Skip to main content

HA Topologies

Five supported shapes for an HA deployment. They differ in where the broker cluster lives and what platform runs it — the Linkiir side of the build is the same in all five, so choosing between them is an infrastructure decision, not a Linkiir one.

At a glance

TopologyServersLayoutChoose when
A: Windows52 Windows for Linkiir, 3 Linux for brokersYour estate is Windows-standard
B: Linux co-located3Brokers on all three, Linkiir on two of themLinux, and the fewest servers
B+: Linux separated52 for Linkiir, 3 for brokersLinux, with full role separation
C: KubernetesVariesLinkiir and brokers as workloads in your clusterKubernetes is already your standard
D: CloudVariesLinkiir across availability zones, managed broker and databaseYou are already on a public cloud

Every topology needs the same four things around it: shared storage, a log database with its own HA, a front door, and time synchronisation. See System Requirements.


Topology A: Windows

Five servers. Linkiir on Windows, brokers on Linux.

Choose it whenWindows is your standard platform for application servers
Servers2 Windows + 3 Linux
Shared storageSMB share. The Linkiir service account needs full control on it
Watch out forBrokers on Windows are supported but not recommended in production. Run Linkiir on Windows and the brokers on Linux, as drawn

This is the most common shape in hospital estates, where Windows is standard for applications but a small Linux footprint for infrastructure services is acceptable.


Topology B: Linux co-located

Three servers. The brokers run on all three; Linkiir runs on two of them.

Choose it whenYou run Linux and want the cheapest correct deployment
Servers3. The third runs brokers only, so it can be smaller
Shared storageNFS or a clustered filesystem
Watch out forMemory contention — a Linkiir server and a broker on one host compete. Size RAM for both
Do not run the front door on either Linkiir host

Losing that host would take the load balancer with it, so a Linkiir failure and a front-door failure become the same event. Put it on a separate host, or use a load balancer you already run.

This is the minimum correct HA deployment: three servers, plus the shared storage, database, and front door that most sites already have.


Topology B+: Linux separated

Five servers. The same as B, with the brokers moved onto their own hosts.

Choose it whenYou run Linux and want each tier patched, sized, and owned separately
Servers2 + 3
Shared storageNFS or a clustered filesystem
Watch out forNothing specific. This is the most conventional shape, and the easiest to hand to separate infrastructure teams

Prefer this over B when the messaging tier has a different owner, a different patch cycle, or a different growth curve from the application tier.


Topology C: Kubernetes

Linkiir and the brokers as workloads in a cluster you already operate.

Choose it whenKubernetes is already your deployment standard and your team runs it well
ServersHowever many nodes your cluster has. Linkiir is still fixed at two instances
Shared storageA volume that supports simultaneous access from both pods, with file locking
Watch out forPod identity. Because hostnames are not stable in a cluster, each instance needs a pinned identity — support configures this during the build
Kubernetes adds no availability here

It is no simpler than three Linux servers and no more scalable, because Linkiir is two instances either way. Choose it because it fits how you already deploy and operate software, not in the expectation of better availability.


Topology D: Cloud

Linkiir on virtual machines spread across availability zones, with managed infrastructure services around it.

Choose it whenYou are already on a public cloud and want its managed services doing the heavy lifting
Servers2 VMs, plus managed broker and database services
Shared storageManaged file storage, premium tier. Standard tiers are usually too slow for version-controlled project operations
Watch out forZone placement, storage tier, and whether the managed broker is a real cluster

Three specifics worth settling early:

  • Spread the two virtual machines across availability zones. Two VMs in one zone survive a host failure, not a zone failure.
  • Use a managed database with its own HA, in a multi-zone configuration.
  • Confirm the managed broker offering with Linkiir support. It must be a real broker cluster with replication and quorum, not only a protocol-compatible endpoint. Managed Kafka services from the major clouds and dedicated Kafka providers are both fine.

Choosing

If this is your constraintChoose
Windows-standard estateA
Fewest possible serversB
Separate teams own application and messaging tiersB+
Everything already runs in KubernetesC
Cloud-first, prefer managed servicesD
You need to survive losing a whole siteNone of these on their own — see Disaster Recovery

Once you have chosen, Planning Your Deployment lists what to have ready before the build.

Next