feat(mount-disks): mount auto-provisioned disks by UUID

pkg/cluster/manager/deploy_volume_server.go

@@ -3,11 +3,13 @@ package manager
 import (
 	"bytes"
 	"fmt"
+	"strings"
+	"time"
+
 	"github.com/seaweedfs/seaweed-up/pkg/cluster/spec"
 	"github.com/seaweedfs/seaweed-up/pkg/disks"
 	"github.com/seaweedfs/seaweed-up/pkg/operator"
 	"github.com/seaweedfs/seaweed-up/scripts"
-	"strings"
 )
 
 func (m *Manager) DeployVolumeServer(masters []string, volumeServerSpec *spec.VolumeServerSpec, index int) error {
@@ -119,14 +121,22 @@ func (m *Manager) prepareUnmountedDisks(op operator.CommandOperator) error {
 	}
 	fmt.Printf("disks2: %+v\n", disks)
 
-	// format disk if no fstype
+	// format disk if no fstype, then resolve the resulting UUID so the fstab
+	// entry written below can mount by UUID instead of by device path.
 	for _, dev := range disks {
 		if dev.FilesystemType == "" {
 			info("mkfs " + dev.Path)
 			if err := m.sudo(op, fmt.Sprintf("mkfs.ext4 %s", dev.Path)); err != nil {
 				return fmt.Errorf("create file system on %s: %v", dev.Path, err)
 			}
 		}
+		if dev.UUID == "" {
+			uuid, err := m.probeDiskUUID(op, dev.Path)
+			if err != nil {
+				return fmt.Errorf("resolve UUID for %s: %v", dev.Path, err)
+			}
+			dev.UUID = uuid
+		}
 	}
 
 	// mount them
@@ -148,6 +158,7 @@ func (m *Manager) prepareUnmountedDisks(op operator.CommandOperator) error {
 
 			data := map[string]interface{}{
 				"DevicePath": dev.Path,
+				"DeviceUUID": dev.UUID,
 				"MountPoint": targetMountPoint,
 			}
 			prepareScript, err := scripts.RenderScript("prepare_disk.sh", data)
@@ -160,8 +171,8 @@ func (m *Manager) prepareUnmountedDisks(op operator.CommandOperator) error {
 				return fmt.Errorf("error received during upload mount script: %s", err)
 			}
 
-			info("mount " + dev.DeviceName + "...")
-			err = op.Execute(fmt.Sprintf("cat /tmp/mount_%s.sh | SUDO_PASS=\"%s\" sh -\n", dev.DeviceName, m.sudoPass))
+			info(fmt.Sprintf("mount %s (UUID=%s) at %s", dev.DeviceName, dev.UUID, targetMountPoint))
+			err = op.Execute(fmt.Sprintf("cat /tmp/mount_%s.sh | SUDO_PASS=%s sh -\n", dev.DeviceName, shellSingleQuote(m.sudoPass)))
 			if err != nil {
 				return fmt.Errorf("error received during mount: %s", err)
 			}
@@ -171,3 +182,43 @@ func (m *Manager) prepareUnmountedDisks(op operator.CommandOperator) error {
 
 	return nil
 }
+
+// probeDiskUUID reads the filesystem UUID of path via blkid. After mkfs the
+// superblock is written but udev may not yet have re-read it, so we let it
+// settle and retry a few times before giving up. blkid needs root to read
+// raw block devices on most distros (and -p always needs root), so we wrap
+// it the same way m.sudo does. Returning an error here aborts the deploy
+// rather than writing a broken fstab entry that would leave the host unable
+// to boot.
+func (m *Manager) probeDiskUUID(op operator.CommandOperator, path string) (string, error) {
+	// Best-effort settle. Ignore errors — `udevadm` is missing on some
+	// minimal images and that's OK; the retry loop below will still pick
+	// up the UUID once it's available.
+	_ = m.sudo(op, "command -v udevadm >/dev/null 2>&1 && udevadm settle || true")
+
+	// -p bypasses the blkid cache and re-probes the superblock directly,
+	// which is what we want right after mkfs.
+	probeCmd := fmt.Sprintf("blkid -p -s UUID -o value %s", shellSingleQuote(path))
+	if m.sudoPass != "" {
+		probeCmd = fmt.Sprintf("echo %s | sudo -S %s", shellSingleQuote(m.sudoPass), probeCmd)
+	}
+
+	const attempts = 5
+	var lastErr error
+	for i := 0; i < attempts; i++ {
+		out, err := op.Output(probeCmd)
+		if err == nil {
+			uuid := strings.TrimSpace(string(out))
+			if uuid != "" {
+				return uuid, nil
+			}
+		} else {
+			lastErr = err
+		}
+		time.Sleep(500 * time.Millisecond)
+	}
+	if lastErr != nil {
+		return "", fmt.Errorf("blkid returned no UUID after %d attempts: %v", attempts, lastErr)
+	}
+	return "", fmt.Errorf("blkid returned no UUID after %d attempts", attempts)
+}

scripts/prepare_disk.sh

@@ -27,16 +27,25 @@ setup_env() {
 
   MOUNT_POINT={{.MountPoint}}
   DEVICE_PATH={{.DevicePath}}
+  DEVICE_UUID={{.DeviceUUID}}
 }
 
 setup_mount() {
 
   info "Setup Mount Point"
   $SUDO mkdir -p -m 755 ${MOUNT_POINT}
-  info "add ${DEVICE_PATH} ${MOUNT_POINT} to fstab"
-  echo "${DEVICE_PATH} ${MOUNT_POINT} ext4 noatime 0 2" | $SUDO tee -a /etc/fstab
-  info "mount ${DEVICE_PATH} ${MOUNT_POINT}"
-  $SUDO mount ${DEVICE_PATH} ${MOUNT_POINT}
+  if grep -qE "^UUID=${DEVICE_UUID}[[:space:]]" /etc/fstab; then
+    info "UUID=${DEVICE_UUID} already in fstab; skipping append"
+  else
+    info "add UUID=${DEVICE_UUID} (${DEVICE_PATH}) ${MOUNT_POINT} to fstab"
+    echo "UUID=${DEVICE_UUID} ${MOUNT_POINT} ext4 noatime 0 2" | $SUDO tee -a /etc/fstab
+  fi
+  if mountpoint -q "${MOUNT_POINT}"; then
+    info "${MOUNT_POINT} already mounted; skipping mount"
+  else
+    info "mount ${DEVICE_PATH} (UUID=${DEVICE_UUID}) at ${MOUNT_POINT}"
+    $SUDO mount ${MOUNT_POINT}
+  fi
 
   return 0
 }