The beds name images the way a machine would find them

Twenty-eight integration tests each carried their own copy of the same two helpers,
which pointed a manifest and the substrate bundle at whatever the lab's registry had
assigned. They now share two in the harness, and the difference is the point: ours is
rewritten to the ID the machine holds it under, and everything else is left exactly as
written so the machine pulls it.

**The substrate bundle is where the fiction was most load-bearing.** mesh-host's
`examples/substrate-first-node.lock` pins all three of its images at
`192.0.2.250:5000/…`, which is the address the lab's registry served from — it was
written for a target, and the target was the lab. Two of those are ordinary third-party
images and become the digests mesh-catalog's own postgres and lavinmq modules pin, so
the substrate's store and broker are literally the images the mesh runs. mesh-control
exists in no registry at all and becomes the ID the machine was handed. **The bundle
itself should be fixed in mesh-host and this substitution deleted with it.**

Beds that wrote a manifest by hand named an image by repository and let the rewrite
supply a digest. There is nothing to supply one now, so `onTheMachine` refuses an
unpinned reference and hands back the digest the catalogue pins — a bed runs the image
the mesh ships, and a bed that drifts from the catalogue is testing a different
postgres.

Three beds took a third-party image out of the raised list, which no longer contains
one: certificates (pebble), objectstore (minio and its client) and provisioner
(postgres) now name theirs and pull it. builds and mesh publish into the MESH's own
artifact store — the `registry` module's image, on the node, on 5000 — rather than into
scenery the lab raised. That is a different claim, and only one of them exists in
production.

New unit tests cover what a full raise would otherwise be the only way to check: the
routes an egress machine gets (that its gateway is still the path to the rest of the
scenario, that a range with no path is unreachable rather than leaked to the uplink,
that each family gets its own next hop), which machine is handed which of our images,
and the `images:` rule that refuses a third-party entry. The "shipped scenarios are
valid" test now loads every scenario rather than two of them.

Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
This commit is contained in:
2026-09-10 23:16:41 +02:00
parent 5c91c0ecd2
commit 675facdb0d
40 changed files with 898 additions and 705 deletions
+106
View File
@@ -9,11 +9,117 @@
*/
import assert from "node:assert/strict";
import { readFileSync } from "node:fs";
import { isReachable, pools, supportedDrivers } from "../../src/incus/client.ts";
import { destroy, list } from "../../src/lifecycle/operate.ts";
import { diagramFromLive } from "../../src/diagram/from-live.ts";
import { duplicateAddresses, describeConflicts, type Held } from "../../src/lifecycle/invariants.ts";
import type { Scenario } from "../../src/declaration/types.ts";
import { isMeshBuilt, pinnedInto, referenceFor, repositoryOf, type HeldImage } from "../../src/pinning.ts";
// --- the substrate bundle, and what its three images are on a real machine ---------------------
/**
* The example bundle in mesh-host names a registry that no longer exists.
*
* `examples/substrate-first-node.lock` was written **for a target**, and the target was the lab: it
* pins `192.0.2.250:5000/…` because that is where the registry the lab used to raise served from.
* That registry is gone, so those three references name nothing.
*
* Two of them are ordinary third-party images and belong to the internet. Rather than invent
* digests here, they are the ones the mesh's own modules already pin — mesh-catalog's `postgres`
* and `lavinmq` — so the substrate's store and broker are literally the images the mesh runs. The
* third, mesh-control, exists in no registry at all and becomes the ID the machine holds it under.
*
* **The bundle itself should be fixed in mesh-host**, and this substitution deleted with it. It is
* here because the file lives in another repository and because a fixture that lies about where an
* image comes from is exactly what this change is removing.
*/
const UPSTREAM_STORE =
"postgres@sha256:7456ef82e5f5bc43d997f4781bbd7c0d6389bff397564649a356e206ba473aee";
const UPSTREAM_BROKER =
"cloudamqp/lavinmq@sha256:3eb54c12916d700a978c2ea86e6362cd4974b0e3189508718006d4e6d341246b";
/**
* The substrate bundle as a machine should receive it.
*
* Third-party references become upstream ones, which the machine pulls over its uplink; ours
* become the ID the machine was handed. Nothing points inside the scenario any more, which is the
* whole of this change: what the bed proves about a bootstrap is now what would happen anywhere.
*/
export function substrateBundle(path: string, held: HeldImage[]): string {
let text = readFileSync(path, "utf8");
text = text.replaceAll(/[A-Za-z0-9_.:-]+\/postgres@sha256:[0-9a-f]{64}/g, UPSTREAM_STORE);
text = text.replaceAll(
/[A-Za-z0-9_.:-]+\/cloudamqp\/lavinmq@sha256:[0-9a-f]{64}/g, UPSTREAM_BROKER);
return pinnedInto(text, held);
}
/**
* The upstream reference for a third-party image, as the mesh's own catalogue pins it.
*
* A bed that writes a manifest by hand still has to name an image exactly — mesh-host refuses a
* tag, and rightly (novox/hq ADR 0006). While the lab had a registry the beds sidestepped that by
* naming a repository and letting the rewrite supply a digest; there is nothing to supply one now,
* so the digest has to be written down.
*
* These are the digests mesh-catalog's own modules pin, taken from `mesh-catalog/modules/*` — so a
* bed runs the image the mesh runs, and a bed that drifts from the catalogue is a bed testing a
* different postgres than the mesh ships.
*/
const UPSTREAM = new Map<string, string>([
["alpine", "alpine@sha256:28bd5fe8b56d1bd048e5babf5b10710ebe0bae67db86916198a6eec434943f8b"],
["baserow/baserow", "baserow/baserow@sha256:834424a10413798567f76428f255dc259445b7f8dcec56598c05b4073bb2a124"],
["cloudamqp/lavinmq", "cloudamqp/lavinmq@sha256:3eb54c12916d700a978c2ea86e6362cd4974b0e3189508718006d4e6d341246b"],
["eclipse-mosquitto", "eclipse-mosquitto@sha256:6f8d8a947c506f8a2290ec65cd4bd2bc7cb4d43fb5f6271f861cb013e2ef9797"],
["ghcr.io/umami-software/umami", "ghcr.io/umami-software/umami@sha256:fa32d116cf20cad52cbc3fad9a63b46e7fa02299d8f967168eb453d49c476b4a"],
["letta/letta", "letta/letta@sha256:1d2e0692514287c5ed1a483e14e16ed945f8632d315539f5e66373bb7d7c471b"],
["lscr.io/linuxserver/radarr", "lscr.io/linuxserver/radarr@sha256:119aaa4a4f7349bcd2a136c5373a0d7925b5479915c7dfe0c0ad352db2a6d438"],
["lscr.io/linuxserver/sonarr", "lscr.io/linuxserver/sonarr@sha256:c19aa4ecdf03d73e1d5c901da33744cb7eb4d921f89bafed1ca264601d7fa224"],
["lscr.io/linuxserver/unifi-controller", "lscr.io/linuxserver/unifi-controller@sha256:fcd5d8b13a77a588c79c1b49e5fc9ad08115aa3bb1a3576c589c64908a68845f"],
["minio/minio", "minio/minio@sha256:aefec8a86702aff0b0dcfdd9284bd7ab7c5631cbf9be63275799e6edcb30dfa2"],
["mongo", "mongo@sha256:e3fa459b4f4b72f3257c67a23c145e250b8b5700f033860392c68539b998bbe3"],
["ollama/ollama", "ollama/ollama@sha256:32931b46719f673c05fdbaa81ccb26da18ea4a1c57590a754874ab28ba269eb2"],
["postgres", "postgres@sha256:7456ef82e5f5bc43d997f4781bbd7c0d6389bff397564649a356e206ba473aee"],
["redis", "redis@sha256:ff02b58f971e7d7d156a1267e283fcbbeee91773b6aa36c49dac28ecfe28eadf"],
["registry", "registry@sha256:a3d8aaa63ed8681a604f1dea0aa03f100d5895b6a58ace528858a7b332415373"],
["synesthesiam/marytts", "synesthesiam/marytts@sha256:45970ecb3e21a2981c66c60563a70cf00be8e95c02565e7d74b3a73dcec7db2c"],
]);
/**
* What a manifest's image reference becomes on the machine.
*
* Three cases, and the middle one is the whole change:
*
* - **Ours** becomes the ID the machine holds it under. Nothing serves it, and nothing needs to.
* - **Anything already pinned by digest** is returned exactly as written. The machine pulls it
* from the internet, over its uplink, which is what a real machine does and what the lab spent
* a long time serving from a registry of its own instead.
* - **A bare repository or tag** is one of ours in spirit — a bed naming an image by hand — and
* is given the digest the catalogue pins. A tag would be refused by mesh-host anyway, and
* refusing here says why rather than failing on the machine.
*/
export function onTheMachine(reference: string, held: HeldImage[]): string {
if (isMeshBuilt(reference)) {
const found = referenceFor(held, repositoryOf(reference));
assert.ok(
found,
`nothing loaded ${reference} onto the machines. They hold:\n ` +
held.map((i) => `${i.repository} ${i.reference}`).join("\n "),
);
return found;
}
if (reference.includes("@sha256:")) return reference;
const upstream = UPSTREAM.get(repositoryOf(reference));
assert.ok(
upstream,
`${reference} is not pinned and this harness does not know an upstream digest for it. ` +
`Add the one mesh-catalog pins, or write the reference out in full — a tag moves, and the ` +
`host refuses one.`,
);
return upstream;
}
export interface Capability {
usable: boolean;