Skip to main content

High Availability

Linkiir High Availability runs two Linkiir servers as an active / warm-standby pair. One serves users and runs your integrations. The other stays up, idle, watching. If the active server fails, the standby takes over automatically in about half a minute, and no messages are lost.

These pages explain what HA is, what it costs you in infrastructure, and which shape of deployment fits your estate — so you can plan and budget before anyone installs anything.

These pages cover the concepts, the design, and how to use HA once you have it. Using the HA Settings and Operating an HA Pair are the two you will come back to.

The initial build is done with Linkiir support

Installing the pair and configuring your load balancer are done with Linkiir support, against your environment. Once it is built, everything you do with HA day to day is in the Linkiir UI and documented here.

When you have chosen a topology, email support@linkiir.com with your platform, your chosen topology, and your message volumes.

Where to start

You want toRead
Learn the vocabulary before a design discussionHA Terminology
Know what HA costs to licenseHA Licensing
Understand how it works, and see the designHA Architecture
Size the servers and storageSystem Requirements
Choose the shape of your deploymentHA Topologies
Understand how HA relates to backups and DRBackup, Restore, and Disaster Recovery
Prepare for the buildPlanning Your Deployment
Use the High Availability page — enable HA, name the servers, step downUsing the HA Settings
Run the pair — test a failover, patch without downtime, monitor itOperating an HA Pair

What HA protects against

FailureCoveredHow
The Linkiir process crashes or hangsYesThe standby detects that the active server is no longer serving, and promotes itself
The active server dies — power, hardware, operating systemYesSame
Planned maintenance and patchingYesMove the role to the other server, patch the idle one, move it back
One message broker node lostYesA three-node broker cluster keeps a majority and keeps serving
A database failoverDepends on your databaseLinkiir reconnects. Your database platform's own HA does the work

What HA does not protect against

HA is not a backup

The two servers share one working directory and one log database. That sharing is what makes failover fast — and it means anything that damages the data damages it for both servers at the same instant.

Not coveredWhyWhat you need instead
A deleted project, or a bad configuration changeBoth servers read the same directoryBackups
Database corruptionThere is one database, sharedDatabase backups
Malicious or ransomware deletionIt reaches shared storage tooOff-site, versioned backups
Shared storage failureIt is the single resource both servers depend onStorage-level redundancy, plus backups
Loss of the whole siteBoth servers are in one siteDisaster recovery — a second site

HA gives you uptime. Backups give you recoverability. A production deployment needs both.

What a failover looks like

EventTime before the pair is serving again
Unplanned failure — the server or process diesAbout 18 seconds, plus a few seconds for your load balancer
Planned hand-over — you move the role deliberatelyUnder 10 seconds
Messages lostNone. In-flight messages wait in the broker and are processed by the promoted server

Three behaviors that surprise people, all of them deliberate:

  • There is no primary to nominate. The two servers are interchangeable. Whichever one currently holds the role is the active one, which is why a failover needs no decision from you.
  • Failback is manual. A recovered server rejoins as standby and does not take the role back on its own. An automatic failback would be a second unplanned interruption.
  • Users stay signed in. A session created on one server is accepted by the other, so nobody is logged out when the role moves.

What a complete deployment contains

PartWho provides it
Two Linkiir serversLinkiir release, installed twice
Three-node message broker clusterYou, or a managed offering
Front door that sends clients to the active serverYou
Shared working directoryYou
Log database with its own HAYou

You supply the storage, the database, and the front door. Each is a mature product category with its own high-availability story, and Linkiir works with them rather than reimplementing them.

Two numbers that are not negotiable

Three broker nodes, never two. The broker cluster needs a majority of its nodes reachable to accept writes. With two nodes the majority is two, so losing either one halts the cluster. A two-node cluster costs more than a single node and buys no availability. Use three, or five for very large deployments.

Two Linkiir servers, not three. Only one can be active at a time, so a third adds cost and no availability.

Next