Skip to main content

Planning Your HA Deployment

What to settle before an HA build, and what the build itself involves. Work through this page, then bring the answers to Linkiir support.

The build is done with Linkiir support

Installing the pair, enabling HA, configuring your front door, and testing failover are done together with Linkiir support, against your environment. That is deliberate: the parts most likely to go wrong are your shared storage, your database, and your load balancer, and they are quicker to get right with someone who has done it before than from a written procedure.

Email support@linkiir.com with the completed planning decisions below.

Step 1 — Confirm HA is what you need

You want toYou need
Survive a server or process failure with seconds of interruptionHA
Recover a deleted project or a bad changeBackups
Survive losing the whole siteDisaster recovery
Process more messages per secondNeither — a single correctly sized server handles a few thousand messages per second. See System Requirements

HA is worth doing when an outage of minutes to hours is unacceptable for your feeds. If your feeds tolerate a maintenance window and your recovery need is really about data, invest in backups and testing them first.

Step 2 — Choose a topology

Pick from the five in HA Topologies. The decision is driven by your platform and how many servers you can run, not by anything in Linkiir — the Linkiir side of the build is identical in all five.

Record: the topology, and where each server will live.

Step 3 — Decide the four things around the pair

These are yours to provide, and they are where HA deployments succeed or fail.

DecisionWhat to recordReference
Shared storagePlatform, the absolute path both servers will use, measured latency, whether file locking worksSystem Requirements
Log databasePostgreSQL or MS SQL, its HA arrangement, and confirmation both servers can reach itSystem Requirements
Front doorWhich load balancer or virtual IP, and who administers itSystem Requirements
Broker clusterThree nodes, self-hosted or managed, and which offeringSystem Requirements
Validate shared storage before the build date

Confirm the same absolute path on both servers, write access for the service account, that a file created on one is visible on the other, latency under 2 ms, and working file locks. If any of that is unresolved, fix it first — the build cannot succeed over storage that does not meet these requirements.

Step 4 — Decide how integration traffic follows a failover

Your load balancer covers the UI and API. Inbound HL7 and HTTP feeds arrive on node listener ports, and they need their own answer.

OptionWhat it means for your partners
A virtual IP covering the listener portsNothing changes on their side
A load-balancer rule per listener portNothing changes on their side
Partners reconfigured after a failoverManual, slow, and not a real HA posture

Decide this now, with names against it. It is the most commonly missed part of an HA rollout, and it is much cheaper to design in than to retrofit.

Step 5 — Arrange licensing

One Enterprise license with the HA feature covers the pair, and it must be issued for an HA pair. Request or re-issue it as part of planning rather than on the build day. See HA Licensing.

Step 6 — Plan backups alongside HA

HA does not back anything up. Before go-live, have all three configured: instance backup, a remote per project, and database backups — with destinations reachable from both servers. See Backup, Restore, and Disaster Recovery.

What the build involves

So you know what to expect and can schedule the right people.

StageWhat happensWho needs to be there
ValidationShared storage, database reachability, broker health, and time sync are confirmedInfrastructure and database
InstallThe same release is installed on both servers, pointed at the one shared directoryServer administrators
ConfigureDatabase and license are set once, and read by both serversLinkiir administrator
Enable HAReadiness checks are cleared and HA is switched on. No restart, and running workflows are not interruptedLinkiir administrator
Front doorHealth check, timeouts, and session handling are configured and verifiedLoad-balancer administrator
Failover testA planned hand-over, then an unplanned failure, both measuredEveryone above

Two things worth knowing in advance: enabling HA needs no restart and does not interrupt running workflows, and the second server does not need to be told about the first — both register themselves in the shared directory and find each other.

The readiness checks

Linkiir will not let HA be enabled until all seven of these pass, so a half-built pair cannot be switched on. Knowing what they are tells you what to have ready.

CheckWhat must be true
Working directory writableThe shared working directory is writable by the service account
Log backend supports concurrent writersThe log database is PostgreSQL or MS SQL
License includes HAEnterprise with the HA feature
Node identity configuredEach server has a stable identity
Pair has exactly two nodesBoth servers registered, and no third installation present
Peer reachableEach server can reach the other
Kafka has more than one brokerThe three-node broker cluster is reachable

Before you go live

Infrastructure

  • Three broker nodes, healthy, with replication factor 3
  • Log database is PostgreSQL or MS SQL, reachable from both servers, with its own HA
  • Shared directory at the same absolute path on both, file locking verified, latency under 2 ms
  • Time synchronised on both servers

The pair

  • Both servers on the same release, against the same working directory
  • The license shows Enterprise with HA among its features
  • All seven readiness checks passing, and exactly one server showing ACTIVE

Front door

  • Health-checking correctly, with only a success response treated as healthy
  • Session affinity off, session cookie passed through unmodified
  • Idle timeout of at least 65 seconds, response buffering off
  • Verified it never routes to the standby
  • Not running on either Linkiir server

Integration traffic

  • A decision recorded for how inbound feeds follow a failover

Tested

  • A planned hand-over completed in under 10 seconds
  • An unplanned failure recovered in under a minute
  • A signed-in session survived the failover
  • No duplicate log records afterwards
  • One broker node stopped, and the cluster kept serving
  • The recovered server rejoined as standby and did not take the role back

Understood by the team

  • HA is not a backup, and backups are configured for instance, projects, and database
  • Failback is manual by design
  • Who does what after an unplanned failover, and who to call

Adding HA to an existing installation

You can start with a single server and add HA later, without rebuilding your projects. It involves:

  1. Moving the working directory onto shared storage.
  2. Migrating the log database from SQLite to PostgreSQL or MS SQL, if it is not there already.
  3. Adding the second Linkiir server, the three-node broker cluster, and the front door.
  4. Having the license re-issued for a pair.

Steps 1 and 2 need a planned downtime window. Plan the migration with Linkiir support — email support@linkiir.com with your current setup and target topology.

Next