LPIC-3 High Availability and Storage Clusters key terms

Cluster resource manager that starts, stops, moves, and monitors services across nodes according to configured constraints and policies. It uses Corosync for messaging and quorum, and fencing to isolate failed nodes.

Cluster messaging and membership layer that provides reliable communication, node liveness detection, and quorum services to Pacemaker. It can use unicast or multicast UDP, or Knet transports, depending on configuration and version.

A majority of cluster nodes that must be present for the cluster to make decisions and run resources. Without quorum, Pacemaker normally stops resources unless a configured policy says otherwise.

Shoot The Other Node In The Head; a fencing mechanism that forcibly powers off or resets a suspected-failed node so it cannot access shared storage or corrupt cluster state. Common agents control IPMI, PDUs, or hypervisor APIs.

A failure condition where cluster nodes lose communication and each partition believes it is the only valid cluster, potentially running the same services and writing to shared storage simultaneously. Fencing and quorum policies prevent it.

A script or program that implements start, stop, monitor, and metadata actions for a specific service so the cluster manager can control and health-check it. Common standards are OCF, LSB, systemd, and service.

A rule in Pacemaker that controls resource placement or ordering, such as location, colocation, order, or ticket constraints. Constraints let administrators express dependencies and preferred nodes without hard-coding commands.

Command-line and web-based administration tool for Pacemaker and Corosync clusters. It creates cluster configuration, manages nodes, resources, constraints, fencing, and quorum settings through simple commands.

Ticket-based coordination service used with Pacemaker to manage resources across separate clusters or sites. A ticket is held by one site at a time, helping prevent both sites from activating the same service.

Storage-Based Death; a fencing method where nodes use a shared block device and watchdog to self-terminate when they lose cluster communication. It avoids relying solely on external power fencing.

A Pacemaker resource type that runs on multiple nodes but allows some instances to be in a promoted (master) role and others in an unpromoted (slave) role, commonly used for DRBD or databases.

A Pacemaker feature that lets cluster resources run on nodes outside the Corosync membership, such as virtual machines or remote sites, while still being managed by the cluster.

Distributed Replicated Block Device; a Linux kernel module that mirrors a block device between two or more nodes over the network. It is often used with Pacemaker and a promoted clone for high availability.

Cluster Logical Volume Manager; an extension of LVM that allows logical volumes to be managed and activated consistently across cluster nodes. It uses a locking service so only one node modifies metadata at a time.

Linux subsystem that presents multiple physical paths to a SAN LUN as one logical device, providing path failover and load balancing when a cable, switch, or controller fails.

Internet Small Computer Systems Interface; a protocol that carries SCSI commands over TCP/IP, letting a host use remote storage as a block device. It includes an initiator on the host and a target on the storage server.

Global File System 2; a shared-disk cluster filesystem for Linux that allows multiple nodes to mount and write to the same block device concurrently. It requires a distributed lock manager and fencing.

Oracle Cluster File System 2; a shared-disk cluster filesystem supported on Linux that coordinates access through a distributed lock manager. It is often used for Oracle databases and requires proper fencing and cluster stack integration.

Distributed Lock Manager; a cluster service that coordinates locks on shared resources so multiple nodes can safely access a shared filesystem or application data without corrupting it.

A SCSI feature that lets a node reserve a LUN and control which other hosts may access it. Clusters use reservations to fence storage access and prevent simultaneous writes after a failure.

Open-source distributed storage platform that provides object, block, and file storage on a single cluster. It distributes data across many OSDs using CRUSH and is managed by MON and MGR daemons.

Reliable Autonomic Distributed Object Store; the core layer of Ceph that stores objects across OSDs, handles replication or erasure coding, and provides the foundation for RBD, CephFS, and RGW.

Object Storage Daemon; a Ceph process that stores data on a local disk and serves reads and writes to clients. Each OSD participates in replication, recovery, and data placement under the CRUSH algorithm.

A logical bucket that groups Ceph objects for placement and replication. CRUSH maps each placement group to a set of OSDs, reducing the amount of per-object placement metadata the cluster must track.

Controlled Replication Under Scalable Hashing; the Ceph algorithm that computes where objects are stored based on the cluster map, enabling deterministic placement without a central lookup table.

Ceph's POSIX-compliant distributed file system, built on RADOS and served through metadata servers (MDS). It supports multiple clients and is often mounted with the kernel or FUSE driver.

RADOS Block Device; Ceph's block storage interface that presents virtual block devices to clients, commonly used by virtual machines and containers. Images are striped across objects in RADOS pools.

Scale-out distributed file system that aggregates storage bricks from multiple servers into volumes. It supports distributed, replicated, and dispersed volume types and heals data after failures.

A basic storage unit in GlusterFS, typically a directory on a filesystem exported by a Gluster server. Volumes are built from bricks, and their layout determines redundancy and performance.

The GlusterFS process that repairs missing or stale data on a brick after a failure or replacement. It compares file metadata and contents among replicas and copies the correct version to affected bricks.

Kernel driver that combines multiple network interfaces into one logical interface for redundancy or throughput. Common modes include active-backup, balance-rr, and 802.3ad LACP, with monitoring via MII or ARP.

Virtual Router Redundancy Protocol; a standards-based protocol that lets multiple routers or hosts share a virtual IP address. One node is master and answers for the VIP; others take over if it fails.

Linux daemon that implements VRRP for virtual IP failover and can also perform health checks on local services. It is often used for simple two-node high availability without a full Pacemaker cluster.

Linux tool for creating and managing software RAID arrays using the md kernel driver. It supports levels such as RAID 0, 1, 5, 6, and 10, and can monitor and rebuild degraded arrays.

Hardware or software timer that reboots a node if the operating system stops feeding it. In HA clusters it enforces fencing when the node hangs or loses communication.

Network UPS Tools; a suite that monitors uninterruptible power supplies and can trigger orderly shutdown of a single node or cluster when battery power is low.