I85PREP
⌘ K
LPIC-3 High Availability and Storage Clusters study materials
LPIC-3 High Availability and Storage Clusters study resources

LPIC-3 High Availability and Storage Clusters: Single Node High Availability

LPIC-3 High Availability and Storage Clusters · Work through your exam, one topic at a time.

All guide sections

Single Node High Availability

Single Node High Availability

This domain tests redundancy inside one host: mdadm RAID, LVM/DM RAID, multipath, network bonding, boot redundancy, and watchdogs. The goal is to survive a disk, path, link, or hung kernel without cluster-wide failover.

Core rules

  • Use RAID1 or RAID10 for redundancy. RAID5/6 saves capacity but rebuilds are slow and a second failure can be fatal; add a hot spare where appropriate.
  • Manage arrays with mdadm --create, inspect /proc/mdstat and mdadm --detail, and simulate repair with --fail, --remove, --add.
  • Persist the array: mdadm --detail --scan into mdadm.conf, then rebuild the initramfs with dracut -f or update-initramfs -u. Install the bootloader on every mirror member.
  • LVM RAID (lvcreate --type raid1 -m1) adds mirroring above PVs; pvmove evacuates a failing PV. Neither replaces backups.
  • Multipath needs two fabrics or HBAs to the same LUN; use multipath -ll and keep multipathd running.
  • Bonding mode 1 (active-backup) needs no switch configuration; mode 4 (802.3ad) requires LACP on the switch. Set miimon to detect link failure.
  • A watchdog resets a hung node; systemd can arm it with RuntimeWatchdogSec=.

Worked example

Mirror two partitions: mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sdb1 /dev/sdc1. Make a filesystem, add the scan line to /etc/mdadm.conf, rebuild the initramfs, and run grub2-install /dev/sdb and grub2-install /dev/sdc. To test: mdadm /dev/md0 --fail /dev/sdb1 --remove /dev/sdb1, insert a replacement, then mdadm /dev/md0 --add /dev/sdd1; watch the rebuild in /proc/mdstat. Reasoning: RAID1 keeps data available, and bootloader copies on both disks allow boot after a disk loss.

Traps

Putting both mirrors on one physical disk, forgetting the initramfs update, treating RAID as backup, configuring LACP without the switch, leaving multipathd disabled, and setting a watchdog timeout shorter than normal boot or workload stalls.