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: High Availability Cluster Management

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

All guide sections

High Availability Cluster Management

High Availability Cluster Management

This domain tests the cluster control plane: Corosync membership and quorum, Pacemaker scheduling from the CIB, resource agents, constraint scoring, fencing, and operations with pcs and crm_mon.

Core rules

  • Corosync carries membership and messaging; Pacemaker places resources. Quorum is policy: a two-node cluster needs explicit handling such as Corosync two_node: 1 with wait_for_all: 1, or a quorum device, to fail over safely.
  • The CIB is desired state. Change it with pcs or crm configure, not by hand; inspect with cibadmin --query.
  • Prefer OCF agents. Give every long-running resource a monitor operation with realistic interval and timeout so a hung service is detected and fenced.
  • Constraints encode dependencies: order for sequence, colocation for togetherness, location for preference. INFINITY is mandatory, finite scores are preferences, -INFINITY is a ban.
  • STONITH is mandatory for shared data. If the cluster cannot prove a node is dead, it must fence before starting the resource elsewhere.

Worked example

A two-node web cluster needs a floating address before httpd. Create WebIP as ocf:heartbeat:IPaddr2 and WebServer as systemd:httpd, then: pcs constraint order WebIP then WebServer pcs constraint colocation add WebServer with WebIP INFINITY pcs constraint location WebServer prefers node1=100 Reasoning: ordering starts the address first and stops httpd first; colocation prevents httpd from running without its address; the finite location score biases placement without blocking failover. Configure and test STONITH first.

Traps

pcs resource move leaves a temporary location constraint; clear it with pcs resource clear or the resource never fails back. Do not leave stonith-enabled=false in production. Default quorum can stop resources after one node of two dies. Check pcs status --full, pcs constraint --full, and journalctl -u pacemaker before changing scores.