From b91eb6de407c26179a1d52f97ca141795e9eb468 Mon Sep 17 00:00:00 2001
From: henrygd <hank@henrygd.me>
Date: Thu, 26 Feb 2026 18:11:33 -0500
Subject: [PATCH] improve root I/O device detection and fallback (#1772)

- Match FILESYSTEM directly against I/O devices if partition lookup
fails
- Fall back to the most active I/O device if no root device is detected
- Add WARN logs in final fallback case to most active device
---
 agent/disk.go      | 45 +++++++++++++++++++++++++++++++++++++++------
 agent/disk_test.go | 22 ++++++++++++++++++++++
 2 files changed, 61 insertions(+), 6 deletions(-)

diff --git a/agent/disk.go b/agent/disk.go
index 876f6815..917fbc90 100644
--- a/agent/disk.go
+++ b/agent/disk.go
@@ -128,7 +128,15 @@ func (a *Agent) initializeDiskInfo() {
 			}
 		}
 		if !hasRoot {
-			slog.Warn("Partition details not found", "filesystem", filesystem)
+			// FILESYSTEM may name a physical disk absent from partitions (e.g.
+			// ZFS lists dataset paths like zroot/ROOT/default, not block devices).
+			// Try matching directly against diskIoCounters.
+			if ioKey, match := findIoDevice(filesystem, diskIoCounters); match {
+				a.fsStats[ioKey] = &system.FsStats{Root: true, Mountpoint: rootMountPoint}
+				hasRoot = true
+			} else {
+				slog.Warn("Partition details not found", "filesystem", filesystem)
+			}
 		}
 	}
 
@@ -194,13 +202,19 @@ func (a *Agent) initializeDiskInfo() {
 		}
 	}
 
-	// If no root filesystem set, use fallback
+	// If no root filesystem set, try the most active I/O device as a last
+	// resort (e.g. ZFS where dataset names are unrelated to disk names).
 	if !hasRoot {
-		rootKey := filepath.Base(rootMountPoint)
-		if _, exists := a.fsStats[rootKey]; exists {
-			rootKey = "root"
+		rootKey := mostActiveIoDevice(diskIoCounters)
+		if rootKey != "" {
+			slog.Warn("Using most active device for root I/O; set FILESYSTEM to override", "device", rootKey)
+		} else {
+			rootKey = filepath.Base(rootMountPoint)
+			if _, exists := a.fsStats[rootKey]; exists {
+				rootKey = "root"
+			}
+			slog.Warn("Root I/O device not detected; set FILESYSTEM to override")
 		}
-		slog.Warn("Root device not detected; root I/O disabled", "mountpoint", rootMountPoint)
 		a.fsStats[rootKey] = &system.FsStats{Root: true, Mountpoint: rootMountPoint}
 	}
 
@@ -305,6 +319,25 @@ func findIoDevice(filesystem string, diskIoCounters map[string]disk.IOCountersSt
 	return best.name, true
 }
 
+// mostActiveIoDevice returns the device with the highest I/O activity,
+// or "" if diskIoCounters is empty.
+func mostActiveIoDevice(diskIoCounters map[string]disk.IOCountersStat) string {
+	var best ioMatchCandidate
+	for _, d := range diskIoCounters {
+		c := ioMatchCandidate{
+			name:  d.Name,
+			bytes: d.ReadBytes + d.WriteBytes,
+			ops:   d.ReadCount + d.WriteCount,
+		}
+		if best.name == "" || c.bytes > best.bytes ||
+			(c.bytes == best.bytes && c.ops > best.ops) ||
+			(c.bytes == best.bytes && c.ops == best.ops && c.name < best.name) {
+			best = c
+		}
+	}
+	return best.name
+}
+
 // prefixRelated reports whether either identifier is a prefix of the other.
 func prefixRelated(a, b string) bool {
 	if a == "" || b == "" || a == b {
diff --git a/agent/disk_test.go b/agent/disk_test.go
index acfcb751..98f16aa3 100644
--- a/agent/disk_test.go
+++ b/agent/disk_test.go
@@ -206,6 +206,28 @@ func TestFilesystemMatchesPartitionSetting(t *testing.T) {
 	})
 }
 
+func TestMostActiveIoDevice(t *testing.T) {
+	t.Run("returns most active device", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"nda0": {Name: "nda0", ReadBytes: 5000, WriteBytes: 5000, ReadCount: 100, WriteCount: 100},
+			"nda1": {Name: "nda1", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 50, WriteCount: 50},
+		}
+		assert.Equal(t, "nda0", mostActiveIoDevice(ioCounters))
+	})
+
+	t.Run("uses deterministic tie-breaker", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"sdb": {Name: "sdb", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 10, WriteCount: 10},
+			"sda": {Name: "sda", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 10, WriteCount: 10},
+		}
+		assert.Equal(t, "sda", mostActiveIoDevice(ioCounters))
+	})
+
+	t.Run("returns empty for empty map", func(t *testing.T) {
+		assert.Equal(t, "", mostActiveIoDevice(map[string]disk.IOCountersStat{}))
+	})
+}
+
 func TestIsDockerSpecialMountpoint(t *testing.T) {
 	testCases := []struct {
 		name       string