Public Protocol
Platform

Element 13.12

Small clusters of boards

Sheet
13.12
Chapter
13 Digital infrastructure
State
Drafted
Release
16 November 2026
Editor
Contributed by Praecise as open research
Reviewers
To be named at review
Safety
S1, as the chapter
Licence
CC BY-SA 4.0
Resources
6

The element

Requirement

A community joins several small computers into one cluster only when it needs a service to survive the loss of one machine or to spread work across many, and it can keep the cluster running with the skills it has.

How to tell

The reason is written: availability (the service survives one node failing) or capacity (more work than one board can do). If availability, the control plane has an odd number of at least three server nodes, because quorum for n servers is (n/2)+1 (K3s high availability with embedded etcd). Each server node meets the stated minimum of 2 cores and 2 GB of memory, each agent 1 core and 512 MB (K3s requirements). Nodes run from SSDs: the K3s documentation warns that etcd is write-intensive and that SD cards and eMMC cannot handle the load (K3s requirements), and that embedded etcd may perform poorly on Raspberry Pis with SD cards (K3s high availability with embedded etcd). One node is switched off on purpose and the service keeps answering.

Options
  • No cluster. One machine with a tested backup (13.5) and a spare board. For most services at 30 adults this does the job with far less to keep.
  • Two machines, one standing by. One runs, one holds a copy and can take over by hand. Simple and well understood.
  • A lightweight Kubernetes cluster. K3s is a fully compliant Kubernetes distribution in one binary of less than 100 MB, made for edge, IoT and single-board ARM computers, and for air-gapped sites (K3s). With three server nodes and embedded etcd it tolerates one server failing (K3s high availability with embedded etcd).
  • A cluster of agents under one server. Gains capacity without availability: the single server is the single point of failure (K3s high availability with embedded etcd).

Storage. On Raspberry Pi 5 an NVMe SSD through an M.2 adapter (Raspberry Pi SSD). Steady power still matters: SSDs too can fail under power faults (Understanding the Robustness of SSDs under Power Fault).

Settled, open, contested

Settled. Availability needs at least three server nodes for quorum, and write-heavy datastores need SSDs (K3s high availability with embedded etcd, K3s requirements). Open. No source was found that measures how often small-board clusters at community scale fail, or what they cost to keep. Contested. Whether Kubernetes is worth learning for a small community. For: one way to run, move and update every service. Against: a large body of knowledge for a few services that one machine could carry.

The four stages

Learn: quorum and why an odd number of servers is needed (K3s high availability with embedded etcd). Design: write the reason; count nodes and storage (K3s requirements). Build: install K3s on SSD-backed nodes (K3s, Raspberry Pi SSD). Deploy: rehearse a node failure, back up the datastore and the applications (13.5), and log node temperatures (Raspberry Pi hardware documentation).

Resources at each stage

2 Design

Size and choose

3 Build

Build or obtain

4 Deploy

Install, run and measure

What changes between 30 adults and 300 people

At 30 adults a cluster is rarely worth its upkeep; a single machine and a spare are the usual answer. At 300 people services that must not stop, such as sign-in or the shared records, can justify three server nodes, kept by at least two people who can rebuild them.

Reference designs

Open for contributions Designs with drawings, specifications and a bill of materials. Add to it

Parts list

Every resource that supports this element. Those marked External carry the label “External resource. Not made or verified by Public Protocol”.

ResourceTypeOriginMaturitySafetyLicence, as stated at its source
K3s high availability with embedded etcdGuideExternalEstablishedS1Not stated on the page opened
K3s requirementsDatasheetExternalEstablishedS1Not stated on the page opened
K3sToolExternalEstablishedS1Not stated on the page opened
Raspberry Pi hardware documentationDatasheetExternalEstablishedS1Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0); some content from the eLinux wiki under CC BY-SA 3.0 Unported
Raspberry Pi SSDDatasheetExternalEstablishedS0Not stated on the page opened
Understanding the Robustness of SSDs under Power FaultPaperExternalReplicatedS0Not stated on the page opened; USENIX states that its proceedings are free to everyone (open access)

Depends on, and completes

Depends on 13.1, 13.5.

It is one of 25 elements that complete Digital infrastructure.

The standing questions

Who can afford this: three boards with SSDs cost several times one. Who does the work: at least two keepers who understand the cluster. Who decides: which services must survive a failure, and so justify a cluster, is a community decision (15.2).

Environments

No source found on clusters of small boards in particular climates; the operating ranges of each board apply (see the element on choosing an edge computer).

Safety

State and evidence

Drafted

An editor has written the element's text from the resources listed under it, citing them. It is checked when it is reviewed.

Evidence from a real place or system is recorded here, with what did not work. What the states mean

History