Linux HA Cluster Overview

One, high availability principle

System availability = Average trouble-free time/(mean time to repair + average no downtime)

Improve system availability:

Reduced mean time to repair (redundancy mechanism)

Extended mean time-to-average trouble-free

Methods of improving system availability:

HA cluster:

Manually switch

Automatic switching

Cluster switching:

Failover: Failover

failback: Fix Transfer

Cluster resources:

Vip:float IP

Ipvs rules

Constraint type:

Position constraint (location)

Permutation constraints (colocation)

Sequential constraint (order)

Solution:

Vrrp+script:keepalived

Ais:

Heartbeat

Corosync

Cman (Openais)

Type of service:

No Ha-aware

Ha-aware

Framework of HA:

Ha tiering

Messaging Layer: Basic transaction tier, delivering cluster transactions and heartbeat information

Providers: Heartbeat v1, V2, v3 corosync (OpenAIS) Cman (OpenAIS)

Crm:cluster Resource Manager

Provider: Heartbeat v1: Haresources with Resource Manager (Configuration Interface: Configuration file, file name also called haresources)

Heartbeat v2: Bring your own Resource Manager CRM (each node runs the CRMD process, configures the interface: command-line client Crmsh,gui client Hb-gui)

Heartbeat v3 = heartbeat + pacemaker + Cluster-glue

Packmaker:

Cli:crm (SuSE), PCs

Gui:hawk, LCMC, PACEMAKER-MGMT

Cman + Rgmanager:

Resource group Manager:failover Domain, node priority

To configure the interface:

Clustat, Cman_tool

Conga:luci+ricci

Lrm:local Resource Manager

Provided by CRM

Ra:resource Agent

Heartbeat legacy:heartbeat and traditional types, usually scripts in the/etc/ha.d/haresources.d/directory;

LSB:/etc/init.d/*

OCF (Open Cluster Framework):

Provider

STONITH:

Quorum: Statutory number of votes (more than half of the total number of votes)

In order to determine the cluster splitting scenario, some nodes can continue to run in cluster mode;

With quorum has legal votes, cluster normal

With out quorum no legal votes, the cluster has a brain fissure

Arbitration equipment:

Ping node

Ping Node Group

Quorum Disk:qdisk

Resource Isolation Mechanisms:

Node Level: STONITH

Power switch

Service Hardware Management Module

Resource level:

Stopped

Ignore

Freeze

Suicide

Second, the working model of a high-availability cluster for CentOS or RHEL systems:

CentOS 5:

Rhcs:cman+rgmanager

Choose a third-party scenario: Corosync+pacemaker, Heartbeat (v1 or v2), keepalived

CentOS 6:

Rhcs:cman+rgmanager

Corosync + Rgmanager

Cman + Pacemaker

Heartbeat v3 + Pacemaker

Keepalived

Third, the premise of configuring a Highly available cluster: (Take the heartbeat of two nodes as an example)

1. Time must be kept in sync

Using an NTP server

2. Nodes must have names to communicate with each other

Resolving node names

Edit the hostname that is used in the/etc/hosts cluster with the hostname ' uname-n ';

3. Ping node

Only the even-numbered nodes are required;

4, SSH key authentication for communication;

Linux HA Cluster Overview