LPIC-3 High Availability and Storage Clusters flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Corosync votequorum

Corosync votequorum tracks node votes and determines whether a partition has quorum; Pacemaker permits resource actions only when quorum is present, unless no-quorum-policy is set to ignore.

Pacemaker STONITH

STONITH fencing power-cycles or blocks a failed node so it cannot access shared storage; Pacemaker requires a working fencing device before recovering resources when stonith-enabled is true.

Pacemaker resource stickiness

Stickiness is a resource meta-attribute score that biases a resource toward its current node; higher positive values reduce migration, while zero permits balancing and negative values encourage movement.

Pacemaker location constraint

A location constraint assigns a score for a resource on a node or node set; positive values prefer, negative values avoid, and -INFINITY forbids placement there.

Pacemaker order constraint

An order constraint defines start and stop sequencing between two resources; mandatory means the first starts before the second and stops after it, while advisory only affects scoring.

Pacemaker colocation constraint

A colocation constraint specifies that resources should run on the same node or on the same or different nodes; positive scores prefer together, negative scores prefer apart, and INFINITY enforces.

pcs cluster standby

The pcs cluster standby command places a node in standby mode, causing Pacemaker to move managed resources away while Corosync membership and cluster services remain active.

Quorum device qdevice

A qdevice supplies an additional vote to a Pacemaker cluster through corosync-qnetd; it helps a two-node cluster retain quorum after one node fails but does not replace fencing.

DM-Multipath

Device mapper multipath presents multiple SAN paths to one LUN as a single block device; multipathd monitors paths and switches I/O according to path priority and failback policy.

LUN masking

LUN masking configures a storage array to expose a LUN only to specified host initiator WWNs or IQNs; fabric zoning separately controls which ports can communicate.

iSCSI target and initiator

An iSCSI target exports block LUNs over TCP using an IQN; an initiator logs in to the target, discovers LUNs, and creates local SCSI devices, optionally authenticated with CHAP.

GFS2 cluster filesystem

GFS2 is a shared-disk cluster filesystem that uses the distributed lock manager and requires cluster quorum plus fencing; all nodes mount the same filesystem concurrently.

OCFS2 cluster filesystem

OCFS2 is a shared-disk cluster filesystem for Oracle and Linux; it uses a cluster stack, slot maps, and fencing, and supports concurrent mounts from multiple cluster nodes.

DLM Distributed Lock Manager

The distributed lock manager coordinates locks among cluster nodes for shared filesystems such as GFS2; lock state is maintained across nodes and released after node failure and fencing.

lvmlockd

lvmlockd coordinates LVM metadata changes in shared volume groups using sanlock or DLM; it prevents conflicting activation and is the modern replacement for clvmd.

SCSI-3 persistent reservation

SCSI-3 persistent reservations let cluster nodes register and reserve a LUN so only authorized nodes write; cluster agents use them to block I/O from a failed or partitioned node.

Ceph monitor

A Ceph monitor maintains the cluster map, including monitor, OSD, and placement group maps; monitors form a quorum, and an odd number, commonly three, is recommended.

Ceph OSD

A Ceph object storage daemon stores objects on a local disk or device and participates in replication, recovery, and backfill; OSDs report to monitors and serve client I/O.

CRUSH map

The CRUSH map describes storage devices, buckets, and failure domains and defines rules that map placement groups to OSDs without a central lookup table.

Ceph placement group

A placement group is a logical partition of a pool that groups objects for placement and replication; CRUSH maps each PG to a set of OSDs according to the pool rule.

Ceph RBD

RADOS Block Device exports a Ceph pool object as a virtual block device through librbd; clients can snapshot, clone, and map images, often with exclusive-lock enabled.

GlusterFS brick

A GlusterFS brick is an exported directory on a server that stores volume data; volumes are built from bricks, and clients access them through a FUSE or NFS mount.

DRBD protocol C

DRBD protocol C reports write completion only after the local and remote nodes have acknowledged the write; it provides synchronous replication but increases latency with distance.

DRBD split-brain

DRBD split-brain occurs when both nodes promote to Primary while disconnected and diverge; recovery requires choosing a source, discarding or merging data, and reconnecting.

RAID 1

RAID 1 mirrors identical data across two or more disks; it tolerates at least one disk failure per mirror, offers no capacity gain, and can improve read performance.

RAID 6

RAID 6 stripes data with two independent parity blocks and survives two disk failures; usable capacity is total minus two disks, and write performance is lower than RAID 5.

mdadm

mdadm creates, assembles, and manages Linux software RAID arrays; common operations include --create, --detail, --fail, --remove, --add, and monitoring /proc/mdstat.

Network bonding

Linux bonding combines NICs into one logical interface; active-backup uses mode 1, LACP uses mode 4 with 802.3ad, and miimon monitors link state for failover.

SMART

SMART monitors disk health attributes such as reallocated sectors and pending sectors; smartctl -a reads attributes, -H shows health, and -t long starts a self-test.

kdump

kdump reserves crashkernel memory and boots a capture kernel after a panic; it writes a vmcore file, usually under /var/crash, for later analysis with crash.

Hardware watchdog

A hardware watchdog timer resets the system if software fails to refresh it; systemd can manage it with RuntimeWatchdogSec, and softdog provides a software fallback.

NUT UPS monitoring

Network UPS Tools monitors a UPS and can trigger orderly shutdown when battery is low; upsmon defines monitored systems, and upsd serves UPS status.