CLI: amend 'performance throughput'

* compute all throughputs on the client side - given interval and the corresponding `KindSize` metric - Clouds now separately * add `SizeToThroughput` api helper; add "naming conventions" Signed-off-by: Alex Aizman <[email protected]>
NVIDIA · Dec 27, 2024 · a8374ce · a8374ce
1 parent 0996dc6
commit a8374ce
Show file tree

Hide file tree

Showing 6 changed files with 82 additions and 37 deletions.
diff --git a/cmd/cli/cli/performance.go b/cmd/cli/cli/performance.go
@@ -171,11 +171,10 @@ func showCountersHandler(c *cli.Context) error {
 	return showPerfTab(c, selected, nil, cmdShowCounters, nil, false)
 }
 
-// TODO -- FIXME: revisit computing over totals (compare with latency)
+// TODO -- FIXME: support regex, verbose
 func showThroughputHandler(c *cli.Context) error {
 	var (
 		totals       = make(map[string]int64, 4) // throughput metrics ("columns") to tally up
-		regexStr     = parseStrFlag(c, regexColsFlag)
 		metrics, err = getMetricNames(c)
 		verbose      = flagIsSet(c, verboseFlag)
 	)
@@ -187,31 +186,41 @@ func showThroughputHandler(c *cli.Context) error {
 
 	selected := make(cos.StrKVs, len(metrics))
 	for name, kind := range metrics {
-		switch name {
-		case stats.GetSize, stats.GetCount, stats.PutSize, stats.PutCount:
+		// - always show io-errors
+		// - other errors only if (get|put) and verbose
+		// - otherwise, skip anything other than the two relevant kinds
+		if stats.IsIOErrMetric(name) {
 			selected[name] = kind
 			continue
 		}
-		if !verbose && regexStr == "" {
-			if cos.StringInSlice(name, verboseCounters[:]) {
+		if stats.IsErrMetric(name) {
+			if !verbose {
 				continue
 			}
+			if !strings.Contains(name, "get") && !strings.Contains(name, "put") {
+				continue
+			}
+			selected[name] = kind
+			continue
 		}
 
-		switch {
-		case kind == stats.KindThroughput:
-			// 1. all throughput
-			selected[name] = kind
-			totals[name] = 0
-		case strings.HasSuffix(name, "."+stats.GetCount) || strings.HasSuffix(name, "."+stats.GetSize):
-			selected[name] = kind
-		case strings.HasSuffix(name, "."+stats.PutCount) || strings.HasSuffix(name, "."+stats.PutSize):
-			selected[name] = kind
-		case stats.IsErrMetric(name):
-			// 3. errors (compare with latency selection below)
-			if strings.Contains(name, "get") || strings.Contains(name, "put") {
+		switch kind {
+		case stats.KindCounter:
+			if name == stats.GetCount || name == stats.PutCount ||
+				strings.HasSuffix(name, "."+stats.GetCount) || strings.HasSuffix(name, "."+stats.PutCount) {
 				selected[name] = kind
 			}
+			continue
+		case stats.KindSize:
+			if name == stats.GetSize || name == stats.PutSize ||
+				strings.HasSuffix(name, "."+stats.GetSize) || strings.HasSuffix(name, "."+stats.PutSize) {
+				selected[name] = kind
+
+				if bpsName := stats.SizeToThroughput(name, stats.KindSize); bpsName != "" {
+					selected[bpsName] = stats.KindThroughput
+				}
+			}
+			continue
 		}
 	}
 	// `true` to show average get/put sizes
@@ -232,24 +241,34 @@ func _throughput(c *cli.Context, metrics cos.StrKVs, mapBegin, mapEnd teb.StstMa
 			continue
 		}
 		for name, v := range begin.Tracker {
-			if kind, ok := metrics[name]; !ok || kind != stats.KindThroughput {
+			kind, ok := metrics[name]
+			if !ok || kind != stats.KindSize {
+				continue
+			}
+			bpsName := stats.SizeToThroughput(name, stats.KindSize)
+			if bpsName == "" {
 				continue
 			}
 			vend := end.Tracker[name]
+
 			if vend.Value <= v.Value {
 				// no changes, nothing to show
 				v.Value = 0
 				begin.Tracker[name] = v
 				continue
 			}
 
+			//
+			// given this (KindSize) metric change and elapsed time, add computed throughput:
+			//
 			v.Value = (vend.Value - v.Value) / seconds
-			begin.Tracker[name] = v
+			begin.Tracker[bpsName] = v
 			num++
 		}
 	}
+
 	idle = num == 0
-	return
+	return idle
 }
 
 // otherwise, skip computing (TODO: add comdline option)

diff --git a/cmd/cli/go.mod b/cmd/cli/go.mod
@@ -3,7 +3,7 @@ module github.com/NVIDIA/aistore/cmd/cli
 go 1.23.2
 
 require (
-	github.com/NVIDIA/aistore v1.3.26-0.20241226184445-e6814a27d17d
+	github.com/NVIDIA/aistore v1.3.26-0.20241227164600-fbb55bc59304
 	github.com/fatih/color v1.18.0
 	github.com/json-iterator/go v1.1.12
 	github.com/onsi/ginkgo/v2 v2.21.0

diff --git a/cmd/cli/go.sum b/cmd/cli/go.sum
@@ -1,7 +1,7 @@
 code.cloudfoundry.org/bytefmt v0.0.0-20190710193110-1eb035ffe2b6/go.mod h1:wN/zk7mhREp/oviagqUXY3EwuHhWyOvAdsn5Y4CzOrc=
 github.com/BurntSushi/toml v1.4.0/go.mod h1:ukJfTF/6rtPPRCnwkur4qwRxa8vTRFBF0uk2lLoLwho=
-github.com/NVIDIA/aistore v1.3.26-0.20241226184445-e6814a27d17d h1:hhv9YZyWW0QCgxv82OBgq3+nFBn+YS8ztkmg1zC7Re4=
-github.com/NVIDIA/aistore v1.3.26-0.20241226184445-e6814a27d17d/go.mod h1:mjhY9OGIZULaC79+iRfzEUvUZw7aIWklJ8um321QVpw=
+github.com/NVIDIA/aistore v1.3.26-0.20241227164600-fbb55bc59304 h1:4BQqN3OSXWS04ciLhSOrS+VuEYRIAoXO8ci6J3W8+gk=
+github.com/NVIDIA/aistore v1.3.26-0.20241227164600-fbb55bc59304/go.mod h1:mjhY9OGIZULaC79+iRfzEUvUZw7aIWklJ8um321QVpw=
 github.com/OneOfOne/xxhash v1.2.8 h1:31czK/TI9sNkxIKfaUfGlU47BAxQ0ztGgd9vPyqimf8=
 github.com/OneOfOne/xxhash v1.2.8/go.mod h1:eZbhyaAYD41SGSSsnmcpxVoRiQ/MPUTjUdIIOT9Um7Q=
 github.com/VividCortex/ewma v1.1.1/go.mod h1:2Tkkvm3sRDVXaiyucHiACn4cqf7DpdyLvmxzcbUokwA=

diff --git a/stats/api.go b/stats/api.go
@@ -12,6 +12,7 @@ import (
 	"github.com/NVIDIA/aistore/api/apc"
 	"github.com/NVIDIA/aistore/cmn"
 	"github.com/NVIDIA/aistore/cmn/cos"
+	"github.com/NVIDIA/aistore/cmn/debug"
 	"github.com/NVIDIA/aistore/core"
 	"github.com/NVIDIA/aistore/core/meta"
 	"github.com/NVIDIA/aistore/fs"
@@ -154,10 +155,15 @@ func IsIOErrMetric(name string) bool {
 	return strings.HasPrefix(name, ioErrPrefix) // e.g., "err.io.get.n" (see ioErrNames)
 }
 
+//
+// name translations, to recompute latency and throughput over client-controlled intervals
+// see "Naming conventions"
+//
+
 // compare with base.init() at ais/backend/common
-func LatencyToCounter(latency string) string {
+func LatencyToCounter(latName string) string {
 	// 1. basics first
-	switch latency {
+	switch latName {
 	case GetLatency, GetRedirLatency, GetLatencyTotal:
 		return GetCount
 	case PutLatency, PutRedirLatency, PutLatencyTotal:
@@ -170,17 +176,17 @@ func LatencyToCounter(latency string) string {
 		return AppendCount
 	}
 	// 2. filter out
-	if !strings.Contains(latency, "get.") && !strings.Contains(latency, "put.") {
+	if !strings.Contains(latName, "get.") && !strings.Contains(latName, "put.") {
 		return ""
 	}
 	// backend first
-	if strings.HasSuffix(latency, ".ns.total") {
+	if strings.HasSuffix(latName, ".ns.total") {
 		for prefix := range apc.Providers {
 			if prefix == apc.AIS {
 				prefix = apc.RemAIS
 			}
-			if strings.HasPrefix(latency, prefix) {
-				if strings.Contains(latency, ".get.") {
+			if strings.HasPrefix(latName, prefix) {
+				if strings.Contains(latName, ".get.") {
 					return prefix + "." + GetCount
 				}
 				return prefix + "." + PutCount
@@ -189,3 +195,14 @@ func LatencyToCounter(latency string) string {
 	}
 	return ""
 }
+
+func SizeToThroughput(name, kind string) string {
+	if kind != KindSize {
+		return ""
+	}
+	if !strings.HasSuffix(name, ".size") { // see "Naming conventions"
+		debug.Assert(false, name)
+		return ""
+	}
+	return strings.TrimSuffix(name, ".size") + ".bps"
+}
diff --git a/stats/common.go b/stats/common.go
@@ -32,7 +32,14 @@ import (
 	jsoniter "github.com/json-iterator/go"
 )
 
-// All error counters must have "err_" prefix (see `errPrefix`)
+// Naming conventions:
+// ========================================================
+// "*.n"    - KindCounter
+// "*.ns"   - KindLatency, KindTotal (nanoseconds)
+// "*.size" - KindSize (bytes)
+// "*.bps"  - KindThroughput, KindComputedThroughput
+//
+// all error counters must have "err_" prefix (see `errPrefix`)
 
 // Linkage:
 // - this source is common for both Prometheus (common_prom.go) and StatsD (common_statsd.go)

diff --git a/stats/target_stats.go b/stats/target_stats.go
@@ -22,13 +22,15 @@ import (
 	"github.com/NVIDIA/aistore/fs"
 )
 
-// Naming Convention:
+// Naming conventions:
+// ========================================================
+// "*.n"    - KindCounter
+// "*.ns"   - KindLatency, KindTotal (nanoseconds)
+// "*.size" - KindSize (bytes)
+// "*.bps"  - KindThroughput, KindComputedThroughput
 //
-//	-> "*.n" - counter
-//	-> "*.ns" - latency (nanoseconds)
-//	-> "*.size" - size (bytes)
-//	-> "*.bps" - throughput (in byte/s)
-//	-> "*.id" - ID
+// all error counters must have "err_" prefix (see `errPrefix`)
+
 const (
 	// KindCounter & KindSize - always incremented