vmcluster: re-routing enhancement (#5293)

* app/vmstorage: close vminsert connections gradually before stopping storage

Implements graceful shutdown approach suggested here - https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4922#issuecomment-1768146878

Test results for this can be found here - https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4922#issuecomment-1790640274

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

* app/vmstorage: update graceful shutdown logic

- close connections from vminsert in determenistic order
- update flag description
- lower default timeout to 25 seconds. 25 seconds value was chosen because the lowest default value used in default configuration deployments is 30s(default value in Kubernetes and ansible-playbooks).

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

* docs/cluster: add information about re-routing enhancement during restart

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

* docs/changelog: add entry for new command-line flag

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

* {app/vmstorage,lib/ingestserver}: address review feedback

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

* docs/cluster: add note to update workload scheduler timeout

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

* wip

---------

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>

docs/CHANGELOG.md

@@ -81,6 +81,7 @@ The sandbox cluster installation is running under the constant load generated by
 * FEATURE: all: track requests with wrong auth key and wrong basic auth at `vm_http_request_errors_total` [metric](https://docs.victoriametrics.com/#monitoring) with `reason="wrong_auth_key"` and `reason="wrong_basic_auth"`. See [this issue](https://github.com/victoriaMetrics/victoriaMetrics/issues/4590). Thanks to @venkatbvc for the [pull request](https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5166).
 * FEATURE: [vmauth](https://docs.victoriametrics.com/vmauth.html): add ability to drop the specified number of `/`-delimited prefix parts from the request path before proxying the request to the matching backend. See [these docs](https://docs.victoriametrics.com/vmauth.html#dropping-request-path-prefix).
 * FEATURE: [vmauth](https://docs.victoriametrics.com/vmauth.html): add ability to skip TLS verification and to specify TLS Root CA when connecting to backends. See [these docs](https://docs.victoriametrics.com/vmauth.html#backend-tls-setup) and [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5240).
+* FEATURE: `vmstorage`: gradually close `vminsert` connections during 25 seconds at [graceful shutdown](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#updating--reconfiguring-cluster-nodes). This should reduce data ingestion slowdown during rolling restarts. The duration for gradual closing of `vminsert` connections can be configured via `-storage.vminsertConnsShutdownDuration` command-line flag. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4922) and [these docs](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#improving-re-routing-performance-during-restart) for details.
 * FEATURE: `vmstorage`: add `-blockcache.missesBeforeCaching` command-line flag, which can be used for fine-tuning RAM usage for `indexdb/dataBlocks` cache when queries touching big number of time series are executed.
 * FEATURE: add `-loggerMaxArgLen` command-line flag for fine-tuning the maximum lengths of logged args.
 

docs/Cluster-VictoriaMetrics.md

@@ -323,6 +323,10 @@ See more details about cardinality limiter in [these docs](https://docs.victoria
 
 ## Troubleshooting
 
+- If your VictoriaMetrics cluster experiences data ingestion delays during
+  [rolling restarts and configuration updates](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#updating--reconfiguring-cluster-nodes),
+  then see [these docs](#improving-re-routing-performance-during-restart).
+
 [Troubleshooting docs for single-node VictoriaMetrics](https://docs.victoriametrics.com/Troubleshooting.html) apply to VictoriaMetrics cluster as well.
 
 ## Readonly mode
@@ -473,6 +477,16 @@ This strategy allows upgrading the cluster without downtime if the following con
 If at least a single condition isn't met, then the rolling restart may result in cluster unavailability
 during the config update / version upgrade. In this case the following strategy is recommended.
 
+#### Improving re-routing performance during restart
+
+`vmstorage` nodes may experience increased usage for CPU, RAM and disk IO during
+[rolling restarts](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#no-downtime-strategy),
+since they need to process higher load when some of `vmstorage` nodes are temporarily unavailable in the cluster.
+It is possible to reduce resource usage spikes by running more `vminsert` nodes and by passing bigger values
+to `-storage.vminsertConnsShutdownDuration` command-line flag at `vmstorage` nodes.
+Make sure that the `-storage.vminsertConnsShutdownDuration` is smaller than the graceful timeout configured at the system which manages `vmstorage`
+(e.g. Docker, Kubernetes, systemd, etc.). Otherwise the system may kill `vmstorage` node before it finishes gradual closing of `vminsert` connections.
+
 ### Minimum downtime strategy
 
 1. Gracefully stop all the `vminsert` and `vmselect` nodes in parallel.
@@ -866,6 +880,8 @@ Below is the output for `/path/to/vminsert -help`:
      Whether to skip verification of TLS certificates provided by -storageNode nodes if -cluster.tls flag is set. Note that disabled TLS certificate verification breaks security. This flag is available only in VictoriaMetrics enterprise. See https://docs.victoriametrics.com/enterprise.html
   -cluster.tlsKeyFile string
      Path to client-side TLS key file to use when connecting to -storageNode if -cluster.tls flag is set. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection . This flag is available only in VictoriaMetrics enterprise. See https://docs.victoriametrics.com/enterprise.html
+  -clusternative.vminsertConnsShutdownDuration duration
+     The time needed for gradual closing of upstream vminsert connections during graceful shutdown. Bigger duration reduces spikes in CPU, RAM and disk IO load on the remaining lower-level clusters during rolling restart. Smaller duration reduces the time needed to close all the upstream vminsert connections, thus reducing the time for graceful shutdown. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#improving-re-routing-performance-during-restart (default 25s)
   -clusternativeListenAddr string
      TCP address to listen for data from other vminsert nodes in multi-level cluster setup. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#multi-level-cluster-setup . Usually :8400 should be set to match default vmstorage port for vminsert. Disabled work if empty
   -csvTrimTimestamp duration
@@ -974,6 +990,8 @@ Below is the output for `/path/to/vminsert -help`:
      Allows renaming fields in JSON formatted logs. Example: "ts:timestamp,msg:message" renames "ts" to "timestamp" and "msg" to "message". Supported fields: ts, level, caller, msg
   -loggerLevel string
      Minimum level of errors to log. Possible values: INFO, WARN, ERROR, FATAL, PANIC (default "INFO")
+  -loggerMaxArgLen int
+     The maximum length of a single logged argument. Longer arguments are replaced with 'arg_start..arg_end', where 'arg_start' and 'arg_end' is prefix and suffix of the arg with the length not exceeding -loggerMaxArgLen / 2 (default 500)
   -loggerOutput string
      Output for the logs. Supported values: stderr, stdout (default "stderr")
   -loggerTimezone string
@@ -1177,6 +1195,8 @@ Below is the output for `/path/to/vmselect -help`:
      Allows renaming fields in JSON formatted logs. Example: "ts:timestamp,msg:message" renames "ts" to "timestamp" and "msg" to "message". Supported fields: ts, level, caller, msg
   -loggerLevel string
      Minimum level of errors to log. Possible values: INFO, WARN, ERROR, FATAL, PANIC (default "INFO")
+  -loggerMaxArgLen int
+     The maximum length of a single logged argument. Longer arguments are replaced with 'arg_start..arg_end', where 'arg_start' and 'arg_end' is prefix and suffix of the arg with the length not exceeding -loggerMaxArgLen / 2 (default 500)
   -loggerOutput string
      Output for the logs. Supported values: stderr, stdout (default "stderr")
   -loggerTimezone string
@@ -1430,6 +1450,8 @@ Below is the output for `/path/to/vmstorage -help`:
      Allows renaming fields in JSON formatted logs. Example: "ts:timestamp,msg:message" renames "ts" to "timestamp" and "msg" to "message". Supported fields: ts, level, caller, msg
   -loggerLevel string
      Minimum level of errors to log. Possible values: INFO, WARN, ERROR, FATAL, PANIC (default "INFO")
+  -loggerMaxArgLen int
+     The maximum length of a single logged argument. Longer arguments are replaced with 'arg_start..arg_end', where 'arg_start' and 'arg_end' is prefix and suffix of the arg with the length not exceeding -loggerMaxArgLen / 2 (default 500)
   -loggerOutput string
      Output for the logs. Supported values: stderr, stdout (default "stderr")
   -loggerTimezone string
@@ -1509,6 +1531,8 @@ Below is the output for `/path/to/vmstorage -help`:
   -storage.minFreeDiskSpaceBytes size
      The minimum free disk space at -storageDataPath after which the storage stops accepting new data
      Supports the following optional suffixes for size values: KB, MB, GB, TB, KiB, MiB, GiB, TiB (default 10000000)
+  -storage.vminsertConnsShutdownDuration duration
+     The time needed for gradual closing of vminsert connections during graceful shutdown. Bigger duration reduces spikes in CPU, RAM and disk IO load on the remaining vmstorage nodes during rolling restart. Smaller duration reduces the time needed to close all the vminsert connections, thus reducing the time for graceful shutdown. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#improving-re-routing-performance-during-restart (default 25s)
   -storageDataPath string
      Path to storage data (default "vmstorage-data")
   -tls

lib/ingestserver/conns_map.go

@@ -2,18 +2,25 @@ package ingestserver
 
 import (
 	"net"
+	"sort"
 	"sync"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 )
 
 // ConnsMap is used for tracking active connections.
 type ConnsMap struct {
+	clientName string
+
 	mu       sync.Mutex
 	m        map[net.Conn]struct{}
 	isClosed bool
 }
 
 // Init initializes cm.
-func (cm *ConnsMap) Init() {
+func (cm *ConnsMap) Init(clientName string) {
+	cm.clientName = clientName
 	cm.m = make(map[net.Conn]struct{})
 	cm.isClosed = false
 }
@@ -36,12 +43,51 @@ func (cm *ConnsMap) Delete(c net.Conn) {
 	cm.mu.Unlock()
 }
 
-// CloseAll closes all the added conns.
-func (cm *ConnsMap) CloseAll() {
+// CloseAll gradually closes all the cm conns with during the given shutdownDuration.
+func (cm *ConnsMap) CloseAll(shutdownDuration time.Duration) {
 	cm.mu.Lock()
+	conns := make([]net.Conn, len(cm.m))
 	for c := range cm.m {
-		_ = c.Close()
+		conns = append(conns, c)
+		delete(cm.m, c)
 	}
 	cm.isClosed = true
 	cm.mu.Unlock()
+
+	if shutdownDuration <= 0 {
+		// Close all the connections at once.
+		for _, c := range conns {
+			_ = c.Close()
+		}
+		return
+	}
+	if len(conns) == 0 {
+		return
+	}
+	if len(conns) == 1 {
+		// Simple case - just close a single connection and that's it!
+		_ = conns[0].Close()
+		return
+	}
+
+	// Sort vminsert conns in order to make the order of closing connections deterministic across vmstorage nodes.
+	// This should reduce resource usage spikes at vmstorage nodes during rolling restarts.
+	sort.Slice(conns, func(i, j int) bool {
+		return conns[i].RemoteAddr().String() < conns[j].RemoteAddr().String()
+	})
+
+	shutdownInterval := shutdownDuration / time.Duration(len(conns)-1)
+	startTime := time.Now()
+	logger.Infof("closing %d %s connections with %dms interval between them", len(conns), cm.clientName, shutdownInterval.Milliseconds())
+	remoteAddr := conns[0].RemoteAddr().String()
+	_ = conns[0].Close()
+	logger.Infof("closed %s connection %s", cm.clientName, remoteAddr)
+	conns = conns[1:]
+	for _, c := range conns {
+		time.Sleep(shutdownInterval)
+		remoteAddr := c.RemoteAddr().String()
+		_ = c.Close()
+		logger.Infof("closed %s connection %s", cm.clientName, remoteAddr)
+	}
+	logger.Infof("closed %d %s connections in %s", len(conns), cm.clientName, time.Since(startTime))
 }

lib/ingestserver/graphite/server.go

@@ -59,7 +59,7 @@ func MustStart(addr string, useProxyProtocol bool, insertHandler func(r io.Reade
 		lnTCP: lnTCP,
 		lnUDP: lnUDP,
 	}
-	s.cm.Init()
+	s.cm.Init("graphite")
 	s.wg.Add(1)
 	go func() {
 		defer s.wg.Done()
@@ -85,7 +85,7 @@ func (s *Server) MustStop() {
 	if err := s.lnUDP.Close(); err != nil {
 		logger.Errorf("cannot close UDP Graphite server: %s", err)
 	}
-	s.cm.CloseAll()
+	s.cm.CloseAll(0)
 	s.wg.Wait()
 	logger.Infof("TCP and UDP Graphite servers at %q have been stopped", s.addr)
 }

lib/ingestserver/influx/server.go

@@ -59,7 +59,7 @@ func MustStart(addr string, useProxyProtocol bool, insertHandler func(r io.Reade
 		lnTCP: lnTCP,
 		lnUDP: lnUDP,
 	}
-	s.cm.Init()
+	s.cm.Init("influx")
 	s.wg.Add(1)
 	go func() {
 		defer s.wg.Done()
@@ -85,7 +85,7 @@ func (s *Server) MustStop() {
 	if err := s.lnUDP.Close(); err != nil {
 		logger.Errorf("cannot close UDP InfluxDB server: %s", err)
 	}
-	s.cm.CloseAll()
+	s.cm.CloseAll(0)
 	s.wg.Wait()
 	logger.Infof("TCP and UDP InfluxDB servers at %q have been stopped", s.addr)
 }

lib/ingestserver/opentsdb/server.go

@@ -67,7 +67,7 @@ func MustStart(addr string, useProxyProtocol bool, telnetInsertHandler func(r io
 		httpServer: httpServer,
 		lnUDP:      lnUDP,
 	}
-	s.cm.Init()
+	s.cm.Init("opentsdb")
 	s.wg.Add(1)
 	go func() {
 		defer s.wg.Done()
@@ -103,7 +103,7 @@ func (s *Server) MustStop() {
 	if err := s.lnUDP.Close(); err != nil {
 		logger.Errorf("cannot stop UDP OpenTSDB server: %s", err)
 	}
-	s.cm.CloseAll()
+	s.cm.CloseAll(0)
 	s.wg.Wait()
 	logger.Infof("TCP and UDP OpenTSDB servers at %q have been stopped", s.addr)
 }