server: don't double-count RAID volumes in disk metrics

Previously, we wouldn't exclude volumes from disk counters
that are likely to be double-counted such as RAID logical
volumes that are composed of physical volumes that are also
independently present in disk metrics. This change adds a
regex-based filter, overridable with env vars, that excludes
common double-counted volume patterns.

Fixes cockroachdb#97867.

Epic: none

Release note (bug fix): Avoids double-counting disk read/write bytes
in disk metrics if Cockroach observes volumes that are likely to
be duplicated in reported disk counters, such as RAID logical vs
physical volumes.

pkg/server/status/BUILD.bazel

@@ -9,8 +9,10 @@ go_library(
         "health_check.go",
         "recorder.go",
         "runtime.go",
+        "runtime_generic.go",
         "runtime_jemalloc.go",
         "runtime_jemalloc_darwin.go",
+        "runtime_linux.go",
         "runtime_log.go",
     ],
     # keep
@@ -129,6 +131,7 @@ go_test(
         "jemalloc_test.go",
         "main_test.go",
         "recorder_test.go",
+        "runtime_linux_test.go",
         "runtime_stats_test.go",
         "runtime_test.go",
     ],

pkg/server/status/runtime.go

@@ -13,12 +13,14 @@ package status
 import (
 	"context"
 	"os"
+	"regexp"
 	"runtime"
 	"runtime/debug"
 	"time"
 
 	"github.com/cockroachdb/cockroach/pkg/build"
 	"github.com/cockroachdb/cockroach/pkg/util/cgroups"
+	"github.com/cockroachdb/cockroach/pkg/util/envutil"
 	"github.com/cockroachdb/cockroach/pkg/util/goschedstats"
 	"github.com/cockroachdb/cockroach/pkg/util/hlc"
 	"github.com/cockroachdb/cockroach/pkg/util/log"
@@ -251,6 +253,13 @@ var (
 	}
 )
 
+// diskMetricsIgnoredDevices is a regex that matches any block devices that must be
+// ignored for disk metrics (eg. sys.host.disk.write.bytes), as those devices
+// have likely been counted elsewhere. This prevents us from double-counting,
+// for instance, RAID volumes under both the logical volume and under the
+// physical volume(s).
+var diskMetricsIgnoredDevices = envutil.EnvOrDefaultString("COCKROACH_DISK_METRICS_IGNORED_DEVICES", getDefaultIgnoredDevices())
+
 // getCgoMemStats is a function that fetches stats for the C++ portion of the code.
 // We will not necessarily have implementations for all builds, so check for nil first.
 // Returns the following:
@@ -692,7 +701,7 @@ func getSummedDiskCounters(ctx context.Context) (DiskStats, error) {
 		return DiskStats{}, err
 	}
 
-	return sumDiskCounters(diskCounters), nil
+	return sumAndFilterDiskCounters(diskCounters)
 }
 
 func getSummedNetStats(ctx context.Context) (net.IOCountersStat, error) {
@@ -704,11 +713,26 @@ func getSummedNetStats(ctx context.Context) (net.IOCountersStat, error) {
 	return sumNetworkCounters(netCounters), nil
 }
 
-// sumDiskCounters returns a new disk.IOCountersStat whose values are the sum of the
-// values in the slice of disk.IOCountersStats passed in.
-func sumDiskCounters(disksStats []DiskStats) DiskStats {
+// sumAndFilterDiskCounters returns a new disk.IOCountersStat whose values are
+// the sum of the values in the slice of disk.IOCountersStats passed in. It
+// filters out any disk counters that are likely reflecting values already
+// counted elsewhere, eg. md* logical volumes that are created out of RAIDing
+// underlying drives. The filtering regex defaults to a platform-dependent one.
+func sumAndFilterDiskCounters(disksStats []DiskStats) (DiskStats, error) {
 	output := DiskStats{}
+	var ignored *regexp.Regexp
+	if diskMetricsIgnoredDevices != "" {
+		var err error
+		ignored, err = regexp.Compile(diskMetricsIgnoredDevices)
+		if err != nil {
+			return output, err
+		}
+	}
+
 	for _, stats := range disksStats {
+		if ignored != nil && ignored.MatchString(stats.Name) {
+			continue
+		}
 		output.ReadBytes += stats.ReadBytes
 		output.readCount += stats.readCount
 		output.readTime += stats.readTime
@@ -722,7 +746,7 @@ func sumDiskCounters(disksStats []DiskStats) DiskStats {
 
 		output.iopsInProgress += stats.iopsInProgress
 	}
-	return output
+	return output, nil
 }
 
 // subtractDiskCounters subtracts the counters in `sub` from the counters in `from`,

pkg/server/status/runtime_generic.go

@@ -0,0 +1,18 @@
+// Copyright 2023 The Cockroach Authors.
+//
+// Use of this software is governed by the Business Source License
+// included in the file licenses/BSL.txt.
+//
+// As of the Change Date specified in that file, in accordance with
+// the Business Source License, use of this software will be governed
+// by the Apache License, Version 2.0, included in the file
+// licenses/APL.txt.
+
+//go:build !linux
+// +build !linux
+
+package status
+
+func getDefaultIgnoredDevices() string {
+	return ""
+}

pkg/server/status/runtime_linux.go

@@ -0,0 +1,18 @@
+// Copyright 2023 The Cockroach Authors.
+//
+// Use of this software is governed by the Business Source License
+// included in the file licenses/BSL.txt.
+//
+// As of the Change Date specified in that file, in accordance with
+// the Business Source License, use of this software will be governed
+// by the Apache License, Version 2.0, included in the file
+// licenses/APL.txt.
+
+//go:build linux
+// +build linux
+
+package status
+
+func getDefaultIgnoredDevices() string {
+	return "^(ram|loop|fd|(h|s|v|xv)d[a-z]|nvme\\d+n\\d+p)\\d+$"
+}

pkg/server/status/runtime_linux_test.go

@@ -0,0 +1,66 @@
+// Copyright 2023 The Cockroach Authors.
+//
+// Use of this software is governed by the Business Source License
+// included in the file licenses/BSL.txt.
+//
+// As of the Change Date specified in that file, in accordance with
+// the Business Source License, use of this software will be governed
+// by the Apache License, Version 2.0, included in the file
+// licenses/APL.txt.
+
+//go:build linux
+// +build linux
+
+package status
+
+import (
+	"reflect"
+	"testing"
+
+	"github.com/cockroachdb/cockroach/pkg/util/leaktest"
+)
+
+func TestSumAndFilterDiskCountersLinux(t *testing.T) {
+	defer leaktest.AfterTest(t)()
+
+	counters := []DiskStats{
+		{
+			Name:           "nvme0",
+			ReadBytes:      1,
+			readCount:      1,
+			iopsInProgress: 1,
+			WriteBytes:     1,
+			writeCount:     1,
+		},
+		{
+			Name:           "sda1",
+			ReadBytes:      1,
+			readCount:      1,
+			iopsInProgress: 1,
+			WriteBytes:     1,
+			writeCount:     1,
+		},
+		{ // This must be excluded from the sum.
+			Name:           "nvme1n1",
+			ReadBytes:      1,
+			readCount:      1,
+			iopsInProgress: 1,
+			WriteBytes:     1,
+			writeCount:     1,
+		},
+	}
+	summed, err := sumAndFilterDiskCounters(counters)
+	if err != nil {
+		t.Fatalf("error: %s", err.Error())
+	}
+	expected := DiskStats{
+		ReadBytes:      2,
+		readCount:      2,
+		WriteBytes:     2,
+		writeCount:     2,
+		iopsInProgress: 2,
+	}
+	if !reflect.DeepEqual(summed, expected) {
+		t.Fatalf("expected %+v; got %+v", expected, summed)
+	}
+}

pkg/server/status/runtime_test.go

@@ -18,7 +18,7 @@ import (
 	"github.com/shirou/gopsutil/v3/net"
 )
 
-func TestSumDiskCounters(t *testing.T) {
+func TestSumAndFilterDiskCounters(t *testing.T) {
 	defer leaktest.AfterTest(t)()
 
 	counters := []DiskStats{
@@ -37,7 +37,10 @@ func TestSumDiskCounters(t *testing.T) {
 			writeCount:     1,
 		},
 	}
-	summed := sumDiskCounters(counters)
+	summed, err := sumAndFilterDiskCounters(counters)
+	if err != nil {
+		t.Fatalf("error: %s", err.Error())
+	}
 	expected := DiskStats{
 		ReadBytes:      2,
 		readCount:      2,