fix(agent): handle 32-bit wrap of disk I/O time counters (#2407)

agent/disk.go

@@ -3,6 +3,7 @@ package agent
 import (
 	"context"
 	"log/slog"
+	"math"
 	"os"
 	"path/filepath"
 	"runtime"
@@ -686,25 +687,27 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 			// This is the total number of milliseconds spent by all reads (as
 			// measured from __make_request() to end_that_request_last()).
 			// https://www.kernel.org/doc/Documentation/iostats.txt (fields 4, 8)
-			diskReadTime := utils.TwoDecimals(float64(d.ReadTime-prev.readTime) / float64(msElapsed) * 100)
-			diskWriteTime := utils.TwoDecimals(float64(d.WriteTime-prev.writeTime) / float64(msElapsed) * 100)
+			deltaReadTime := ioTimeDelta(d.ReadTime, prev.readTime)
+			deltaWriteTime := ioTimeDelta(d.WriteTime, prev.writeTime)
+			diskReadTime := utils.TwoDecimals(float64(deltaReadTime) / float64(msElapsed) * 100)
+			diskWriteTime := utils.TwoDecimals(float64(deltaWriteTime) / float64(msElapsed) * 100)
 
 			// I/O utilization %: fraction of wall time the device had any I/O in progress (0-100).
-			diskIoUtilPct := utils.TwoDecimals(float64(d.IoTime-prev.ioTime) / float64(msElapsed) * 100)
+			diskIoUtilPct := utils.TwoDecimals(float64(ioTimeDelta(d.IoTime, prev.ioTime)) / float64(msElapsed) * 100)
 
 			// Weighted I/O: queue-depth weighted I/O time, normalized to interval (can exceed 100%).
 			// Linux kernel field 11: incremented by iops_in_progress × ms_since_last_update.
 			// Used to display queue depth. Multipled by 100 to increase accuracy of digit truncation (divided by 100 in UI).
-			diskWeightedIO := utils.TwoDecimals(float64(d.WeightedIO-prev.weightedIO) / float64(msElapsed) * 100)
+			diskWeightedIO := utils.TwoDecimals(float64(ioTimeDelta(d.WeightedIO, prev.weightedIO)) / float64(msElapsed) * 100)
 
 			// r_await / w_await: average time per read/write operation in milliseconds.
 			// Equivalent to r_await and w_await in iostat.
 			var rAwait, wAwait float64
 			if deltaReadCount := d.ReadCount - prev.readCount; deltaReadCount > 0 {
-				rAwait = utils.TwoDecimals(float64(d.ReadTime-prev.readTime) / float64(deltaReadCount))
+				rAwait = utils.TwoDecimals(float64(deltaReadTime) / float64(deltaReadCount))
 			}
 			if deltaWriteCount := d.WriteCount - prev.writeCount; deltaWriteCount > 0 {
-				wAwait = utils.TwoDecimals(float64(d.WriteTime-prev.writeTime) / float64(deltaWriteCount))
+				wAwait = utils.TwoDecimals(float64(deltaWriteTime) / float64(deltaWriteCount))
 			}
 
 			// Update global fsStats baseline for cross-interval correctness
@@ -740,6 +743,21 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 	}
 }
 
+// ioTimeDelta returns the increase of a cumulative millisecond counter from
+// the disk I/O stats. Linux prints these fields of /proc/diskstats as 32-bit
+// unsigned ints, so they wrap to zero at 2^32. A busy disk reaches that in
+// days for the weighted I/O time. Other platforms report 64-bit counters,
+// so a lower value there is a reset.
+func ioTimeDelta(current, previous uint64) uint64 {
+	if current >= previous {
+		return current - previous
+	}
+	if runtime.GOOS == "linux" && previous <= math.MaxUint32 {
+		return current + (math.MaxUint32 + 1 - previous)
+	}
+	return 0
+}
+
 // getRootMountPoint returns the appropriate root mount point for the system.
 // On Windows it returns the system drive (e.g. "C:").
 // For immutable systems like Fedora Silverblue, it returns /sysroot instead of /

agent/disk_io_linux_test.go

@@ -0,0 +1,75 @@
+//go:build linux
+
+package agent
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// Linux prints four millisecond fields of /proc/diskstats as 32-bit unsigned ints:
+// read time, write time, io time and weighted io time. They wrap to zero at 2^32.
+func TestUpdateDiskIoTimeCounterWrap(t *testing.T) {
+	const wrap = uint64(1) << 32
+
+	tests := []struct {
+		name string
+		base uint64 // added to every previous time counter
+	}{
+		{"no wrap", 0},
+		{"32-bit wrap", wrap - 1000},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Deltas over 60s: read 300ms / 10 ops, write 400ms / 20 ops,
+			// io time 1200ms, weighted io 3000ms.
+			prev := prevDisk{
+				readBytes:  20000 * 512,
+				writeBytes: 10000 * 512,
+				readTime:   tt.base + 900,
+				writeTime:  tt.base + 700,
+				ioTime:     tt.base + 400,
+				weightedIO: tt.base,
+				readCount:  1000,
+				writeCount: 500,
+				at:         time.Now().Add(-60 * time.Second),
+			}
+			cur := func(v uint64) uint64 { return v % wrap }
+			line := fmt.Sprintf("   8       0 sda %d 0 %d %d %d 0 %d %d 0 %d %d\n",
+				1010, 21200, cur(prev.readTime+300),
+				520, 10400, cur(prev.writeTime+400),
+				cur(prev.ioTime+1200), cur(prev.weightedIO+3000))
+
+			dir := t.TempDir()
+			require.NoError(t, os.WriteFile(filepath.Join(dir, "diskstats"), []byte(line), 0o644))
+			t.Setenv("HOST_PROC", dir)
+			t.Setenv("HOST_SYS", dir)
+			t.Setenv("HOST_DEV", dir)
+			t.Setenv("HOST_RUN", dir)
+
+			fs := &system.FsStats{Root: true}
+			a := &Agent{
+				fsNames:  []string{"sda"},
+				fsStats:  map[string]*system.FsStats{"sda": fs},
+				diskPrev: map[uint16]map[string]prevDisk{60000: {"sda": prev}},
+			}
+			var stats system.Stats
+			a.updateDiskIo(60000, &stats)
+
+			// Same order as DiskIoStats in system.FsStats.
+			want := [6]float64{0.5, 0.67, 2, 30, 20, 5}
+			for i := range want {
+				assert.InDelta(t, want[i], fs.DiskIoStats[i], 0.01, "DiskIoStats[%d]", i)
+				assert.InDelta(t, want[i], stats.DiskIoStats[i], 0.01, "system DiskIoStats[%d]", i)
+			}
+		})
+	}
+}

agent/disk_test.go

@@ -3,7 +3,9 @@
 package agent
 
 import (
+	"math"
 	"os"
+	"runtime"
 	"strings"
 	"testing"
 	"time"
@@ -1041,3 +1043,17 @@ func TestInitializeDiskIoStatsResetsTrackedDevices(t *testing.T) {
 	assert.Equal(t, uint64(50), agent.fsStats["sdb"].TotalRead)
 	assert.Equal(t, uint64(60), agent.fsStats["sdb"].TotalWrite)
 }
+
+func TestIoTimeDelta(t *testing.T) {
+	assert.Equal(t, uint64(300), ioTimeDelta(1200, 900))
+
+	// A lower value is a 32-bit wrap only on Linux. Other platforms
+	// report 64-bit counters, so there it is a reset.
+	var want uint64
+	if runtime.GOOS == "linux" {
+		want = 1200
+	}
+	assert.Equal(t, want, ioTimeDelta(200, math.MaxUint32+1-1000))
+
+	assert.Equal(t, uint64(0), ioTimeDelta(200, math.MaxUint32+1000))
+}