RIG

rig builds a disposable Kubernetes environment on your machine so you can explore how a system fits together without needing access to any of it. Its job is onboarding and exploration, not a production replica.

Most services in it are deliberately not real. What has to be faithful is the topology — the names, the ports, the dependency order, who can reach whom, and how it fails. The workloads themselves are noise. This is what makes the whole estate fit on a laptop: a real 20-service platform will not fit even once on 14 GB, but mocks are about 30 MB each, so three faithful copies do.

The only prerequisite

Docker. No curl, no jq, no python, no apt repositories to configure.

# then, in the environment directory:
make check     # is this machine ready? reports, never fixes
make deps        # install the pinned toolchain
make cluster up  # cluster + registry + addons; ports derive by themselves

Read make check before make deps. It never changes anything — it prints what it found and, at the end, the steps it cannot perform for you.

1 · make check

Asks whether this machine is ready. It changes nothing — it reports what it found and, at the end, the things only a human can do (anything needing sudo, or a Windows-side restart). Read it before installing anything; it is faster than discovering the same problems one failure at a time.

make check

2 · make deps

Installs the pinned toolchain — only what is missing — and tells you if its directory is not on PATH yet. Running it twice is safe.

make deps

3 · make cluster up

Builds the cluster, starts its registry and installs the profile's addons — there is nothing else to run first. It prints what the profile locks in before spending the time, because the kind config is fixed at creation and cannot be changed afterwards.

Re-running is safe and, more importantly, convergent: if a first attempt was interrupted before the CNI was installed, running it again finishes the job rather than reporting "already exists" and leaving every node permanently NotReady.

make cluster up                 # built-in defaults — no profile needed
make cluster up PROFILE=client  # after copying env.d/client.env.example: cached registry
make cluster reset              # destroy and rebuild — how an edited kind config takes effect

4 · make docs

Serves this page from a throwaway container. Works with no cluster and no toolchain, which is deliberate: these pages are the instructions for building everything else, so they cannot depend on it.

make docs

Checking on things

make cluster list

Every cluster on the machine, its memory cost and its port block. The usual reason a new one will not start is an old one you forgot about; make cluster free frees them without deleting.

make check

Short: host, toolchain, and whether this cluster fits, its ports, registry and addons. Details only appear when something needs attention; make check all prints every one.

make check mem

Memory in depth: what caps it, how far it really climbs, and on WSL the .wslconfig backup and restore.

Running more than one

Copy the directory, rename it, and run make cluster up. Cluster name, context, image tags and the port block all follow the directory name, so the second environment collides with nothing and neither one's teardown can reach the other.

cp -r rig ../platform-v2 && cd ../platform-v2
make cluster up

The installer is a container, not a shell script, for a specific reason: a stock slim Debian has no curl, no wget, no jq, no python3 and no CA bundle. A shell installer could not make a verified HTTPS request, let alone check one. The container carries its own toolchain, so the host needs nothing but Docker.

The cluster never runs inside that container. Everything it installs — kind, kubectl, tilt, jq — runs natively afterwards, so nothing pays a container tax during daily work.

Pinned and verified

Every tool is a single binary fetched at a pinned version and checked against a published SHA256. Node images are pinned by digest, so upgrading kind cannot silently move your Kubernetes version.

Not every machine should get cluster tooling

A managed or corporate-issued machine — the kind that holds the access you cannot get anywhere else — is not somewhere to install development tools by default. So the toolchain comes in two tiers:

tier	installs	for
core	kubectl, jq	talk to a cluster someone else runs
dev	+ kind, tilt	build clusters and hot-reload into them

make deps core   # kubectl and jq only — nothing that creates a cluster
make deps        # dev, the default

Testing in situ on a managed machine is still possible — install the dev tier deliberately when you need it. The point is that it should be a decision rather than a side effect of installing.

The documentation itself needs neither tier: make docs wants only Docker.

Air-gapped

make deps image full                       # bakes every binary into the image
docker save …-deps:full | gzip > rig.tgz
# carry that one file in, then:
docker load < rig.tgz && make cluster up PROFILE=offline   # from env.d/offline.env.example

Running several versions of a system at once means several clusters on one machine, not several machines. Everything that could collide is derived from the directory name:

cluster + context

acmebank/ builds acmebank on kind-acmebank.

port block

Ten ports from a hash of the name, in the 20000+ range — clear of 80, 443, 3000, 5432, 8000 and 8080.

registry + images

Named after the environment, so two copies never share one.

Two copies therefore never collide, and neither one's make cluster down can touch the other. make check shows the block; bash ctrl/ports.sh persist freezes it into ctrl/.env if you want it fixed rather than derived.

Configuration layers

Weakest first, later wins: built-in defaults → pinned versions → a profile, if you name one → ctrl/.env → the environment. So make cluster up PROFILE=<name> always beats every file.

example	registry	for
none	local	the built-in defaults; no profile needed
client.env.example	mirror	images through a corporate registry
offline.env.example	local	air-gapped
data.env.example	local	postgres, redis, airflow

LoadBalancer services

Real manifests use type: LoadBalancer, because a real cluster has one. On a bare local cluster those Services sit at EXTERNAL-IP <pending> forever, with no error anywhere — the deployment looks healthy and simply is not reachable.

The metallb addon fixes that, so the same manifests work here as upstream and nothing has to be rewritten to NodePort. Its address pool is derived from the cluster's Docker network at install time rather than hardcoded, because Docker picks that subnet and it differs between machines.

Networking

The cluster uses kind's built-in networking, which does enforce standard NetworkPolicy — verified against a no-policy control, not assumed. The widely repeated claim that it accepts policies and silently ignores them is out of date.

A pluggable CNI was tried and removed: it only added GlobalNetworkPolicy, policy tiers and egress-CIDR rules, none of which are needed yet, in exchange for a slower boot and one more thing that has to be right at creation time. Worth revisiting only when a policy the built-in cannot express actually comes up.

Memory

Every cluster is a running container tree whether you are using it or not. make cluster list shows what exists and what it costs; make cluster free stops the others without deleting them.

mode	what it does
none	images are built straight into the node
local	a registry container wired into the cluster
mirror	that container as a pull-through cache of the corporate registry
remote	no local container; pull direct with an imagePullSecret

mirror is what a locked-down network actually looks like: images originate from the corporate registry, you do not hammer it, and you keep working when the connection drops.

Reachability also depends on where you are: if the registry is only routable from a managed workspace, mirror and remote will not resolve from a laptop at all. That is what local and offline are for.

Each component is one of three things, and switching between them should be a one-line change rather than a rewrite:

real

Built from source and hot-reloaded. The thing you are actually working on — usually exactly one.

mock

A generic stub with canned responses. Everything you do not care about today.

remote

No pod at all: a Service of type ExternalName pointing at the real system. In-cluster DNS resolves identically, so callers never change.

The intended end state is that this diagram is generated from the running cluster rather than drawn by hand — so it becomes a report of what exists instead of a picture of what was once intended.

Tilt stops noticing file changes

Almost always inotify limits, and it fails silently — nothing errors, changes just stop being picked up. Defaults on WSL are far too low. make check reports it and prints the fix.

Cluster creation dies halfway with a port error

Docker reports failed to bind host port … address already in use partway through creating the cluster. Run make check first — it checks every port in this environment's block before anything is built.

Every node stays NotReady

Usually a cluster created with the default CNI disabled but the real CNI never installed — typically an interrupted first run. Just run make cluster up again: it converges rather than exiting early, and will finish the missing steps.

x509: certificate signed by unknown authority

Corporate CA trust has not reached one of the three places it needs to be. See Registry.

kubectl says the context does not exist

The cluster can exist while its context does not — a reset or a switched KUBECONFIG loses it. make cluster up detects this and re-exports the context.