline(r)
-	labels, err := netstorage.GetLabels(deadline)
-	if err != nil {
-		return fmt.Errorf("cannot obtain labels: %s", err)
+	deadline := getDeadlineForQuery(r)
+
+	if err := r.ParseForm(); err != nil {
+		return fmt.Errorf("cannot parse form values: %s", err)
+	}
+	var labels []string
+	if len(r.Form["match[]"]) == 0 && len(r.Form["start"]) == 0 && len(r.Form["end"]) == 0 {
+		var err error
+		labels, err = netstorage.GetLabels(deadline)
+		if err != nil {
+			return fmt.Errorf("cannot obtain labels: %s", err)
+		}
+	} else {
+		// Extended functionality that allows filtering by label filters and time range
+		// i.e. /api/v1/labels?match[]=foobar{baz="abc"}&start=...&end=...
+		matches := r.Form["match[]"]
+		if len(matches) == 0 {
+			matches = []string{"{__name__!=''}"}
+		}
+		ct := currentTime()
+		end, err := getTime(r, "end", ct)
+		if err != nil {
+			return err
+		}
+		start, err := getTime(r, "start", end-defaultStep)
+		if err != nil {
+			return err
+		}
+		labels, err = labelsWithMatches(matches, start, end, deadline)
+		if err != nil {
+			return fmt.Errorf("cannot obtain labels for match[]=%q, start=%d, end=%d: %s", matches, start, end, err)
+		}
 	}
 
 	w.Header().Set("Content-Type", "application/json")
@@ -357,12 +393,57 @@ func LabelsHandler(w http.ResponseWriter, r *http.Request) error {
 	return nil
 }
 
+func labelsWithMatches(matches []string, start, end int64, deadline netstorage.Deadline) ([]string, error) {
+	if len(matches) == 0 {
+		logger.Panicf("BUG: matches must be non-empty")
+	}
+	tagFilterss, err := getTagFilterssFromMatches(matches)
+	if err != nil {
+		return nil, err
+	}
+	if start >= end {
+		end = start + defaultStep
+	}
+	sq := &storage.SearchQuery{
+		MinTimestamp: start,
+		MaxTimestamp: end,
+		TagFilterss:  tagFilterss,
+	}
+	rss, err := netstorage.ProcessSearchQuery(sq, false, deadline)
+	if err != nil {
+		return nil, fmt.Errorf("cannot fetch data for %q: %s", sq, err)
+	}
+
+	m := make(map[string]struct{})
+	var mLock sync.Mutex
+	err = rss.RunParallel(func(rs *netstorage.Result, workerID uint) {
+		mLock.Lock()
+		tags := rs.MetricName.Tags
+		for i := range tags {
+			t := &tags[i]
+			m[string(t.Key)] = struct{}{}
+		}
+		m["__name__"] = struct{}{}
+		mLock.Unlock()
+	})
+	if err != nil {
+		return nil, fmt.Errorf("error when data fetching: %s", err)
+	}
+
+	labels := make([]string, 0, len(m))
+	for label := range m {
+		labels = append(labels, label)
+	}
+	sort.Strings(labels)
+	return labels, nil
+}
+
 var labelsDuration = metrics.NewSummary(`vm_request_duration_seconds{path="/api/v1/labels"}`)
 
 // SeriesCountHandler processes /api/v1/series/count request.
 func SeriesCountHandler(w http.ResponseWriter, r *http.Request) error {
 	startTime := time.Now()
-	deadline := getDeadline(r)
+	deadline := getDeadlineForQuery(r)
 	n, err := netstorage.GetSeriesCount(deadline)
 	if err != nil {
 		return fmt.Errorf("cannot obtain series count: %s", err)
@@ -402,7 +483,7 @@ func SeriesHandler(w http.ResponseWriter, r *http.Request) error {
 	if err != nil {
 		return err
 	}
-	deadline := getDeadline(r)
+	deadline := getDeadlineForQuery(r)
 
 	tagFilterss, err := getTagFilterssFromMatches(matches)
 	if err != nil {
@@ -471,7 +552,7 @@ func QueryHandler(w http.ResponseWriter, r *http.Request) error {
 	if err != nil {
 		return err
 	}
-	deadline := getDeadline(r)
+	deadline := getDeadlineForQuery(r)
 	lookbackDelta, err := getMaxLookback(r)
 	if err != nil {
 		return err
@@ -594,7 +675,7 @@ func QueryRangeHandler(w http.ResponseWriter, r *http.Request) error {
 }
 
 func queryRangeHandler(w http.ResponseWriter, query string, start, end, step int64, r *http.Request, ct int64) error {
-	deadline := getDeadline(r)
+	deadline := getDeadlineForQuery(r)
 	mayCache := !getBool(r, "nocache")
 	lookbackDelta, err := getMaxLookback(r)
 	if err != nil {
@@ -786,12 +867,21 @@ func getMaxLookback(r *http.Request) (int64, error) {
 	return getDuration(r, "max_lookback", d)
 }
 
-func getDeadline(r *http.Request) netstorage.Deadline {
+func getDeadlineForQuery(r *http.Request) netstorage.Deadline {
+	dMax := int64(maxQueryDuration.Seconds() * 1e3)
+	return getDeadlineWithMaxDuration(r, dMax)
+}
+
+func getDeadlineForExport(r *http.Request) netstorage.Deadline {
+	dMax := int64(maxExportDuration.Seconds() * 1e3)
+	return getDeadlineWithMaxDuration(r, dMax)
+}
+
+func getDeadlineWithMaxDuration(r *http.Request, dMax int64) netstorage.Deadline {
 	d, err := getDuration(r, "timeout", 0)
 	if err != nil {
 		d = 0
 	}
-	dMax := int64(maxQueryDuration.Seconds() * 1e3)
 	if d <= 0 || d > dMax {
 		d = dMax
 	}

app/vmselect/promql/aggr.go

@@ -27,7 +27,7 @@ var aggrFuncs = map[string]aggrFunc{
 	"topk":         newAggrFuncTopK(false),
 	"quantile":     aggrFuncQuantile,
 
-	// Extended PromQL funcs
+	// PromQL extension funcs
 	"median":         aggrFuncMedian,
 	"limitk":         aggrFuncLimitK,
 	"distinct":       newAggrFunc(aggrFuncDistinct),

docs/CaseStudies.md

@@ -18,6 +18,7 @@ Numbers:
 * The average time series churn rate is ~3M per day.
 * The average query rate is ~1K per minute (mostly alert queries).
 * Query duration: median is ~70ms, 99th percentile is ~2sec.
+* Retention: 6 months.
 
 > Alternatives that we’ve played with before choosing VictoriaMetrics are: federated Prometheus, Cortex, IronDB and Thanos.
 > Points that were critical to us when we were choosing a central tsdb, in order of importance:
@@ -29,7 +30,7 @@ Numbers:
 * Enough head room/scaling capacity for future growth, up to 100M active time series.
 * Ability to split DB replicas per workload. Alert queries go to one replica, user queries go to another (speed for users, effective cache).
 
-> Optimizing for those points and our specific workload VictoriaMetrics proved to be the best option. As an icing on a cake we’ve got an extended PromQL - `default 0` and `histogram` are my favorite ones, for example. What we specially like is having a lot of tsdb params easily available via config options, that makes tsdb easy to tune for specific use case. Also worth noting is a great community in [Slack channel](http://slack.victoriametrics.com/) and of course maintainer support.
+> Optimizing for those points and our specific workload VictoriaMetrics proved to be the best option. As an icing on a cake we’ve got [PromQL extensions](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL) - `default 0` and `histogram` are my favorite ones, for example. What we specially like is having a lot of tsdb params easily available via config options, that makes tsdb easy to tune for specific use case. Also worth noting is a great community in [Slack channel](http://slack.victoriametrics.com/) and of course maintainer support.
 
 Alex Ulstein, Head of Monitoring, Wix.com
 
@@ -45,3 +46,19 @@ Numbers:
 * Query duration: median is ~2ms, 99th percentile is ~50ms.
 
 > We like configuration simplicity and zero maintenance for VictoriaMetrics - once installed and forgot about it. It works out of the box without any issues.
+
+
+### Dreamteam
+
+[Dreamteam](https://dreamteam.gg/) successfully uses single-node VictoriaMetrics in multiple environments.
+
+Numbers:
+
+* Active time series: from 350K to 725K.
+* Total number of time series: from 100M to 320M.
+* Total number of datapoints: from 120 billions to 155 billions.
+* Retention: 3 months.
+
+VictoriaMetrics in production environment runs on 2 M5 EC2 instances in "HA" mode, managed by Terraform and Ansible TF module.
+2 Prometheus instances are writing to both VMs, with 2 [Promxy](https://github.com/jacksontj/promxy) replicas
+as load balancer for reads.

docs/ExtendedPromQL.md

@@ -1,9 +1,24 @@
-# Extended PromQL
+# MetricsQL
 
-VictoriaMetrics supports [standard PromQL](https://prometheus.io/docs/prometheus/latest/querying/basics/)
-including [subqueries](https://prometheus.io/blog/2019/01/28/subquery-support/).
-Additionally it supports useful extensions mentioned below.
-Try these extensions on [an editable Grafana dashboard](http://play-grafana.victoriametrics.com:3000/d/4ome8yJmz/node-exporter-on-victoriametrics-demo).
+VictoriaMetrics implements MetricsQL - query language inspired by [PromQL](https://prometheus.io/docs/prometheus/latest/querying/basics/).
+It is backwards compatible with PromQL, so Grafana dashboards backed by Prometheus datasource should work the same after switching from Prometheus to VictoriaMetrics.
+
+The following functionality is implemented differently in MetricsQL comparing to PromQL in order to improve user experience:
+* MetricsQL takes into account the previous point before the window in square brackets for range functions such as `rate` and `increase`.
+  It also doesn't extrapolate range function results. This addresses [this issue from Prometheus](https://github.com/prometheus/prometheus/issues/3746).
+* MetricsQL returns the expected non-empty responses for requests with `step` values smaller than scrape interval. This addresses [this issue from Grafana](https://github.com/grafana/grafana/issues/11451).
+* MetricsQL treats `scalar` type the same as `instant vector` without labels, since subtle difference between these types usually confuses users.
+  See [the corresponding Prometheus docs](https://prometheus.io/docs/prometheus/latest/querying/basics/#expression-language-data-types) for details.
+
+Other PromQL functionality should work the same in MetricsQL. [File an issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues)
+if you notice discrepancies between PromQL and MetricsQL results other than mentioned above.
+
+MetricsQL provides additional functionality mentioned below, which is aimed towards solving practical cases.
+Feel free [filing a feature request](https://github.com/VictoriaMetrics/VictoriaMetrics/issues) if you think MetricsQL misses certain useful functionality.
+
+*Note that the functionality mentioned below doesn't work in PromQL, so it is impossible switching back to Prometheus after you start using it.*
+
+This functionality can be tried at [an editable Grafana dashboard](http://play-grafana.victoriametrics.com:3000/d/4ome8yJmz/node-exporter-on-victoriametrics-demo).
 
 - [`WITH` templates](https://play.victoriametrics.com/promql/expand-with-exprs). This feature simplifies writing and managing complex queries. Go to [`WITH` templates playground](https://victoriametrics.com/promql/expand-with-exprs) and try it.
 - Metric names and metric labels may contain escaped chars. For instance, `foo\-bar{baz\=aa="b"}` is valid expression. It returns time series with name `foo-bar` containing label `baz=aa` with value `b`. Additionally, `\xXX` escape sequence is supported, where `XX` is hexadecimal representation of escaped char.

docs/Home.md

@@ -3,9 +3,11 @@
 * [Quick start](Quick-Start)
 * [`WITH` templates playground](https://play.victoriametrics.com/promql/expand-with-exprs)
 * [Grafana playground](http://play-grafana.victoriametrics.com:3000/d/4ome8yJmz/node-exporter-on-victoriametrics-demo)
-* [Extended PromQL](ExtendedPromQL)
+* [MetricsQL](ExtendedPromQL)
 * [Single-node version](Single-server-VictoriaMetrics)
 * [FAQ](FAQ)
 * [Cluster version](Cluster-VictoriaMetrics)
 * [Articles](Articles)
-
+* [Case Studies](CaseStudies)
+* [vmbackup](vmbackup)
+* [vmrestore](vmrestore)

docs/Single-server-VictoriaMetrics.md

@@ -12,12 +12,13 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
 
 * [Wix.com](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#wixcom)
 * [Wedos.com](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#wedoscom)
+* [Dreamteam](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#dreamteam)
 
 
 ## Prominent features
 
 * Supports [Prometheus querying API](https://prometheus.io/docs/prometheus/latest/querying/api/), so it can be used as Prometheus drop-in replacement in Grafana.
-  Additionally, VictoriaMetrics extends PromQL with opt-in [useful features](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL).
+  VictoriaMetrics implements [MetricsQL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL) query language, which is inspired by PromQL.
 * Supports global query view. Multiple Prometheus instances may write data into VictoriaMetrics. Later this data may be used in a single query.
 * High performance and good scalability for both [inserts](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b)
   and [selects](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4).
@@ -394,6 +395,18 @@ VictoriaMetrics supports the following handlers from [Prometheus querying API](h
 
 These handlers can be queried from Prometheus-compatible clients such as Grafana or curl.
 
+VictoriaMetrics accepts additional args for `/api/v1/labels` and `/api/v1/label/.../values` handlers.
+See [this feature request](https://github.com/prometheus/prometheus/issues/6178) for details:
+
+* Any number [time series selectors](https://prometheus.io/docs/prometheus/latest/querying/basics/#time-series-selectors) via `match[]` query arg.
+* Optional `start` and `end` query args for limiting the time range for the selected labels or label values.
+
+Additionally VictoriaMetrics provides the following handlers:
+
+* `/api/v1/series/count` - it returns the total number of time series in the database. Note that this handler scans all the inverted index,
+  so it can be slow if the database contains tens of millions of time series.
+* `/api/v1/labels/count` - it returns a list of `label: values_count` entries. It can be used for determining labels with the maximum number of values.
+
 
 ### How to build from sources
 
@@ -509,7 +522,8 @@ before actually deleting the metrics.
 
 Send a request to `http://<victoriametrics-addr>:8428/api/v1/export?match[]=<timeseries_selector_for_export>`,
 where `<timeseries_selector_for_export>` may contain any [time series selector](https://prometheus.io/docs/prometheus/latest/querying/basics/#time-series-selectors)
-for metrics to export. The response would contain all the data for the selected time series in [JSON streaming format](https://en.wikipedia.org/wiki/JSON_streaming#Line-delimited_JSON).
+for metrics to export. Use `{__name__!=""}` selector for fetching all the time series.
+The response would contain all the data for the selected time series in [JSON streaming format](https://en.wikipedia.org/wiki/JSON_streaming#Line-delimited_JSON).
 Each JSON line would contain data for a single time series. An example output:
 
 ```
@@ -710,7 +724,7 @@ mkfs.ext4 ... -O 64bit,huge_file,extent -T huge
 
 VictoriaMetrics exports internal metrics in Prometheus format on the `/metrics` page.
 Add this page to Prometheus' scrape config in order to collect VictoriaMetrics metrics.
-There is [an official Grafana dashboard for single-node VictoriaMetrics](https://grafana.com/dashboards/10229).
+There are officials Grafana dashboards for [single-node VictoriaMetrics](https://grafana.com/dashboards/10229) and [clustered VictoriaMetrics](https://grafana.com/grafana/dashboards/11176).
 
 The most interesting metrics are:
 

docs/vmbackup.md

@@ -0,0 +1,181 @@
+## vmbackup
+
+`vmbackup` creates VictoriaMetrics data backups from [instant snapshots](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots).
+
+Supported storage systems for backups:
+
+* [GCS](https://cloud.google.com/storage/). Example: `gcs://<bucket>/<path/to/backup>`
+* [S3](https://aws.amazon.com/s3/). Example: `s3://<bucket>/<path/to/backup>`
+* Any S3-compatible storage such as [MinIO](https://github.com/minio/minio), [Ceph](https://docs.ceph.com/docs/mimic/radosgw/s3/) or [Swift](https://www.swiftstack.com/docs/admin/middleware/s3_middleware.html). See `-customS3Endpoint` command-line flag.
+* Local filesystem. Example: `fs://</absolute/path/to/backup>`
+
+Incremental backups and full backups are supported. Incremental backups are created automatically if the destination path already contains data from the previous backup.
+Full backups can be sped up with `-origin` pointing to already existing backup on the same remote storage. In this case `vmbackup` makes server-side copy for the shared
+data between the existing backup and new backup. This saves time and costs on data transfer.
+
+Backup process can be interrupted at any time. It is automatically resumed from the interruption point when restarting `vmbackup` with the same args.
+
+Backed up data can be restored with [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md).
+
+See [this article](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883) for more details.
+
+
+### Use cases
+
+#### Regular backups
+
+Regular backup can be performed with the following command:
+
+```
+vmbackup -storageDataPath=</path/to/victoria-metrics-data> -snapshotName=<local-snapshot> -dst=gcs://<bucket>/<path/to/new/backup>
+```
+
+* `</path/to/victoria-metrics-data>` - path to VictoriaMetrics data pointed by `-storageDataPath` command-line flag in single-node VictoriaMetrics or in cluster `vmstorage`.
+  There is no need to stop VictoriaMetrics for creating backups, since they are performed from immutable [instant snapshots](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots).
+* `<local-snapshot>` is the snapshot to backup. See [how to create instant snapshots](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots).
+* `<bucket>` is already existing name for [GCS bucket](https://cloud.google.com/storage/docs/creating-buckets).
+* `<path/to/new/backup>` is the destination path where new backup will be placed.
+
+
+#### Regular backups with server-side copy from existing backup
+
+If the destination GCS bucket already contains the previous backup at `-origin` path, then new backup can be sped up
+with the following command:
+
+```
+vmbackup -storageDataPath=</path/to/victoria-metrics-data> -snapshotName=<local-snapshot> -dst=gcs://<bucket>/<path/to/new/backup> -origin=gcs://<bucket>/<path/to/existing/backup>
+```
+
+This saves time and network bandwidth costs by performing server-side copy for the shared data from the `-origin` to `-dst`.
+
+
+#### Incremental backups
+
+Incremental backups are performed if `-dst` points to already existing backup. In this case only new data is uploaded to remote storage.
+This saves time and network bandwidth costs when working with big backups:
+
+```
+vmbackup -storageDataPath=</path/to/victoria-metrics-data> -snapshotName=<local-snapshot> -dst=gcs://<bucket>/<path/to/existing/backup>
+```
+
+
+#### Smart backups
+
+Smart backups mean storing full daily backups into `YYYYMMDD` folders and creating incremental hourly backup into `latest` folder:
+
+* Run the following command every hour:
+
+```
+vmbackup -snapshotName=<latest-snapshot> -dst=gcs://<bucket>/latest
+```
+
+Where `<latest-snapshot>` is the latest [snapshot](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots).
+The command will upload only changed data to `gcs://<bucket>/latest`.
+
+* Run the following command once a day:
+
+```
+vmbackup -snapshotName=<daily-snapshot> -dst=gcs://<bucket>/<YYYYMMDD> -origin=gcs://<bucket>/latest
+```
+
+Where `<daily-snapshot>` is the snapshot for the last day `<YYYYMMDD>`.
+
+
+This apporach saves network bandwidth costs on hourly backups (since they are incremental) and allows recovering data from either the last hour (`latest` backup)
+or from any day (`YYYYMMDD` backups). Note that hourly backup shouldn't run when creating daily backup.
+
+Do not forget removing old snapshots and backups when they are no longer needed for saving storage costs.
+
+
+### How does it work?
+
+The backup algorithm is the following:
+
+1. Collect information about files in the `-snapshotName`, in the `-dst` and in the `-origin`.
+2. Determine files in `-dst`, which are missing in `-snapshotName`, and delete them. These are usually small files, which are already merged into bigger files in the snapshot.
+3. Determine files from `-snapshotName`, which are missing in `-dst`. These are usually small new files and bigger merged files.
+4. Determine files from step 3, which exist in the `-origin`, and perform server-side copy of these files from `-origin` to `-dst`.
+   This are usually the biggest and the oldest files, which are shared between backups.
+5. Upload the remaining files from setp 3 from `-snapshotName` to `-dst`.
+
+The algorithm splits source files into 100MB chunks in the backup. Each chunk is stored as a separate file in the backup.
+Such splitting minimizes the amounts of data to re-transfer after temporary errors.
+
+`vmbackup` relies on [instant snapshot](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282) properties:
+
+- All the files in the snapshot are immutable.
+- Old files are periodically merged into new files.
+- Smaller files have higher probability to be merged.
+- Consecutive snapshots share many identical files.
+
+These properties allow performing fast and cheap incremental backups and server-side copying from `-origin` paths.
+See [this article](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883) for more details.
+`vmbackup` can work improperly or slowly when these properties are violated.
+
+
+### Troubleshooting
+
+* If the backup is slow, then try setting higher value for `-concurrency` flag. This will increase the number of concurrent workers that upload data to backup storage.
+* If `vmbackup` eats all the network bandwidth, then set `-maxBytesPerSecond` to the desired value.
+* If `vmbackup` has been interrupted due to temporary error, then just restart it with the same args. It will resume the backup process.
+
+
+### Advanced usage
+
+Run `vmbackup -help` in order to see all the available options:
+
+```
+  -concurrency int
+    	The number of concurrent workers. Higher concurrency may reduce backup duration (default 10)
+  -configFilePath string
+    	Path to file with S3 configs. Configs are loaded from default location if not set.
+    	See https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html
+  -configProfile string
+    	Profile name for S3 configs (default "default")
+  -credsFilePath string
+    	Path to file with GCS or S3 credentials. Credentials are loaded from default locations if not set.
+    	See https://cloud.google.com/iam/docs/creating-managing-service-account-keys and https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html
+  -customS3Endpoint string
+    	Custom S3 endpoint for use with S3-compatible storages (e.g. MinIO). S3 is used if not set
+  -dst string
+    	Where to put the backup on the remote storage. Example: gcs://bucket/path/to/backup/dir, s3://bucket/path/to/backup/dir or fs:///path/to/local/backup/dir
+    	-dst can point to the previous backup. In this case incremental backup is performed, i.e. only changed data is uploaded
+  -loggerLevel string
+    	Minimum level of errors to log. Possible values: INFO, ERROR, FATAL, PANIC (default "INFO")
+  -maxBytesPerSecond int
+    	The maximum upload speed. There is no limit if it is set to 0
+  -memory.allowedPercent float
+    	Allowed percent of system memory VictoriaMetrics caches may occupy (default 60)
+  -origin string
+    	Optional origin directory on the remote storage with old backup for server-side copying when performing full backup. This speeds up full backups
+  -snapshotName string
+    	Name for the snapshot to backup. See https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots
+  -storageDataPath string
+    	Path to VictoriaMetrics data. Must match -storageDataPath from VictoriaMetrics or vmstorage (default "victoria-metrics-data")
+  -version
+    	Show VictoriaMetrics version
+```
+
+
+### How to build from sources
+
+It is recommended using [binary releases](https://github.com/VictoriaMetrics/VictoriaMetrics/releases) - see `vmutils-*` archives there.
+
+
+#### Development build
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.12.
+2. Run `make vmbackup` from the root folder of the repository.
+   It builds `vmbackup` binary and puts it into the `bin` folder.
+
+#### Production build
+
+1. [Install docker](https://docs.docker.com/install/).
+2. Run `make vmbackup-prod` from the root folder of the repository.
+   It builds `vmbackup-prod` binary and puts it into the `bin` folder.
+
+#### Building docker images
+
+Run `make package-vmbackup`. It builds `victoriametrics/vmbackup:<PKG_TAG>` docker image locally.
+`<PKG_TAG>` is auto-generated image tag, which depends on source code in the repository.
+The `<PKG_TAG>` may be manually set via `PKG_TAG=foobar make package-vmbackup`.

docs/vmrestore.md

@@ -0,0 +1,86 @@
+## vmrestore
+
+`vmrestore` restores data from backups created by [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md).
+VictoriaMetrics `v1.29.0` and newer versions must be used for working with the restored data.
+
+Restore process can be interrupted at any time. It is automatically resumed from the inerruption point
+when restarting `vmrestore` with the same args.
+
+
+### Usage
+
+VictoriaMetrics must be stopped during the restore process.
+
+```
+vmrestore -src=gcs://<bucket>/<path/to/backup> -storageDataPath=<local/path/to/restore>
+
+```
+
+* `<bucket>` is [GCS bucket](https://cloud.google.com/storage/docs/creating-buckets) name.
+* `<path/to/backup>` is the path to backup made with [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md) on GCS bucket.
+* `<local/path/to/restore>` is the path to folder where data will be restored. This folder must be passed
+  to VictoriaMetrics in `-storageDataPath` command-line flag after the restore process is complete.
+
+The original `-storageDataPath` directory may contain old files. They will be susbstituted by the files from backup.
+
+
+### Troubleshooting
+
+* If `vmrestore` eats all the network bandwidth, then set `-maxBytesPerSecond` to the desired value.
+* If `vmrestore` has been interrupted due to temporary error, then just restart it with the same args. It will resume the restore process.
+
+
+### Advanced usage
+
+Run `vmrestore -help` in order to see all the available options:
+
+```
+  -concurrency int
+    	The number of concurrent workers. Higher concurrency may reduce restore duration (default 10)
+  -configFilePath string
+    	Path to file with S3 configs. Configs are loaded from default location if not set.
+    	See https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html
+  -configProfile string
+    	Profile name for S3 configs (default "default")
+  -credsFilePath string
+    	Path to file with GCS or S3 credentials. Credentials are loaded from default locations if not set.
+    	See https://cloud.google.com/iam/docs/creating-managing-service-account-keys and https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html
+  -customS3Endpoint string
+    	Custom S3 endpoint for use with S3-compatible storages (e.g. MinIO). S3 is used if not set
+  -loggerLevel string
+    	Minimum level of errors to log. Possible values: INFO, ERROR, FATAL, PANIC (default "INFO")
+  -maxBytesPerSecond int
+    	The maximum download speed. There is no limit if it is set to 0
+  -memory.allowedPercent float
+    	Allowed percent of system memory VictoriaMetrics caches may occupy (default 60)
+  -src string
+    	Source path with backup on the remote storage. Example: gcs://bucket/path/to/backup/dir, s3://bucket/path/to/backup/dir or fs:///path/to/local/backup/dir
+  -storageDataPath string
+    	Destination path where backup must be restored. VictoriaMetrics must be stopped when restoring from backup. -storageDataPath dir can be non-empty. In this case only missing data is downloaded from backup (default "victoria-metrics-data")
+  -version
+    	Show VictoriaMetrics version
+```
+
+
+### How to build from sources
+
+It is recommended using [binary releases](https://github.com/VictoriaMetrics/VictoriaMetrics/releases) - see `vmutils-*` archives there.
+
+
+#### Development build
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.12.
+2. Run `make vmrestore` from the root folder of the repository.
+   It builds `vmrestore` binary and puts it into the `bin` folder.
+
+#### Production build
+
+1. [Install docker](https://docs.docker.com/install/).
+2. Run `make vmrestore-prod` from the root folder of the repository.
+   It builds `vmrestore-prod` binary and puts it into the `bin` folder.
+
+#### Building docker images
+
+Run `make package-vmrestore`. It builds `victoriametrics/vmrestore:<PKG_TAG>` docker image locally.
+`<PKG_TAG>` is auto-generated image tag, which depends on source code in the repository.
+The `<PKG_TAG>` may be manually set via `PKG_TAG=foobar make package-vmrestore`.

go.mod

@@ -4,8 +4,8 @@ require (
 	cloud.google.com/go v0.49.0 // indirect
 	cloud.google.com/go/storage v1.4.0
 	github.com/VictoriaMetrics/fastcache v1.5.4
-	github.com/VictoriaMetrics/metrics v1.9.2
-	github.com/aws/aws-sdk-go v1.25.48
+	github.com/VictoriaMetrics/metrics v1.9.3
+	github.com/aws/aws-sdk-go v1.26.0
 	github.com/cespare/xxhash/v2 v2.1.1
 	github.com/golang/groupcache v0.0.0-20191027212112-611e8accdfc9 // indirect
 	github.com/golang/snappy v0.0.1
@@ -19,10 +19,10 @@ require (
 	go.opencensus.io v0.22.2 // indirect
 	golang.org/x/exp v0.0.0-20191129062945-2f5052295587 // indirect
 	golang.org/x/lint v0.0.0-20191125180803-fdd1cda4f05f // indirect
-	golang.org/x/net v0.0.0-20191207000613-e7e4b65ae663 // indirect
+	golang.org/x/net v0.0.0-20191209160850-c0dbc17a3553 // indirect
 	golang.org/x/oauth2 v0.0.0-20191202225959-858c2ad4c8b6 // indirect
-	golang.org/x/sys v0.0.0-20191206220618-eeba5f6aabab
-	golang.org/x/tools v0.0.0-20191206204035-259af5ff87bd // indirect
+	golang.org/x/sys v0.0.0-20191210023423-ac6580df4449
+	golang.org/x/tools v0.0.0-20191212051200-825cb0626375 // indirect
 	google.golang.org/api v0.14.0
 	google.golang.org/appengine v1.6.5 // indirect
 	google.golang.org/genproto v0.0.0-20191206224255-0243a4be9c8f // indirect

go.sum

@@ -22,12 +22,12 @@ github.com/BurntSushi/toml v0.3.1/go.mod h1:xHWCNGjB5oqiDr8zfno3MHue2Ht5sIBksp03
 github.com/BurntSushi/xgb v0.0.0-20160522181843-27f122750802/go.mod h1:IVnqGOEym/WlBOVXweHU+Q+/VP0lqqI8lqeDx9IjBqo=
 github.com/VictoriaMetrics/fastcache v1.5.4 h1:0BaXbRH01RycJk79OOBwMCXlNryko9z4yEf6RqbP+Xo=
 github.com/VictoriaMetrics/fastcache v1.5.4/go.mod h1:ptDBkNMQI4RtmVo8VS/XwRY6RoTu1dAWCbrk+6WsEM8=
-github.com/VictoriaMetrics/metrics v1.9.2 h1:+CNV5OOPe1PuHff3AsOa5cbN8qCWkWwZZA7eOmFjOwQ=
-github.com/VictoriaMetrics/metrics v1.9.2/go.mod h1:LU2j9qq7xqZYXz8tF3/RQnB2z2MbZms5TDiIg9/NHiQ=
+github.com/VictoriaMetrics/metrics v1.9.3 h1:+1kZnOIb8RY825Nb9q9yMrPcOYuPE2GrZWxUh59XnHI=
+github.com/VictoriaMetrics/metrics v1.9.3/go.mod h1:LU2j9qq7xqZYXz8tF3/RQnB2z2MbZms5TDiIg9/NHiQ=
 github.com/allegro/bigcache v1.2.1-0.20190218064605-e24eb225f156 h1:eMwmnE/GDgah4HI848JfFxHt+iPb26b4zyfspmqY0/8=
 github.com/allegro/bigcache v1.2.1-0.20190218064605-e24eb225f156/go.mod h1:Cb/ax3seSYIx7SuZdm2G2xzfwmv3TPSk2ucNfQESPXM=
-github.com/aws/aws-sdk-go v1.25.48 h1:J82DYDGZHOKHdhx6hD24Tm30c2C3GchYGfN0mf9iKUk=
-github.com/aws/aws-sdk-go v1.25.48/go.mod h1:KmX6BPdI08NWTb3/sm4ZGu5ShLoqVDhKgpiN924inxo=
+github.com/aws/aws-sdk-go v1.26.0 h1:uddjCeunG1nroOMjcMO57g/vfwY7jJ1zYSEGmdeCeEs=
+github.com/aws/aws-sdk-go v1.26.0/go.mod h1:KmX6BPdI08NWTb3/sm4ZGu5ShLoqVDhKgpiN924inxo=
 github.com/census-instrumentation/opencensus-proto v0.2.1/go.mod h1:f6KPmirojxKA12rnyqOA5BBL4O983OfeGPqjHWSTneU=
 github.com/cespare/xxhash/v2 v2.1.1 h1:6MnRN8NT7+YBpUIWxHtefFZOKTAPgGjpQSxqLNn0+qY=
 github.com/cespare/xxhash/v2 v2.1.1/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
@@ -150,8 +150,8 @@ golang.org/x/net v0.0.0-20190501004415-9ce7a6920f09/go.mod h1:t9HGtf8HONx5eT2rtn
 golang.org/x/net v0.0.0-20190503192946-f4e77d36d62c/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190603091049-60506f45cf65/go.mod h1:HSz+uSET+XFnRR8LxR5pz3Of3rY3CfYBVs4xY44aLks=
 golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
-golang.org/x/net v0.0.0-20191207000613-e7e4b65ae663 h1:Dd5RoEW+yQi+9DMybroBctIdyiwuNT7sJFMC27/6KxI=
-golang.org/x/net v0.0.0-20191207000613-e7e4b65ae663/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/net v0.0.0-20191209160850-c0dbc17a3553 h1:efeOvDhwQ29Dj3SdAV/MJf8oukgn+8D8WgaCaRMchF8=
+golang.org/x/net v0.0.0-20191209160850-c0dbc17a3553/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/oauth2 v0.0.0-20180821212333-d2e6202438be/go.mod h1:N/0e6XlmueqKjAGxoOufVs8QHGRruUQn6yWY3a++T0U=
 golang.org/x/oauth2 v0.0.0-20190226205417-e64efc72b421/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
 golang.org/x/oauth2 v0.0.0-20190604053449-0f29369cfe45 h1:SVwTIAaPC2U/AvvLNZ2a7OVsmBpC8L5BlwK1whH3hm0=
@@ -172,8 +172,8 @@ golang.org/x/sys v0.0.0-20190502145724-3ef323f4f1fd/go.mod h1:h1NjWce9XRLGQEsW7w
 golang.org/x/sys v0.0.0-20190507160741-ecd444e8653b/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20190606165138-5da285871e9c/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20190624142023-c5567b49c5d0/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20191206220618-eeba5f6aabab h1:FvshnhkKW+LO3HWHodML8kuVX8rnJTxKm9dFPuI68UM=
-golang.org/x/sys v0.0.0-20191206220618-eeba5f6aabab/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20191210023423-ac6580df4449 h1:gSbV7h1NRL2G1xTg/owz62CST1oJBmxy4QpMMregXVQ=
+golang.org/x/sys v0.0.0-20191210023423-ac6580df4449/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.1-0.20180807135948-17ff2d5776d2/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2 h1:tW2bmiBqwgJj/UpqtC8EpXEZVYOwU0yG4iWbprSVAcs=
@@ -197,8 +197,8 @@ golang.org/x/tools v0.0.0-20190911174233-4f2ddba30aff/go.mod h1:b+2E5dAYhXwXZwtn
 golang.org/x/tools v0.0.0-20191012152004-8de300cfc20a/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
 golang.org/x/tools v0.0.0-20191115202509-3a792d9c32b2/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
 golang.org/x/tools v0.0.0-20191125144606-a911d9008d1f/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
-golang.org/x/tools v0.0.0-20191206204035-259af5ff87bd h1:Zc7EU2PqpsNeIfOoVA7hvQX4cS3YDJEs5KlfatT3hLo=
-golang.org/x/tools v0.0.0-20191206204035-259af5ff87bd/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.0.0-20191212051200-825cb0626375 h1:hrBCt+pkb1irxFjg/GeadSn24gLxi5/Z861I53OkQbc=
+golang.org/x/tools v0.0.0-20191212051200-825cb0626375/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 google.golang.org/api v0.4.0/go.mod h1:8k5glujaEP+g9n7WNsDg8QP6cUVNI86fCNMcbazEtwE=
 google.golang.org/api v0.7.0/go.mod h1:WtwebWUNSVBH/HAw79HIFXZNqEvBhG+Ra+ax0hx3E3M=

lib/httpserver/httpserver.go

@@ -155,9 +155,6 @@ var metricsHandlerDuration = metrics.NewHistogram(`vm_http_request_duration_seco
 
 func handlerWrapper(w http.ResponseWriter, r *http.Request, rh RequestHandler) {
 	requestsTotal.Inc()
-	if !checkAuth(w, r) {
-		return
-	}
 	switch r.URL.Path {
 	case "/health":
 		w.Header().Set("Content-Type", "text/plain")
@@ -172,25 +169,36 @@ func handlerWrapper(w http.ResponseWriter, r *http.Request, rh RequestHandler) {
 		}
 		w.WriteHeader(status)
 		return
-	case "/metrics":
-		startTime := time.Now()
-		metricsRequests.Inc()
-		w.Header().Set("Content-Type", "text/plain")
-		writePrometheusMetrics(w)
-		metricsHandlerDuration.UpdateDuration(startTime)
-		return
 	case "/favicon.ico":
 		faviconRequests.Inc()
 		w.WriteHeader(http.StatusNoContent)
 		return
+	case "/metrics":
+		metricsRequests.Inc()
+		if len(*metricsAuthKey) > 0 && r.FormValue("authKey") != *metricsAuthKey {
+			http.Error(w, "The provided authKey doesn't match -metricsAuthKey", http.StatusUnauthorized)
+			return
+		}
+		startTime := time.Now()
+		w.Header().Set("Content-Type", "text/plain")
+		writePrometheusMetrics(w)
+		metricsHandlerDuration.UpdateDuration(startTime)
+		return
 	default:
 		if strings.HasPrefix(r.URL.Path, "/debug/pprof/") {
 			pprofRequests.Inc()
+			if len(*pprofAuthKey) > 0 && r.FormValue("authKey") != *pprofAuthKey {
+				http.Error(w, "The provided authKey doesn't match -pprofAuthKey", http.StatusUnauthorized)
+				return
+			}
 			DisableResponseCompression(w)
 			pprofHandler(r.URL.Path[len("/debug/pprof/"):], w, r)
 			return
 		}
 
+		if !checkBasicAuth(w, r) {
+			return
+		}
 		if rh(w, r) {
 			return
 		}
@@ -201,27 +209,6 @@ func handlerWrapper(w http.ResponseWriter, r *http.Request, rh RequestHandler) {
 	}
 }
 
-func checkAuth(w http.ResponseWriter, r *http.Request) bool {
-	path := r.URL.Path
-	if path == "/metrics" && len(*metricsAuthKey) > 0 {
-		authKey := r.FormValue("authKey")
-		if *metricsAuthKey == authKey {
-			return true
-		}
-		http.Error(w, "The provided authKey doesn't match -metricsAuthKey", http.StatusUnauthorized)
-		return false
-	}
-	if strings.HasPrefix(path, "/debug/pprof/") && len(*pprofAuthKey) > 0 {
-		authKey := r.FormValue("authKey")
-		if *pprofAuthKey == authKey {
-			return true
-		}
-		http.Error(w, "The provided authKey doesn't match -pprofAuthKey", http.StatusUnauthorized)
-		return false
-	}
-	return checkBasicAuth(w, r)
-}
-
 func checkBasicAuth(w http.ResponseWriter, r *http.Request) bool {
 	if len(*httpAuthUsername) == 0 {
 		// HTTP Basic Auth is disabled.

lib/logger/logger.go

@@ -8,7 +8,6 @@ import (
 	"os"
 	"runtime"
 	"strings"
-	"sync"
 	"sync/atomic"
 	"time"
 
@@ -16,7 +15,10 @@ import (
 	"github.com/VictoriaMetrics/metrics"
 )
 
-var loggerLevel = flag.String("loggerLevel", "INFO", "Minimum level of errors to log. Possible values: INFO, ERROR, FATAL, PANIC")
+var (
+	loggerLevel  = flag.String("loggerLevel", "INFO", "Minimum level of errors to log. Possible values: INFO, ERROR, FATAL, PANIC")
+	loggerFormat = flag.String("loggerFormat", "default", "Format for logs. Possible values: default, json")
+)
 
 // Init initializes the logger.
 //
@@ -25,7 +27,9 @@ var loggerLevel = flag.String("loggerLevel", "INFO", "Minimum level of errors to
 // There is no need in calling Init from tests.
 func Init() {
 	validateLoggerLevel()
+	validateLoggerFormat()
 	go errorsLoggedCleaner()
+	go logMessageWriter()
 	logAllFlags()
 }
 
@@ -38,6 +42,15 @@ func validateLoggerLevel() {
 	}
 }
 
+func validateLoggerFormat() {
+	switch *loggerFormat {
+	case "default", "json":
+	default:
+		// We cannot use logger.Pancif here, since the logger isn't initialized yet.
+		panic(fmt.Errorf("FATAL: unsupported `-loggerFormat` value: %q; supported values are: default, json", *loggerFormat))
+	}
+}
+
 var stdErrorLogger = log.New(&logWriter{}, "", 0)
 
 // StdErrorLogger returns standard error logger.
@@ -101,7 +114,7 @@ func (lw *logWriter) Write(p []byte) (int, error) {
 }
 
 func logMessage(level, msg string, skipframes int) {
-	timestamp := time.Now().UTC().Format("2006-01-02T15:04:05.000+0000")
+	timestamp := time.Now().UTC().Format("2006-01-02T15:04:05.000Z")
 	levelLowercase := strings.ToLower(level)
 	_, file, line, ok := runtime.Caller(skipframes)
 	if !ok {
@@ -115,12 +128,22 @@ func logMessage(level, msg string, skipframes int) {
 	for len(msg) > 0 && msg[len(msg)-1] == '\n' {
 		msg = msg[:len(msg)-1]
 	}
-	logMsg := fmt.Sprintf("%s\t%s\t%s:%d\t%s\n", timestamp, levelLowercase, file, line, msg)
+	var logMsg string
+	switch *loggerFormat {
+	case "json":
+		caller := fmt.Sprintf("%s:%d", file, line)
+		logMsg = fmt.Sprintf(`{"ts":%q,"level":%q,"caller":%q,"msg":%q}`+"\n", timestamp, levelLowercase, caller, msg)
+	default:
+		logMsg = fmt.Sprintf("%s\t%s\t%s:%d\t%s\n", timestamp, levelLowercase, file, line, msg)
+	}
 
-	// Serialize writes to log.
-	mu.Lock()
-	fmt.Fprint(os.Stderr, logMsg)
-	mu.Unlock()
+	select {
+	case logMessageCh <- logMsg:
+	default:
+		// Writing to log can stuck if the log output isn't consumed in timely manner.
+		// Handle this case via `vm_log_messages_dropped_total`
+		logMessagesDropped.Inc()
+	}
 
 	// Increment vm_log_messages_total
 	location := fmt.Sprintf("%s:%d", file, line)
@@ -135,7 +158,15 @@ func logMessage(level, msg string, skipframes int) {
 	}
 }
 
-var mu sync.Mutex
+var logMessagesDropped = metrics.NewCounter(`vm_log_messages_dropped_total`)
+
+func logMessageWriter() {
+	for msg := range logMessageCh {
+		fmt.Fprint(os.Stderr, msg)
+	}
+}
+
+var logMessageCh = make(chan string, 100)
 
 func shouldSkipLog(level string) bool {
 	switch *loggerLevel {

lib/netutil/tcplistener.go

@@ -6,6 +6,7 @@ import (
 	"net"
 	"time"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/metrics"
 )
 
@@ -72,6 +73,8 @@ func (ln *TCPListener) Accept() (net.Conn, error) {
 		ln.accepts.Inc()
 		if err != nil {
 			if ne, ok := err.(net.Error); ok && ne.Temporary() {
+				logger.Errorf("temporary error when listening for TCP addr %q: %s", ln.Addr(), err)
+				time.Sleep(time.Second)
 				continue
 			}
 			ln.acceptErrors.Inc()

lib/storage/partition.go

@@ -61,6 +61,11 @@ const defaultPartsToMerge = 15
 // write amplification.
 const finalPartsToMerge = 3
 
+// The number of shards for rawRow entries per partition.
+//
+// Higher number of shards reduces CPU contention and increases the max bandwidth on multi-core systems.
+var rawRowsShardsPerPartition = (runtime.GOMAXPROCS(-1) + 7) / 8
+
 // getMaxRowsPerPartition returns the maximum number of rows that haven't been converted into parts yet.
 func getMaxRawRowsPerPartition() int {
 	maxRawRowsPerPartitionOnce.Do(func() {
@@ -71,7 +76,7 @@ func getMaxRawRowsPerPartition() int {
 		if n > 500e3 {
 			n = 500e3
 		}
-		maxRawRowsPerPartition = n
+		maxRawRowsPerPartition = n / rawRowsShardsPerPartition
 	})
 	return maxRawRowsPerPartition
 }
@@ -129,16 +134,10 @@ type partition struct {
 	// Contains file-based parts with big number of items.
 	bigParts []*partWrapper
 
-	// rawRowsLock protects rawRows.
-	rawRowsLock sync.Mutex
-
 	// rawRows contains recently added rows that haven't been converted into parts yet.
 	//
 	// rawRows aren't used in search for performance reasons.
-	rawRows []rawRow
-
-	// rawRowsLastFlushTime is the last time rawRows are flushed.
-	rawRowsLastFlushTime time.Time
+	rawRows rawRowsShards
 
 	snapshotLock sync.RWMutex
 
@@ -264,18 +263,18 @@ func openPartition(smallPartsPath, bigPartsPath string, getDeletedMetricIDs func
 }
 
 func newPartition(name, smallPartsPath, bigPartsPath string, getDeletedMetricIDs func() *uint64set.Set) *partition {
-	return &partition{
+	p := &partition{
 		name:           name,
 		smallPartsPath: smallPartsPath,
 		bigPartsPath:   bigPartsPath,
 
 		getDeletedMetricIDs: getDeletedMetricIDs,
 
-		rawRows: getRawRowsMaxSize().rows,
-
 		mergeIdx: uint64(time.Now().UnixNano()),
 		stopCh:   make(chan struct{}),
 	}
+	p.rawRows.init()
+	return p
 }
 
 // partitionMetrics contains essential metrics for the partition.
@@ -322,10 +321,9 @@ type partitionMetrics struct {
 
 // UpdateMetrics updates m with metrics from pt.
 func (pt *partition) UpdateMetrics(m *partitionMetrics) {
-	pt.rawRowsLock.Lock()
-	m.PendingRows += uint64(len(pt.rawRows))
-	m.SmallRowsCount += uint64(len(pt.rawRows))
-	pt.rawRowsLock.Unlock()
+	rawRowsLen := uint64(pt.rawRows.Len())
+	m.PendingRows += rawRowsLen
+	m.SmallRowsCount += rawRowsLen
 
 	pt.partsLock.Lock()
 
@@ -399,29 +397,82 @@ func (pt *partition) AddRows(rows []rawRow) {
 		}
 	}
 
-	// Try adding rows.
-	var rrs []*rawRows
-	pt.rawRowsLock.Lock()
+	pt.rawRows.addRows(pt, rows)
+}
+
+type rawRowsShards struct {
+	lock     sync.Mutex
+	shardIdx int
+
+	// Shards reduce lock contention when adding rows on multi-CPU systems.
+	shards []rawRowsShard
+}
+
+func (rrs *rawRowsShards) init() {
+	rrs.shards = make([]rawRowsShard, rawRowsShardsPerPartition)
+}
+
+func (rrs *rawRowsShards) addRows(pt *partition, rows []rawRow) {
+	rrs.lock.Lock()
+	rrs.shardIdx++
+	if rrs.shardIdx >= len(rrs.shards) {
+		rrs.shardIdx = 0
+	}
+	shard := &rrs.shards[rrs.shardIdx]
+	rrs.lock.Unlock()
+
+	shard.addRows(pt, rows)
+}
+
+func (rrs *rawRowsShards) Len() int {
+	n := 0
+	for i := range rrs.shards[:] {
+		n += rrs.shards[i].Len()
+	}
+	return n
+}
+
+type rawRowsShard struct {
+	lock          sync.Mutex
+	rows          []rawRow
+	lastFlushTime time.Time
+}
+
+func (rrs *rawRowsShard) Len() int {
+	rrs.lock.Lock()
+	n := len(rrs.rows)
+	rrs.lock.Unlock()
+	return n
+}
+
+func (rrs *rawRowsShard) addRows(pt *partition, rows []rawRow) {
+	var rrss []*rawRows
+
+	rrs.lock.Lock()
+	if cap(rrs.rows) == 0 {
+		rrs.rows = getRawRowsMaxSize().rows
+	}
+	maxRowsCount := getMaxRawRowsPerPartition()
 	for {
-		capacity := cap(pt.rawRows) - len(pt.rawRows)
+		capacity := maxRowsCount - len(rrs.rows)
 		if capacity >= len(rows) {
 			// Fast path - rows fit capacity.
-			pt.rawRows = append(pt.rawRows, rows...)
+			rrs.rows = append(rrs.rows, rows...)
 			break
 		}
 
 		// Slow path - rows don't fit capacity.
 		// Fill rawRows to capacity and convert it to a part.
-		pt.rawRows = append(pt.rawRows, rows[:capacity]...)
+		rrs.rows = append(rrs.rows, rows[:capacity]...)
 		rows = rows[capacity:]
 		rr := getRawRowsMaxSize()
-		pt.rawRows, rr.rows = rr.rows, pt.rawRows
-		rrs = append(rrs, rr)
-		pt.rawRowsLastFlushTime = time.Now()
+		rrs.rows, rr.rows = rr.rows, rrs.rows
+		rrss = append(rrss, rr)
+		rrs.lastFlushTime = time.Now()
 	}
-	pt.rawRowsLock.Unlock()
+	rrs.lock.Unlock()
 
-	for _, rr := range rrs {
+	for _, rr := range rrss {
 		pt.addRowsPart(rr.rows)
 		putRawRows(rr)
 	}
@@ -583,7 +634,7 @@ func (pt *partition) MustClose() {
 	startTime = time.Now()
 
 	// Flush raw rows the last time before exit.
-	pt.flushRawRows(nil, true)
+	pt.flushRawRows(true)
 
 	// Flush inmemory parts to disk.
 	var pws []*partWrapper
@@ -637,7 +688,6 @@ func (pt *partition) startRawRowsFlusher() {
 }
 
 func (pt *partition) rawRowsFlusher() {
-	var rawRows []rawRow
 	t := time.NewTimer(rawRowsFlushInterval)
 	for {
 		select {
@@ -646,29 +696,35 @@ func (pt *partition) rawRowsFlusher() {
 		case <-t.C:
 			t.Reset(rawRowsFlushInterval)
 		}
-
-		rawRows = pt.flushRawRows(rawRows[:0], false)
+		pt.flushRawRows(false)
 	}
 }
 
-func (pt *partition) flushRawRows(newRawRows []rawRow, isFinal bool) []rawRow {
-	oldRawRows := newRawRows[:0]
-	mustFlush := false
+func (pt *partition) flushRawRows(isFinal bool) {
+	pt.rawRows.flush(pt, isFinal)
+}
+
+func (rrs *rawRowsShards) flush(pt *partition, isFinal bool) {
+	for i := range rrs.shards[:] {
+		rrs.shards[i].flush(pt, isFinal)
+	}
+}
+
+func (rrs *rawRowsShard) flush(pt *partition, isFinal bool) {
+	var rr *rawRows
 	currentTime := time.Now()
 
-	pt.rawRowsLock.Lock()
-	if isFinal || currentTime.Sub(pt.rawRowsLastFlushTime) > rawRowsFlushInterval {
-		mustFlush = true
-		oldRawRows = pt.rawRows
-		pt.rawRows = newRawRows[:0]
-		pt.rawRowsLastFlushTime = currentTime
+	rrs.lock.Lock()
+	if isFinal || currentTime.Sub(rrs.lastFlushTime) > rawRowsFlushInterval {
+		rr = getRawRowsMaxSize()
+		rrs.rows, rr.rows = rr.rows, rrs.rows
 	}
-	pt.rawRowsLock.Unlock()
+	rrs.lock.Unlock()
 
-	if mustFlush {
-		pt.addRowsPart(oldRawRows)
+	if rr != nil {
+		pt.addRowsPart(rr.rows)
+		putRawRows(rr)
 	}
-	return oldRawRows
 }
 
 func (pt *partition) startInmemoryPartsFlusher() {
@@ -1334,7 +1390,7 @@ func (pt *partition) CreateSnapshotAt(smallPath, bigPath string) error {
 	startTime := time.Now()
 
 	// Flush inmemory data to disk.
-	pt.flushRawRows(nil, true)
+	pt.flushRawRows(true)
 	if _, err := pt.flushInmemoryParts(nil, true); err != nil {
 		return fmt.Errorf("cannot flush inmemory parts: %s", err)
 	}

lib/storage/partition_search_test.go

@@ -185,7 +185,7 @@ func testPartitionSearchEx(t *testing.T, ptt int64, tr TimeRange, partsCount, ma
 		pt.AddRows(rows)
 
 		// Flush just added rows to a separate partition.
-		pt.flushRawRows(nil, true)
+		pt.flushRawRows(true)
 	}
 	testPartitionSearch(t, pt, tsids, tr, rbsExpected, -1)
 	pt.MustClose()

lib/storage/storage.go

@@ -803,6 +803,11 @@ func (s *Storage) add(rows []rawRow, mrs []MetricRow, precisionBits uint8) ([]ra
 	}
 	rows = rows[:rowsLen+len(mrs)]
 	j := 0
+	var (
+		// These vars are used for speeding up bulk imports of multiple adjancent rows for the same metricName.
+		prevTSID          TSID
+		prevMetricNameRaw []byte
+	)
 	minTimestamp, maxTimestamp := s.tb.getMinMaxTimestamps()
 	// Return only the last error, since it has no sense in returning all errors.
 	var lastWarn error
@@ -830,9 +835,17 @@ func (s *Storage) add(rows []rawRow, mrs []MetricRow, precisionBits uint8) ([]ra
 		r.Timestamp = mr.Timestamp
 		r.Value = mr.Value
 		r.PrecisionBits = precisionBits
+		if string(mr.MetricNameRaw) == string(prevMetricNameRaw) {
+			// Fast path - the current mr contains the same metric name as the previous mr, so it contains the same TSID.
+			// This path should trigger on bulk imports when many rows contain the same MetricNameRaw.
+			r.TSID = prevTSID
+			continue
+		}
 		if s.getTSIDFromCache(&r.TSID, mr.MetricNameRaw) {
 			if !dmis.Has(r.TSID.MetricID) {
 				// Fast path - the TSID for the given MetricName has been found in cache and isn't deleted.
+				prevTSID = r.TSID
+				prevMetricNameRaw = mr.MetricNameRaw
 				continue
 			}
 		}
@@ -890,6 +903,12 @@ func (s *Storage) updatePerDateData(rows []rawRow, lastError error) error {
 	var date uint64
 	var hour uint64
 	var prevTimestamp int64
+	var (
+		// These vars are used for speeding up bulk imports when multiple adjancent rows
+		// contain the same (metricID, date) pairs.
+		prevMatchedDate     uint64
+		prevMatchedMetricID uint64
+	)
 	idb := s.idb()
 	hm := s.currHourMetricIDs.Load().(*hourMetricIDs)
 	for i := range rows {
@@ -913,8 +932,14 @@ func (s *Storage) updatePerDateData(rows []rawRow, lastError error) error {
 		}
 
 		// Slower path: check global cache for (date, metricID) entry.
+		if metricID == prevMatchedMetricID && date == prevMatchedDate {
+			// Fast path for bulk import of multiple rows with the same (date, metricID) pairs.
+			continue
+		}
 		if s.dateMetricIDCache.Has(date, metricID) {
 			// The metricID has been already added to per-day inverted index.
+			prevMatchedDate = date
+			prevMatchedMetricID = metricID
 			continue
 		}
 

lib/storage/table.go

@@ -220,7 +220,7 @@ func (tb *table) flushRawRows() {
 	defer tb.PutPartitions(ptws)
 
 	for _, ptw := range ptws {
-		ptw.pt.flushRawRows(nil, true)
+		ptw.pt.flushRawRows(true)
 	}
 }
 

vendor/github.com/VictoriaMetrics/metrics/histogram.go

@@ -34,7 +34,7 @@ const (
 //     - <counter> - the number of hits to the given bucket during Update* calls
 //
 // Histogram buckets can be converted to Prometheus-like buckets with `le` labels
-// with `prometheus_buckets(<metric_name>_bucket)` function from Extended PromQL in VictoriaMetrics.
+// with `prometheus_buckets(<metric_name>_bucket)` function from PromQL extensions in VictoriaMetrics.
 // (see https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL ):
 //
 //     prometheus_buckets(request_duration_bucket)

vendor/github.com/aws/aws-sdk-go/aws/arn/arn.go

@@ -75,10 +75,11 @@ func Parse(arn string) (ARN, error) {
 	}, nil
 }
 
-// IsARN returns whether the given string is an arn
-// by looking for whether the string starts with arn:
+// IsARN returns whether the given string is an ARN by looking for
+// whether the string starts with "arn:" and contains the correct number
+// of sections delimited by colons(:).
 func IsARN(arn string) bool {
-	return strings.HasPrefix(arn, arnPrefix) && strings.Count(arn, ":") > arnSections-1
+	return strings.HasPrefix(arn, arnPrefix) && strings.Count(arn, ":") >= arnSections-1
 }
 
 // String returns the canonical representation of the ARN

vendor/github.com/aws/aws-sdk-go/aws/credentials/stscreds/assume_role_provider.go

@@ -144,6 +144,13 @@ type AssumeRoleProvider struct {
 	// Session name, if you wish to reuse the credentials elsewhere.
 	RoleSessionName string
 
+	// Optional, you can pass tag key-value pairs to your session. These tags are called session tags.
+	Tags []*sts.Tag
+
+	// A list of keys for session tags that you want to set as transitive.
+	// If you set a tag key as transitive, the corresponding key and value passes to subsequent sessions in a role chain.
+	TransitiveTagKeys []*string
+
 	// Expiry duration of the STS credentials. Defaults to 15 minutes if not set.
 	Duration time.Duration
 
@@ -269,10 +276,12 @@ func (p *AssumeRoleProvider) Retrieve() (credentials.Value, error) {
 	}
 	jitter := time.Duration(sdkrand.SeededRand.Float64() * p.MaxJitterFrac * float64(p.Duration))
 	input := &sts.AssumeRoleInput{
-		DurationSeconds: aws.Int64(int64((p.Duration - jitter) / time.Second)),
-		RoleArn:         aws.String(p.RoleARN),
-		RoleSessionName: aws.String(p.RoleSessionName),
-		ExternalId:      p.ExternalID,
+		DurationSeconds:   aws.Int64(int64((p.Duration - jitter) / time.Second)),
+		RoleArn:           aws.String(p.RoleARN),
+		RoleSessionName:   aws.String(p.RoleSessionName),
+		ExternalId:        p.ExternalID,
+		Tags:              p.Tags,
+		TransitiveTagKeys: p.TransitiveTagKeys,
 	}
 	if p.Policy != nil {
 		input.Policy = p.Policy

vendor/github.com/aws/aws-sdk-go/aws/endpoints/defaults.go

@@ -209,12 +209,42 @@ var awsPartition = partition{
 				"eu-west-1":      endpoint{},
 				"eu-west-2":      endpoint{},
 				"eu-west-3":      endpoint{},
-				"me-south-1":     endpoint{},
-				"sa-east-1":      endpoint{},
-				"us-east-1":      endpoint{},
-				"us-east-2":      endpoint{},
-				"us-west-1":      endpoint{},
-				"us-west-2":      endpoint{},
+				"fips-ca-central-1": endpoint{
+					Hostname: "acm-pca-fips.ca-central-1.amazonaws.com",
+					CredentialScope: credentialScope{
+						Region: "ca-central-1",
+					},
+				},
+				"fips-us-east-1": endpoint{
+					Hostname: "acm-pca-fips.us-east-1.amazonaws.com",
+					CredentialScope: credentialScope{
+						Region: "us-east-1",
+					},
+				},
+				"fips-us-east-2": endpoint{
+					Hostname: "acm-pca-fips.us-east-2.amazonaws.com",
+					CredentialScope: credentialScope{
+						Region: "us-east-2",
+					},
+				},
+				"fips-us-west-1": endpoint{
+					Hostname: "acm-pca-fips.us-west-1.amazonaws.com",
+					CredentialScope: credentialScope{
+						Region: "us-west-1",
+					},
+				},
+				"fips-us-west-2": endpoint{
+					Hostname: "acm-pca-fips.us-west-2.amazonaws.com",
+					CredentialScope: credentialScope{
+						Region: "us-west-2",
+					},
+				},
+				"me-south-1": endpoint{},
+				"sa-east-1":  endpoint{},
+				"us-east-1":  endpoint{},
+				"us-east-2":  endpoint{},
+				"us-west-1":  endpoint{},
+				"us-west-2":  endpoint{},
 			},
 		},
 		"api.ecr": service{
@@ -2465,9 +2495,10 @@ var awsPartition = partition{
 				},
 			},
 			Endpoints: endpoints{
-				"eu-west-1": endpoint{},
-				"us-east-1": endpoint{},
-				"us-west-2": endpoint{},
+				"ap-southeast-2": endpoint{},
+				"eu-west-1":      endpoint{},
+				"us-east-1":      endpoint{},
+				"us-west-2":      endpoint{},
 			},
 		},
 		"monitoring": service{
@@ -3078,9 +3109,10 @@ var awsPartition = partition{
 				},
 			},
 			Endpoints: endpoints{
-				"eu-west-1": endpoint{},
-				"us-east-1": endpoint{},
-				"us-west-2": endpoint{},
+				"ap-southeast-2": endpoint{},
+				"eu-west-1":      endpoint{},
+				"us-east-1":      endpoint{},
+				"us-west-2":      endpoint{},
 			},
 		},
 		"runtime.sagemaker": service{

vendor/github.com/aws/aws-sdk-go/aws/request/retryer.go

@@ -83,6 +83,7 @@ var retryableCodes = map[string]struct{}{
 
 var throttleCodes = map[string]struct{}{
 	"ProvisionedThroughputExceededException": {},
+	"ThrottledException":                     {}, // SNS, XRay, ResourceGroupsTagging API
 	"Throttling":                             {},
 	"ThrottlingException":                    {},
 	"RequestLimitExceeded":                   {},

vendor/github.com/aws/aws-sdk-go/aws/version.go

@@ -5,4 +5,4 @@ package aws
 const SDKName = "aws-sdk-go"
 
 // SDKVersion is the version of this SDK
-const SDKVersion = "1.25.48"
+const SDKVersion = "1.26.0"

vendor/github.com/aws/aws-sdk-go/service/s3/service.go

@@ -31,7 +31,7 @@ var initRequest func(*request.Request)
 const (
 	ServiceName = "s3"        // Name of service.
 	EndpointsID = ServiceName // ID to lookup a service endpoint with.
-	ServiceID   = "S3"        // ServiceID is a unique identifer of a specific service.
+	ServiceID   = "S3"        // ServiceID is a unique identifier of a specific service.
 )
 
 // New creates a new instance of the S3 client with a session.

vendor/github.com/aws/aws-sdk-go/service/sts/service.go

@@ -31,7 +31,7 @@ var initRequest func(*request.Request)
 const (
 	ServiceName = "sts"       // Name of service.
 	EndpointsID = ServiceName // ID to lookup a service endpoint with.
-	ServiceID   = "STS"       // ServiceID is a unique identifer of a specific service.
+	ServiceID   = "STS"       // ServiceID is a unique identifier of a specific service.
 )
 
 // New creates a new instance of the STS client with a session.

vendor/golang.org/x/sys/unix/zptrace_arm64_linux.go

@@ -1,53 +0,0 @@
-// Code generated by linux/mkall.go generatePtracePair("arm", "arm64"). DO NOT EDIT.
-
-// +build linux
-// +build arm64
-
-package unix
-
-import "unsafe"
-
-// PtraceRegsArm is the registers used by arm binaries.
-type PtraceRegsArm struct {
-	Uregs [18]uint32
-}
-
-// PtraceGetRegsArm fetches the registers used by arm binaries.
-func PtraceGetRegsArm(pid int, regsout *PtraceRegsArm) error {
-	return ptrace(PTRACE_GETREGS, pid, 0, uintptr(unsafe.Pointer(regsout)))
-}
-
-// PtraceSetRegsArm sets the registers used by arm binaries.
-func PtraceSetRegsArm(pid int, regs *PtraceRegsArm) error {
-	return ptrace(PTRACE_SETREGS, pid, 0, uintptr(unsafe.Pointer(regs)))
-}
-
-// PtraceRegsArm64 is the registers used by arm64 binaries.
-type PtraceRegsArm64 struct {
-	Regs   [31]uint64
-	Sp     uint64
-	Pc     uint64
-	Pstate uint64
-}
-
-// PtraceGetRegsArm64 fetches the registers used by arm64 binaries.
-func PtraceGetRegsArm64(pid int, regsout *PtraceRegsArm64) error {
-	return ptrace(PTRACE_GETREGS, pid, 0, uintptr(unsafe.Pointer(regsout)))
-}
-
-// PtraceSetRegsArm64 sets the registers used by arm64 binaries.
-func PtraceSetRegsArm64(pid int, regs *PtraceRegsArm64) error {
-	return ptrace(PTRACE_SETREGS, pid, 0, uintptr(unsafe.Pointer(regs)))
-}
-
-// PtraceGetRegSetArm64 fetches the registers used by arm64 binaries.
-func PtraceGetRegSetArm64(pid, addr int, regsout *PtraceRegsArm64) error {
-	iovec := Iovec{(*byte)(unsafe.Pointer(regsout)), uint64(unsafe.Sizeof(*regsout))}
-	return ptrace(PTRACE_GETREGSET, pid, uintptr(addr), uintptr(unsafe.Pointer(&iovec)))
-}
-
-// PtraceSetRegSetArm64 sets the registers used by arm64 binaries.
-func PtraceSetRegSetArm64(pid, addr int, regs *PtraceRegsArm64) error {
-	iovec := Iovec{(*byte)(unsafe.Pointer(regs)), uint64(unsafe.Sizeof(*regs))}
-	return ptrace(PTRACE_SETREGSET, pid, uintptr(addr), uintptr(unsafe.Pointer(&iovec)))
-}

vendor/golang.org/x/sys/unix/zptrace_armnn_linux.go

@@ -1,7 +1,7 @@
 // Code generated by linux/mkall.go generatePtracePair("arm", "arm64"). DO NOT EDIT.
 
 // +build linux
-// +build arm
+// +build arm arm64
 
 package unix
 
@@ -39,15 +39,3 @@ func PtraceGetRegsArm64(pid int, regsout *PtraceRegsArm64) error {
 func PtraceSetRegsArm64(pid int, regs *PtraceRegsArm64) error {
 	return ptrace(PTRACE_SETREGS, pid, 0, uintptr(unsafe.Pointer(regs)))
 }
-
-// PtraceGetRegSetArm64 fetches the registers used by arm64 binaries.
-func PtraceGetRegSetArm64(pid, addr int, regsout *PtraceRegsArm64) error {
-	iovec := Iovec{(*byte)(unsafe.Pointer(regsout)), uint32(unsafe.Sizeof(*regsout))}
-	return ptrace(PTRACE_GETREGSET, pid, uintptr(addr), uintptr(unsafe.Pointer(&iovec)))
-}
-
-// PtraceSetRegSetArm64 sets the registers used by arm64 binaries.
-func PtraceSetRegSetArm64(pid, addr int, regs *PtraceRegsArm64) error {
-	iovec := Iovec{(*byte)(unsafe.Pointer(regs)), uint32(unsafe.Sizeof(*regs))}
-	return ptrace(PTRACE_SETREGSET, pid, uintptr(addr), uintptr(unsafe.Pointer(&iovec)))
-}

vendor/golang.org/x/sys/unix/zptrace_linux_arm64.go

@@ -0,0 +1,17 @@
+// Code generated by linux/mkall.go generatePtraceRegSet("arm64"). DO NOT EDIT.
+
+package unix
+
+import "unsafe"
+
+// PtraceGetRegSetArm64 fetches the registers used by arm64 binaries.
+func PtraceGetRegSetArm64(pid, addr int, regsout *PtraceRegsArm64) error {
+	iovec := Iovec{(*byte)(unsafe.Pointer(regsout)), uint64(unsafe.Sizeof(*regsout))}
+	return ptrace(PTRACE_GETREGSET, pid, uintptr(addr), uintptr(unsafe.Pointer(&iovec)))
+}
+
+// PtraceSetRegSetArm64 sets the registers used by arm64 binaries.
+func PtraceSetRegSetArm64(pid, addr int, regs *PtraceRegsArm64) error {
+	iovec := Iovec{(*byte)(unsafe.Pointer(regs)), uint64(unsafe.Sizeof(*regs))}
+	return ptrace(PTRACE_SETREGSET, pid, uintptr(addr), uintptr(unsafe.Pointer(&iovec)))
+}

vendor/golang.org/x/tools/go/packages/golist.go

@@ -810,9 +810,14 @@ func golistDriver(cfg *Config, rootsDirs func() *goInfo, words ...string) (*driv
 		}
 
 		if p.Error != nil {
+			msg := strings.TrimSpace(p.Error.Err) // Trim to work around golang.org/issue/32363.
+			// Address golang.org/issue/35964 by appending import stack to error message.
+			if msg == "import cycle not allowed" && len(p.Error.ImportStack) != 0 {
+				msg += fmt.Sprintf(": import stack: %v", p.Error.ImportStack)
+			}
 			pkg.Errors = append(pkg.Errors, Error{
 				Pos: p.Error.Pos,
-				Msg: strings.TrimSpace(p.Error.Err), // Trim to work around golang.org/issue/32363.
+				Msg: msg,
 			})
 		}
 

vendor/modules.txt

@@ -12,9 +12,9 @@ cloud.google.com/go/storage
 github.com/BurntSushi/toml
 # github.com/VictoriaMetrics/fastcache v1.5.4
 github.com/VictoriaMetrics/fastcache
-# github.com/VictoriaMetrics/metrics v1.9.2
+# github.com/VictoriaMetrics/metrics v1.9.3
 github.com/VictoriaMetrics/metrics
-# github.com/aws/aws-sdk-go v1.25.48
+# github.com/aws/aws-sdk-go v1.26.0
 github.com/aws/aws-sdk-go/aws
 github.com/aws/aws-sdk-go/aws/arn
 github.com/aws/aws-sdk-go/aws/awserr
@@ -124,7 +124,7 @@ golang.org/x/exp/cmd/apidiff
 # golang.org/x/lint v0.0.0-20191125180803-fdd1cda4f05f
 golang.org/x/lint
 golang.org/x/lint/golint
-# golang.org/x/net v0.0.0-20191207000613-e7e4b65ae663
+# golang.org/x/net v0.0.0-20191209160850-c0dbc17a3553
 golang.org/x/net/context
 golang.org/x/net/context/ctxhttp
 golang.org/x/net/http/httpguts
@@ -139,14 +139,14 @@ golang.org/x/oauth2/google
 golang.org/x/oauth2/internal
 golang.org/x/oauth2/jws
 golang.org/x/oauth2/jwt
-# golang.org/x/sys v0.0.0-20191206220618-eeba5f6aabab
+# golang.org/x/sys v0.0.0-20191210023423-ac6580df4449
 golang.org/x/sys/unix
 # golang.org/x/text v0.3.2
 golang.org/x/text/secure/bidirule
 golang.org/x/text/transform
 golang.org/x/text/unicode/bidi
 golang.org/x/text/unicode/norm
-# golang.org/x/tools v0.0.0-20191206204035-259af5ff87bd
+# golang.org/x/tools v0.0.0-20191212051200-825cb0626375
 golang.org/x/tools/cmd/goimports
 golang.org/x/tools/go/analysis
 golang.org/x/tools/go/analysis/passes/inspect