novox/mesh-lab
3
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

The lab had a registry that production does not, so it tested a fiction

Browse Source 
The lab raised a `registry` VM, pushed ~73 images into it from the workstation, and
rewrote every manifest reference — third-party ones included — to point at it. No
production mesh has such a thing. So every bed proved that a machine could fetch an
image from a registry that exists nowhere else, and the bootstrap problems that only
appear when a machine has to fetch for itself went unfound.

What replaces it is the two things that are true in the world:

**Public images come from the public internet.** mesh-lab already created a NAT'd
uplink for exactly this and attached it to any machine declaring `egress`; no scenario
ever declared it. They do now, and third-party references are left exactly as the
catalogue writes them.

**The mesh's own images have no registry and never will.** mesh-control, mesh-builder,
mesh-route-proxy and the per-module runtimes are built from source and exist in no
registry. A machine gets them the way an operator's machine does — they are built here
and loaded onto it — and is then named by the digest of its own image configuration,
which mesh-host now accepts as "an image this machine already holds".

`images:` therefore means only *ours*, and a third-party entry is refused rather than
quietly loaded: otherwise the fiction returns one convenient line at a time. It is
per-machine as well, because "everything, everywhere" was never a description of
anything real — handing whole-mesh-full's union to its two 30GiB workstations would
fill the disk with runtimes nothing on them will start.

**The uplink and the declared gateway would have fought, silently.** A gateway container
and the transit router reach the scenario and nothing else; a default route through
either is a black hole for anything outside, and it beats the uplink's DHCP route on
metric. So a machine with egress states the scenario's ranges explicitly — through the
same gateway or transit it would have defaulted to, so the overlay-across-NAT path is
unchanged — and leaves the default to the uplink. A range with no path inside the
scenario becomes `unreachable` rather than falling through: 192.168.1.0/24 is an
ordinary private range in fact, and letting it escape would put scenario traffic on
whatever network the workstation is sitting on. `scenarioRoutesFor` is pure and tested,
because a decision only a full raise could check is one nobody checks.

The registry-reachability check the raise gained earlier is kept, pointed at the real
thing: every machine with egress must resolve a name and reach the internet before the
raise says it finished. Same failure it was written for — a raise that returns, an apply
that dies on its first pull, an instance left a bare shell — now guarding the path that
actually carries.

The base image's trust of the documentation ranges as plain-HTTP registries STAYS. It
was never only for the lab's registry: the mesh has one of its own, the `registry`
module, serving artifacts to the whole mesh over plain HTTP from whatever node runs it.

Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
This commit is contained in:
jschoubben
2026-09-10 23:16:05 +02:00
parent 0a0c57b610
commit 751948f0f9
13 changed files with 558 additions and 602 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/cli.ts
+6 -4
View File
@@ -233,10 +233,12 @@ async function main(): Promise<void> {
const seconds = ((Date.now() - started) / 1000).toFixed(1);
console.log(`\nraised ${raised.instanceId} in ${seconds}s — ${raised.machines.length} machines usable`);
if (raised.images.length > 0) {
// Printed because this is what a declaration pins, and it is not knowable until the
// scenario has been raised — the digest belongs to this registry.
console.log(`\nimages served, pinned by digest:`);
for (const image of raised.images) console.log(` ${image}`);
// Printed because this is what a declaration names them by, and it is not knowable until
// the image has been built — an image ID is the digest of its own configuration.
console.log(`\nthe mesh's own images, as the machines now hold them:`);
for (const image of raised.images) {
console.log(` ${image.requested} → ${image.reference}`);
}
}
if (scenario.snapshot) {
const took = await snapshot(raised.instanceId, scenario.snapshot);
src/declaration/parse.ts
+3
View File
@@ -41,6 +41,9 @@ function normaliseMachine(raw: unknown): Machine {
if (machine["memory"] !== undefined) result.memory = String(machine["memory"]);
if (machine["cpus"] !== undefined) result.cpus = Number(machine["cpus"]);
if (machine["disk"] !== undefined) result.disk = String(machine["disk"]);
// Absent and empty are different: absent means "all of the scenario's images", an explicit empty
// list means "none". A machine that runs nothing of ours should be able to say so.
if (machine["images"] !== undefined) result.images = toList(machine["images"]);
return result;
}
src/declaration/types.ts
+25 -4
View File
@@ -110,6 +110,19 @@ export interface Machine {
*/
memory?: string;
cpus?: number;
/**
* Which of the scenario's `images:` this machine is handed.
*
* Absent means all of them, which is right for a one-machine bed and wrong for a mesh: a
* workstation running one small module does not want forty runtimes copied onto a 30GiB disk.
* That is not a lab economy, it is what is true — an operator's machine holds the images its own
* modules need, because somebody put them there.
*
* Every entry must appear in the scenario's `images:`. Naming one that does not is refused
* rather than ignored, because a machine silently missing an image fails much later, inside an
* apply, as a container that will not start.
*/
images?: string[];
/**
* Root disk size, e.g. "60GiB". Left unset, the VM uses the storage pool's default, which is
* enough for a handful of modules. A broad install that stocks many runtime + service images
@@ -142,11 +155,19 @@ export interface Scenario {
policy?: Policy[];
place?: Placement;
/**
* Container images this scenario needs inside it.
* **The mesh's own images** — the ones that exist in no registry and are put onto a machine by
* whoever built them.
*
* A sealed machine cannot reach a registry, so the lab raises one on a public segment and
* serves these from it. Written as tags — the digest a declaration pins is the one THIS
* registry assigns, and it is reported when the scenario is raised.
* mesh-control, mesh-builder, mesh-route-proxy, the per-module runtimes and the provisioners are
* built from source and published nowhere. A machine gets them the way an operator's machine
* does: they are built on the workstation, loaded onto the machine, and named by the digest of
* their own image configuration. Written as tags, because a tag is what `docker save` can
* export; what a declaration then pins is the image ID, reported when the scenario is raised.
*
* **Third-party images do not belong here.** postgres, gitea, the mailu stack and everything
* else are pulled from the internet over a machine's `egress` uplink, exactly as they are in
* production. The lab used to serve them from a registry of its own, and that registry did not
* exist anywhere else — so every bootstrap problem it papered over went unfound.
*/
images?: string[];
/** Name the state once placement finishes, so a run can return to it. */
src/declaration/validate.ts
+22 -12
View File
@@ -15,6 +15,7 @@
import type { Scenario, Segment } from "./types.ts";
import { contains, familyOf, parseAddress, parseCidr, type Cidr } from "./net.ts";
import { isMeshBuilt } from "../pinning.ts";
/** RFC 5737 and RFC 3849. The only addresses guaranteed never to route on the real internet. */
const DOCUMENTATION_RANGES = [
@@ -315,29 +316,38 @@ export function validate(scenario: Scenario): void {
}
}
for (const image of scenario.images ?? []) {
const declaredImages = scenario.images ?? [];
for (const image of declaredImages) {
if (!image.trim()) {
problems.push("images: an empty entry names nothing");
} else if (image.includes("@sha256:")) {
// The digest a declaration pins is the one the LAB's registry assigns, which is not
// knowable before the scenario is raised. Naming an upstream digest here would pin
// something this registry will never serve.
// A tag, because a tag is what `docker save` exports. The reference a declaration ends up
// using is the image's own ID, which is not knowable until the image has been built.
problems.push(
`images: '${image}' is pinned by digest. Name it by tag — the lab's registry assigns ` +
`its own digest and reports it when the scenario is raised`,
`images: '${image}' is pinned by digest. Name it by tag — what a declaration uses is the ` +
`ID of the image loaded onto the machine, reported when the scenario is raised`,
);
} else if (!isMeshBuilt(image)) {
// **The rule that replaced the lab's registry.** Anything with somewhere to be fetched from
// is fetched from there, over the machine's uplink, exactly as in production. Serving it
// from inside the scenario instead is what hid the bootstrap faults this lab exists to find.
problems.push(
`images: '${image}' is not one of the mesh's own images, so nothing loads it. It is ` +
`pulled from the internet by the machine that needs it — give that machine 'egress: true' ` +
`and delete this line. Only mesh-* images, which exist in no registry, are placed by hand`,
);
}
}
if ((scenario.images ?? []).length > 0) {
const hasPublicV4 = Object.values(scenario.segments)
.some((s) => s.kind === "public" && s.cidr.some((c) => !c.includes(":")));
if (!hasPublicV4) {
for (const [name, machine] of Object.entries(scenario.machines)) {
for (const image of machine.images ?? []) {
if (!declaredImages.includes(image)) {
problems.push(
"images: this scenario declares images and has no public IPv4 segment to serve them " +
"from. The registry stands in for the outside world, so it sits on a public segment",
`machines.${name}.images: '${image}' is not in this scenario's images:. A machine can ` +
`only be handed one of the images the scenario says it has`,
);
}
}
}
if (problems.length > 0) throw new DeclarationError(problems);
}
src/lifecycle/address.ts
+119
View File
@@ -180,6 +180,20 @@ function withPrefix(scenario: Scenario, segment: string, address: string): strin
*
* The router's inside address is the first host address of the range, chosen rather than
* declared because a scenario has nothing to say about it.
*
* **A machine with `egress` is routed differently, and it has to be.** The scenery inside a
* scenario — a gateway container, the transit router — reaches the scenario and nothing else: it
* has no route to the real internet, and never will, because it exists to reproduce a household
* router rather than to be one. So a default route pointing at it is a black hole for anything
* outside, and it wins over the uplink's DHCP route on metric. A machine that must pull an image
* would then sit there failing, with a default route that looks perfectly reasonable.
*
* So an egress machine keeps the uplink as its default and gets an EXPLICIT route to every other
* segment in the scenario, through the same gateway or transit it would otherwise have defaulted
* to. Where there is no such path, the range is made `unreachable` rather than left to fall
* through: 192.168.1.0/24 in a scenario is a documentation range in spirit but an ordinary private
* one in fact, and letting it escape to the uplink would put scenario traffic on whatever network
* the workstation happens to sit on.
*/
export async function applyDefaultRoutes(
scenario: Scenario,
@@ -195,6 +209,13 @@ export async function applyDefaultRoutes(
// segment routes through transit instead — otherwise it can reach its own network and
// nothing else, which is not what being on the internet means.
const behind = spec.at.find((a) => scenario.segments[a.segment]?.gateway);
if (spec.egress) {
await routeScenarioExplicitly(scenario, machine, name);
log(` routed ${machine} inside the scenario, its default out through the uplink`);
continue;
}
if (!behind) {
await routeViaTransit(scenario, spec, name);
continue;
@@ -219,6 +240,104 @@ export async function applyDefaultRoutes(
}
}
/** One route a machine with egress needs, so the scenario stays reachable and stays inside. */
export interface ScenarioRoute {
/** The range this route is for. */
cidr: string;
/** The next hop inside the scenario, or null when there is none and the range is unreachable. */
via: string | null;
}
/**
* The routes a machine with egress needs into the rest of the scenario.
*
* Pure, and exported, because this is the decision that keeps the uplink and the declared gateway
* from fighting — and a decision only a full raise could check is one nobody checks.
*
* One route per segment the machine is not already on, through whatever it would have defaulted to:
* its gateway if it sits behind one, transit if it sits on a public segment and transit exists.
* What has no such path is `unreachable` — the faithful translation of the state it was in before,
* where its default route pointed into scenery that dropped it, and safer, because an unreachable
* route cannot be answered by whatever network the workstation happens to sit on.
*/
export function scenarioRoutesFor(scenario: Scenario, machine: string): ScenarioRoute[] {
const spec = scenario.machines[machine];
if (!spec || spec.at === "detached") return [];
const at = spec.at;
const onSegments = new Set(at.map((a) => a.segment));
const behindGateway = at.some((a) => scenario.segments[a.segment]?.gateway);
const routes: ScenarioRoute[] = [];
for (const [segment, segmentSpec] of Object.entries(scenario.segments)) {
if (onSegments.has(segment)) continue;
for (const cidr of segmentSpec.cidr) {
const slash = cidr.lastIndexOf("/");
if (slash === -1) continue;
const v6 = cidr.slice(0, slash).includes(":");
routes.push({
cidr,
via: behindGateway ? gatewayInside(scenario, at, v6) : transitOn(scenario, at, v6),
});
}
}
return routes;
}
/**
* Apply those routes, and leave the default to the uplink.
*
* No `dev`: every next hop here is on-link, so the kernel picks the interface, and asking `awk` to
* count links is one more thing that can pick `docker0`.
*/
async function routeScenarioExplicitly(
scenario: Scenario,
machine: string,
name: string,
): Promise<void> {
for (const { cidr, via } of scenarioRoutesFor(scenario, machine)) {
const family = cidr.slice(0, cidr.lastIndexOf("/")).includes(":") ? "-6" : "-4";
const route = via ? `${cidr} via ${via}` : `unreachable ${cidr}`;
await incus(
["exec", name, "--", "sh", "-c", `ip ${family} route replace ${route} 2>/dev/null || true`],
30_000,
);
}
}
/** The inside address of the gateway this machine sits behind: the first host address. */
function gatewayInside(scenario: Scenario, at: Attachment[], v6: boolean): string | null {
const behind = at.find((a) => scenario.segments[a.segment]?.gateway);
if (!behind) return null;
for (const range of scenario.segments[behind.segment]?.cidr ?? []) {
const slash = range.lastIndexOf("/");
if (slash === -1) continue;
const base = range.slice(0, slash);
if (base.includes(":") !== v6) continue;
if (v6) return `${base.replace(/::$/, "")}::1`;
const octets = base.split(".");
octets[3] = "1";
return octets.join(".");
}
return null;
}
/** The transit router's address on the public segment this machine sits on. */
function transitOn(scenario: Scenario, at: Attachment[], v6: boolean): string | null {
// One public segment means everything public is adjacent and no transit router is raised, so
// there is nothing to point at — see raiseTransit.
const publicSegments = Object.values(scenario.segments).filter((s) => s.kind === "public");
if (publicSegments.length < 2) return null;
const onPublic = at.find((a) => scenario.segments[a.segment]?.kind === "public");
if (!onPublic) return null;
for (const cidr of scenario.segments[onPublic.segment]?.cidr ?? []) {
if (cidr.slice(0, cidr.lastIndexOf("/")).includes(":") !== v6) continue;
const via = transitAddress(cidr);
if (via) return via.slice(0, via.lastIndexOf("/"));
}
return null;
}
/** A machine on a public segment reaches the other public networks through transit. */
async function routeViaTransit(
scenario: Scenario,
src/lifecycle/base.ts
+14 -7
View File
@@ -87,10 +87,16 @@ export async function buildBaseImage(
// Trust the documentation ranges as plain-HTTP registries.
//
// A scenario's registry is scenery inside the scenario, serving over HTTP, and a runtime
// will not pull from one without being told. Scoped to RFC 5737 and RFC 3849 ranges rather
// than a specific address, because those never route on the real internet — so this cannot
// make a real machine trust a real registry, whatever it is copied onto.
// **Kept after the lab's own registry was deleted, because it was never only for that.** The
// mesh HAS a registry — the `registry` module, `mesh-registry`, serving artifacts to the whole
// mesh on port 5000 over plain HTTP from whatever node runs it. In a scenario that node's
// address is a documentation-range address, and a runtime will not pull from a plain-HTTP
// registry without being told to. Take this away and the artifact store is unusable from every
// machine but the one hosting it.
//
// Scoped to RFC 5737 and RFC 3849 ranges rather than a specific address, because those never
// route on the real internet — so this cannot make a real machine trust a real registry,
// whatever it is copied onto.
await incus([
"exec", BUILDER, "--", "sh", "-c",
`mkdir -p /etc/docker && printf '%s' '${JSON.stringify({
@@ -162,8 +168,8 @@ export async function buildBaseImage(
// Read back that the runtime will actually pull over plain HTTP from a documentation
// range. Writing the file is not the same as the daemon honouring it, and a base image
// that looks right here fails much later — in a sealed scenario, as a container that
// cannot fetch its image, which is a long way from the cause.
// that looks right here fails much later — as a container that cannot fetch its image
// from the mesh's own artifact store, which is a long way from the cause.
const trusted = await incusOk(
["exec", BUILDER, "--", "docker", "info", "--format", "{{.RegistryConfig.InsecureRegistryCIDRs}}"],
60_000,
@@ -172,7 +178,8 @@ export async function buildBaseImage(
throw new BaseImageError(
`the runtime in ${BUILDER} does not trust the documentation ranges as plain-HTTP ` +
`registries. It reported: ${trusted?.trim() || "nothing"}\n` +
` Every scenario raised from this image would fail to pull from its own registry.`,
` Every scenario raised from this image would fail to pull from the mesh's own ` +
`artifact store, which serves plain HTTP inside the scenario.`,
);
}
log(" trusts the documentation ranges as registries");
src/lifecycle/egress.ts
+126
View File
@@ -0,0 +1,126 @@
/**
* Confirming a machine that says it can reach the outside actually can.
*
* **This is what the registry-reachability check became.** The old one proved that every machine
* could fetch a manifest from the registry the lab raised inside the scenario — a real check of a
* fake path, since no production mesh has such a registry. What a machine actually does is pull
* from the internet, and that is now the thing worth proving before a raise says it is finished.
*
* The failure it exists to stop is the same one, in the same shape: `raise` returns, the caller
* applies a substrate, the first pull fails, no node enrols, and the instance is left a bare
* shell — with the cause several steps back and looking like a mesh fault rather than a lab one.
*
* Two things are checked, in this order, because they fail differently and the difference is the
* whole diagnosis:
*
* - **A name resolves.** Without this the machine has a route and no way to use it, and every
* pull dies inside the runtime saying it cannot look up a host.
* - **The path carries.** A request to the registry every image ultimately comes from, over the
* uplink, through whatever gateway sits in front of this machine. Any HTTP answer counts: what
* is in question is the path, not whether Docker Hub likes us.
*/
import type { Scenario } from "../declaration/types.ts";
import { incus, incusOk } from "../incus/client.ts";
export class EgressError extends Error {
constructor(message: string) {
super(message);
this.name = "EgressError";
}
}
/** The host every image is fetched through, in the end. Asked for, never pulled from, here. */
const UPSTREAM = "registry-1.docker.io";
/**
* Confirm every machine declaring `egress` can resolve and reach the outside.
*
* Run after the routes and the firewalls, because that is the path a pull will take: a home node's
* default is the uplink, its route to the rest of the scenario is through its gateway, and its own
* filtering is in place. Checking earlier would prove something no pull relies on.
*/
export async function confirmEgress(
scenario: Scenario,
machineNames: Map<string, string>,
log: (message: string) => void = () => {},
waitSeconds = 120,
): Promise<void> {
for (const [machine, spec] of Object.entries(scenario.machines)) {
if (!spec.egress || spec.at === "detached") continue;
const name = machineNames.get(machine);
if (!name) continue;
if (!(await resolves(name, waitSeconds))) {
// One repair, then a verdict. The uplink is the lab's own network and its DHCP server is
// also its resolver, so the machine has been told the answer and may simply have nowhere
// to write it — an image without systemd-resolved leaves `UseDNS=yes` inert.
await pointResolverAtTheUplink(name);
if (!(await resolves(name, 30))) {
throw new EgressError(
`${machine} declares egress and cannot resolve ${UPSTREAM}.\n` +
` It has a route out and no way to use it, so every image pulled from the internet ` +
`would fail inside the runtime as a lookup error.\n` +
` The uplink's DHCP server is also its resolver; this machine has not taken it.`,
);
}
}
const code = await reaches(name, waitSeconds);
if (!code) {
throw new EgressError(
`${machine} declares egress, resolves names, and cannot reach ${UPSTREAM}.\n` +
` This is the PATH: its default route, the uplink, or the host's own forwarding. ` +
`Every third-party image this machine needs is pulled from the internet, so anything ` +
`applied to it would stop at the first container.`,
);
}
log(` ${machine} reaches the internet over its uplink (${UPSTREAM} answered ${code})`);
}
}
async function resolves(name: string, waitSeconds: number): Promise<boolean> {
const deadline = Date.now() + waitSeconds * 1_000;
while (Date.now() < deadline) {
const said = await incusOk(
["exec", name, "--", "sh", "-c", `getent hosts ${UPSTREAM} >/dev/null && echo yes`], 30_000,
);
if (said?.trim() === "yes") return true;
await new Promise((r) => setTimeout(r, 3_000));
}
return false;
}
/**
* Any HTTP status at all, which is what "the path carries" means.
*
* Not 200: an unauthenticated `/v2/` is answered 401 by design, and a check demanding 200 would
* fail on a machine whose network is perfect.
*/
async function reaches(name: string, waitSeconds: number): Promise<string | null> {
const deadline = Date.now() + waitSeconds * 1_000;
while (Date.now() < deadline) {
const said = (await incusOk(
["exec", name, "--", "sh", "-c",
`curl -s -o /dev/null -w '%{http_code}' --max-time 15 https://${UPSTREAM}/v2/`], 40_000,
))?.trim();
if (said && /^[1-5][0-9]{2}$/.test(said)) return said;
await new Promise((r) => setTimeout(r, 5_000));
}
return null;
}
/**
* Write a resolver of last resort: the uplink's own gateway, which serves DHCP and DNS both.
*
* Deliberately the machine's default next hop rather than a name looked up somewhere — for a
* machine with egress that is the uplink by construction, since every scenario range is routed
* explicitly and nothing else defaults.
*/
async function pointResolverAtTheUplink(name: string): Promise<void> {
await incus([
"exec", name, "--", "sh", "-c",
`via=$(ip -4 route show default | awk '{print $3}' | head -n1); ` +
`[ -n "$via" ] && printf 'nameserver %s\\n' "$via" > /etc/resolv.conf; true`,
], 30_000);
}
src/lifecycle/place.ts
+129 -5
View File
@@ -19,6 +19,7 @@ import { join } from "node:path";
import { incus, incusOk, succeeds } from "../incus/client.ts";
import { around, log, shorten } from "../log.ts";
import { isMeshBuilt, repositoryOf, type HeldImage } from "../pinning.ts";
/**
* What this stage can put inside a machine.
@@ -263,17 +264,17 @@ export async function placeImage(
// sealed machine cannot reach. Measured, not assumed: the load says `Loaded image ID:`
// instead of `Loaded image:`, and `docker images` then lists nothing.
//
// This collides with novox/hq ADR 0006, which pins bundle images BY DIGEST and has the host
// refuse anything else. Reconciling the two needs a registry inside the scenario, which is
// real design work — see 04-ISSUES/009.
// What an archive DOES keep is the image's own ID — the digest of its configuration — and that
// is how the mesh's own images are named once loaded. See {@link loadHeldImages}.
if (reference.includes("@sha256:")) {
throw new PlacementError(
machine,
`${reference} is pinned by digest, and an image placed from an archive cannot keep its ` +
`digest — a repo digest only exists for an image a registry served.\n` +
` Placing it would load an image with no name, and a container declaring that digest ` +
`would try to reach a registry the machine cannot see.\n` +
` Place it by tag, or give the scenario a registry (novox/hq 04-ISSUES/009).`,
`would try to reach a registry.\n` +
` Place it by tag. What survives being loaded is the image's own ID, which is what a ` +
`declaration names it by (mesh-host: an image the machine already holds).`,
);
}
@@ -397,3 +398,126 @@ async function waitForRuntime(instanceName: string, machine: string): Promise<vo
` systemd is waiting on:\n${jobs.trim() || " (it said nothing)"}`,
);
}
// --- the mesh's own images ----------------------------------------------------------------------
/**
* Which machine is handed which of the mesh's own images.
*
* A machine that says nothing gets the lot, which is right for a one-machine bed. A machine that
* lists some gets those. This is where a workstation running one small module stops paying for
* forty runtimes it will never start.
*/
export function planHeldImages(scenario: Scenario): { machine: string; images: string[] }[] {
const all = scenario.images ?? [];
if (all.length === 0) return [];
return Object.entries(scenario.machines)
.map(([machine, spec]) => ({ machine, images: spec.images ?? all }))
.filter((plan) => plan.images.length > 0);
}
/**
* Put the mesh's own images onto the machines that need them, and say what they are now called.
*
* **This is what replaced the lab's registry**, and the difference is the whole point. A registry
* inside the scenario served every image — third-party ones included — from an address that exists
* in no production mesh, so a bootstrap that could only work against it went green here and would
* have failed anywhere else. There is no such registry now: third-party images are pulled from the
* internet over each machine's `egress` uplink, and the mesh's own arrive the way they arrive on an
* operator's machine — somebody built them and put them there.
*
* The reference a declaration then uses is the image's **own ID**, the digest of its configuration.
* `docker load` preserves it, so the name is identical on the workstation that built the image and
* on every machine handed a copy — immutable, unforgeable, and requiring nothing to have served it
* (mesh-host, *an image may be named by the digest of its own configuration*).
*
* Read back on both sides. The ID is taken from the workstation and then CONFIRMED on the machine,
* because a load that lands a different image than the one exported is exactly the silent fault
* this lab exists to catch — and the reference is what every manifest will be rewritten to.
*/
export async function loadHeldImages(
scenario: Scenario,
machineNames: Map<string, string>,
log: (message: string) => void = () => {},
): Promise<HeldImage[]> {
const plans = planHeldImages(scenario);
if (plans.length === 0) return [];
const held: HeldImage[] = [];
for (const requested of scenario.images ?? []) {
// Refused by the validator, so reaching here would be a validator bug — but the consequence
// is a third-party image quietly loaded from the workstation instead of pulled, which is the
// fiction all of this exists to remove. Cheap to check, expensive to miss.
if (!isMeshBuilt(requested)) {
throw new Error(
`images: '${requested}' is not one of the mesh's own images. It is pulled from the ` +
`internet by the machine that needs it, not loaded from this workstation.`,
);
}
const wanted = plans.filter((plan) => plan.images.includes(requested)).map((p) => p.machine);
if (wanted.length === 0) continue;
const id = (await local(
"docker", ["image", "inspect", "--format", "{{.Id}}", requested], 60_000,
)).stdout.trim();
if (!/^sha256:[0-9a-f]{64}$/.test(id)) {
throw new Error(
`${requested} is not on this workstation, so there is nothing to hand the machines.\n` +
` It is one of the mesh's own images and exists in no registry — nothing can pull it.\n` +
` Build it first (mesh-control's \`make image …\`, or scripts/build-module-runtime.sh).`,
);
}
const image: HeldImage = { requested, repository: repositoryOf(requested), reference: id };
// Exported once, handed to each machine that asked for it. The archive is the expensive part
// and it does not depend on the destination.
const tar = join(tmpdir(), `mesh-lab-held-${process.pid}-${Date.now()}.tar`);
const saved = await local("docker", ["save", requested, "-o", tar], 900_000);
if (!saved.ok) {
await unlink(tar).catch(() => {});
throw new Error(`cannot export ${requested} from this workstation: ${saved.stderr.trim()}`);
}
try {
for (const machine of wanted) {
const name = machineNames.get(machine);
if (!name) continue;
await waitForRuntime(name, machine);
await incus(["file", "push", tar, `${name}/tmp/held.tar`], 900_000);
const loaded = await incusOk(
["exec", name, "--", "docker", "load", "-i", "/tmp/held.tar"], 900_000,
);
if (!loaded?.includes("Loaded image")) {
throw new PlacementError(
machine,
`${requested} was pushed to ${machine} and did not load.\n` +
` The runtime said: ${loaded?.trim() || "nothing"}`,
);
}
// The reference every manifest is about to be rewritten to, confirmed present under
// exactly that name. Asking for the tag would prove the load happened; asking for the ID
// proves the thing a declaration will name is the thing that is there.
const there = (await incusOk(
["exec", name, "--", "docker", "image", "inspect", "--format", "{{.Id}}", id], 120_000,
))?.trim();
if (there !== id) {
throw new PlacementError(
machine,
`${requested} loaded onto ${machine} and is not there as ${id}.\n` +
` The runtime answered '${there || "nothing"}'.\n` +
` Every manifest naming this image would be rewritten to a reference the machine ` +
`does not hold, and nothing serves it — so the apply would stop at the container.`,
);
}
await succeeds(["exec", name, "--", "rm", "-f", "/tmp/held.tar"], 60_000);
}
} finally {
await unlink(tar).catch(() => {});
}
held.push(image);
log(` ${requested} → ${id.slice(0, 19)}… on ${wanted.join(", ")}`);
}
return held;
}
src/lifecycle/raise.ts
+24 -40
View File
@@ -22,9 +22,10 @@ import { applyAddresses, applyDefaultRoutes } from "./address.ts";
import { assertSupported } from "./supported.ts";
import { planRouters, raiseRouters, raiseTransit } from "./router.ts";
import { applyHostFirewalls } from "./firewall.ts";
import { IMAGE_PREFIX, BASE_IMAGE_ALIAS, BASE_IMAGE_HOWTO, planPlacements, applyPlacements } from "./place.ts";
import { IMAGE_PREFIX, BASE_IMAGE_ALIAS, BASE_IMAGE_HOWTO, planPlacements, applyPlacements, loadHeldImages } from "./place.ts";
import { baseImageExists, UPSTREAM_IMAGE } from "./base.ts";
import { confirmRegistryServes, discardStock, raiseRegistry, stockRegistry } from "./registry.ts";
import { confirmEgress } from "./egress.ts";
import type { HeldImage } from "../pinning.ts";
import { log as record } from "../log.ts";
/** Drivers whose snapshots are copy-on-write. On `dir` a snapshot is a full copy. */
@@ -47,12 +48,15 @@ export interface RaisedScenario {
networks: string[];
pool: string;
/**
* Images the scenario's registry serves, as references a declaration can pin.
* The mesh's own images, as loaded onto the machines, and what a declaration should call them.
*
* Reported rather than declared, because the digest is the one this registry assigned and
* is not knowable before it was raised.
* Reported rather than declared: an image built from source has no digest until it has been
* built, and what names it here is the digest of its own configuration.
*
* **Only ours.** Everything third-party is pulled from the internet by the machine that needs
* it, so it is not in this list and nothing rewrites it.
*/
images: string[];
images: HeldImage[];
}
export class RaiseError extends Error {
@@ -314,24 +318,6 @@ export async function raise(
const routers = await raiseRouters(scenario, instanceId, planRouters(scenario, instanceId), log);
if (transit) routers.push(transit);
// Stocked on this workstation, where there is a network, and served from inside the
// scenario, where there is not (novox/hq 04-ISSUES/009).
enter("stocking the registry");
const stock = await stockRegistry(scenario.images ?? [], log);
let registry: Awaited<ReturnType<typeof raiseRegistry>> = null;
try {
enter("raising the registry");
registry = await raiseRegistry(scenario, instanceId, stock, log, pool);
} finally {
// Cleaning up scratch must not fail a raise that succeeded. The scenario is standing
// and usable; a directory left behind is untidy, and saying so is the honest report.
try {
await discardStock(stock);
} catch (err) {
log(` (could not remove the registry's scratch directory: ${(err as Error).message})`);
}
}
enter("routing machines through their gateways");
await applyDefaultRoutes(scenario, byMachine, log);
@@ -340,31 +326,29 @@ export async function raise(
enter("applying host firewalls");
await applyHostFirewalls(scenario, byMachine, log);
// **Only now can "the registry is serving" be said truthfully.** Raising it proved the
// registry answers on its own machine; a machine pulls across a segment, and the home nodes
// pull through a gateway whose route and firewall were applied in the two steps above. So the
// path is checked here, where it is finally the one a pull will take — and before `placing`,
// which is minutes of work that a machine unable to fetch an image cannot use.
//
// The alternative is what happened: `raise` returned, the caller applied a substrate whose
// every image is pinned to this registry, the first pull failed, no node enrolled, and the
// instance was left a bare shell. A raise that reports success owes the next step the fact it
// depends on.
if (registry) {
enter("confirming the registry serves the machines");
await confirmRegistryServes(registry, [...byMachine.values()], stock, log);
}
// **Only now is "this machine can reach the outside" a true statement.** The route, the
// gateway and the machine's own filtering are all in place, so this is the path a pull takes.
// A raise that returned without checking would hand the next step a fact it depends on and
// has no way to test — which is how a substrate apply used to die on its first pull.
enter("confirming egress reaches the internet");
await confirmEgress(scenario, byMachine, log);
// Last, and only once the underlay is real. Placing before the machines can reach each
// other would test the host against a network the scenario does not describe.
enter("placing");
await applyPlacements(scenario, byMachine, log);
// After `placing`, because loading an image needs the container runtime that `placing`
// confirmed. The mesh's own images only — everything third-party is pulled by the machine
// itself, over its uplink, exactly as it is on a real one.
enter("loading the mesh's own images onto the machines");
const images = await loadHeldImages(scenario, byMachine, log);
return {
instanceId,
scenario: scenario.scenario,
images: registry?.pinned ?? [],
machines: [...created, ...routers, ...(registry ? [registry.machine] : [])],
images,
machines: [...created, ...routers],
networks,
pool,
};
src/lifecycle/registry.ts
-487
View File
@@ -1,487 +0,0 @@
/**
* A registry inside the scenario.
*
* A sealed machine cannot reach a registry, and an image placed from an archive cannot keep its
* digest — `docker save` of a digest reference produces an archive with no repo tag, because a
* repo digest only exists for an image a registry served (novox/hq 04-ISSUES/009). So an image
* pinned by digest, which is the only kind the host accepts
* ([ADR 0006](../../02-DECISIONS/0046-the-installer-fetches-what-it-pins.md)), could not be
* placed at all.
*
* The answer is a registry, and it is not a workaround for the lab: ADR 0006 names an OCI
* registry as substrate, and ADR 0006 says a first node fetches "upstream, wherever the image
* ordinarily lives". **This is that upstream** — scenery, like the transit router is the
* internet ([ADR 0016](../../02-DECISIONS/0033-a-router-is-scenery-not-a-node.md)).
*
* The digests it serves are its own, not Docker Hub's, and that is correct rather than a
* compromise. What ADR 0006 requires is a reference that is exact and cannot move. A digest
* assigned by this registry is both.
*/
import { spawn } from "node:child_process";
import { incus, incusOk, succeeds } from "../incus/client.ts";
import { macFor, networkName } from "./names.ts";
import { addressLink } from "./address.ts";
import { around, log, shorten } from "../log.ts";
import { BASE_IMAGE_ALIAS, placeImage } from "./place.ts";
import { mkdtemp, rm } from "node:fs/promises";
import { tmpdir } from "node:os";
import { join } from "node:path";
/** The image the registry itself runs from. Placed by tag, which archives keep. */
export const REGISTRY_IMAGE = "registry:2";
/** Where the registry serves, inside its machine. */
export const REGISTRY_PORT = 5000;
export class RegistryError extends Error {
constructor(message: string) {
super(message);
this.name = "RegistryError";
}
}
export interface StockedImage {
/** What the scenario asked for, as written. */
requested: string;
/** The repository path the registry serves it under. */
repository: string;
/** The digest THIS registry assigned. What a declaration pins. */
digest: string;
}
export interface Stock {
/** A directory holding the registry's data, ready to be placed in a machine. */
dataDir: string;
images: StockedImage[];
}
/**
* Build a registry's data directory on this workstation, with the given images in it.
*
* Runs a throwaway registry here — where there IS a network — pushes into it, and keeps what
* it wrote. Research 012's reframing again: fetch at build time on a machine that has a
* network, apply on a target that needs nothing.
*
* The caller owns the returned directory and must remove it.
*/
export async function stockRegistry(
references: string[],
log: (message: string) => void = () => {},
): Promise<Stock> {
if (references.length === 0) return { dataDir: "", images: [] };
const dataDir = await mkdtemp(join(tmpdir(), "mesh-lab-registry-"));
const container = `mesh-lab-stock-${process.pid}`;
const port = 5000 + (process.pid % 1000);
await docker(["rm", "-f", container], 60_000);
const started = await docker(
["run", "-d", "--name", container, "-p", `${port}:5000`, "-v", `${dataDir}:/var/lib/registry`,
REGISTRY_IMAGE],
300_000,
);
if (!started.ok) {
await rm(dataDir, { recursive: true, force: true });
throw new RegistryError(
`cannot run ${REGISTRY_IMAGE} on this workstation to stock a registry: ${started.stderr.trim()}`,
);
}
try {
await waitForRegistry(port);
const images: StockedImage[] = [];
for (const reference of references) {
// The repository path a machine will pull from. A tag is dropped: what a declaration
// pins is the digest, and carrying the tag as well would invite pinning the wrong one.
const repository = repositoryFor(reference);
const target = `localhost:${port}/${repository}`;
const tagged = await docker(["tag", reference, target], 60_000);
if (!tagged.ok) {
throw new RegistryError(
`${reference} is not on this workstation, and the lab does not fetch on a scenario's ` +
`behalf. Pull it here first.\n ${tagged.stderr.trim()}`,
);
}
const pushed = await docker(["push", target], 900_000);
if (!pushed.ok) throw new RegistryError(`cannot push ${reference}: ${pushed.stderr.trim()}`);
const digest = digestFrom(pushed.stdout + pushed.stderr);
if (!digest) {
throw new RegistryError(
`${reference} was pushed and the registry did not report a digest. Without one there ` +
`is nothing for a declaration to pin.`,
);
}
images.push({ requested: reference, repository, digest });
log(` stocked ${repository}@${digest}`);
}
return { dataDir, images };
} catch (err) {
await discardStock({ dataDir, images: [] });
throw err;
} finally {
await docker(["rm", "-f", container], 60_000);
}
}
/**
* Remove a stocked registry's data.
*
* Through a container, because a container wrote it. The registry runs as root inside, so the
* blobs it writes into a bind mount are owned by root and an ordinary process cannot remove
* them — `rmdir` fails with EACCES on a directory that looks like ours.
*
* Whoever made the files removes them.
*/
export async function discardStock(stock: Stock): Promise<void> {
if (!stock.dataDir) return;
await docker(["run", "--rm", "-v", `${stock.dataDir}:/stock`, REGISTRY_IMAGE,
"sh", "-c", "rm -rf /stock/* /stock/.[!.]* 2>/dev/null || true"], 120_000);
await rm(stock.dataDir, { recursive: true, force: true }).catch(() => {});
}
/** `alpine:3.20` and `alpine` both serve from `alpine`; `foo/bar:1` from `foo/bar`. */
export function repositoryFor(reference: string): string {
const withoutDigest = reference.split("@")[0] ?? reference;
const lastColon = withoutDigest.lastIndexOf(":");
const lastSlash = withoutDigest.lastIndexOf("/");
return lastColon > lastSlash ? withoutDigest.slice(0, lastColon) : withoutDigest;
}
/** `docker push` prints `<tag>: digest: sha256:… size: …` on its last useful line. */
export function digestFrom(output: string): string | null {
const match = output.match(/digest:\s*(sha256:[a-f0-9]{64})/);
return match?.[1] ?? null;
}
async function waitForRegistry(port: number): Promise<void> {
for (let i = 0; i < 30; i++) {
const probe = await docker(["run", "--rm", "--network", "host", REGISTRY_IMAGE,
"sh", "-c", `wget -q -O- http://localhost:${port}/v2/ >/dev/null 2>&1`], 30_000);
if (probe.ok) return;
await new Promise((r) => setTimeout(r, 1_000));
}
throw new RegistryError("a registry was started on this workstation and never answered");
}
function docker(
args: string[],
timeoutMs: number,
): Promise<{ ok: boolean; stdout: string; stderr: string }> {
// The second of the three places the lab runs an external program (novox/hq 04-ISSUES/024).
// `docker push` of a large image is minutes of legitimate silence, which is exactly when a
// heartbeat earns its keep.
return around(`docker ${shorten(args)}`, () => runDocker(args, timeoutMs), { heartbeatMs: 15_000 });
}
function runDocker(
args: string[],
timeoutMs: number,
): Promise<{ ok: boolean; stdout: string; stderr: string }> {
return new Promise((resolve) => {
const child = spawn("docker", args, { stdio: ["ignore", "pipe", "pipe"] });
let stdout = "";
let stderr = "";
const timer = setTimeout(() => child.kill("SIGKILL"), timeoutMs);
child.stdout.on("data", (d) => (stdout += d));
child.stderr.on("data", (d) => (stderr += d));
child.on("error", (err) => {
clearTimeout(timer);
resolve({ ok: false, stdout, stderr: err.message });
});
child.on("close", (code) => {
clearTimeout(timer);
// A docker failure is an answer here rather than an exception, so it would otherwise pass
// through the log looking exactly like a success.
if (code !== 0) {
log.debug(` exit ${code}: ${shorten([stderr.trim() || "(nothing on stderr)"], 400)}`);
}
resolve({ ok: code === 0, stdout, stderr });
});
});
}
// --- the registry inside a scenario ------------------------------------------------------------
/**
* Where the registry sits on its segment.
*
* A convention rather than a declaration, like the router's. `.250` is chosen to sit well away
* from the low addresses scenarios give their machines, so a scenario can be written without
* thinking about it and a collision is obvious when it happens.
*/
export const REGISTRY_HOST_OCTET = 250;
/** The address the registry answers on, given the segment it is attached to. */
export function registryAddress(cidr: string): string {
const [network] = cidr.split("/");
const parts = (network ?? "").split(".");
if (parts.length !== 4) {
throw new RegistryError(
`cannot place a registry on '${cidr}': it is not an IPv4 network, and the registry needs ` +
`an address a machine can be pointed at.`,
);
}
return `${parts[0]}.${parts[1]}.${parts[2]}.${REGISTRY_HOST_OCTET}`;
}
/** What a declaration should pin, once a scenario is raised. */
export function pinnedReference(address: string, image: StockedImage): string {
return `${address}:${REGISTRY_PORT}/${image.repository}@${image.digest}`;
}
// --- raising it inside a scenario ---------------------------------------------------------------
/** What a raised registry is, and what a declaration needs from it. */
export interface RaisedRegistry {
machine: string;
segment: string;
address: string;
/** Each image, as a reference a declaration can pin. */
pinned: string[];
}
/**
* Pick the segment the registry sits on.
*
* A public segment, because that is what stands in for the outside world — a first node fetches
* from upstream, and this is upstream. An IPv4 range, because a machine has to be pointed at it
* by address.
*/
export function registrySegment(
segments: Record<string, { kind: string; cidr: string[] }>,
): { name: string; cidr: string } | null {
for (const [name, segment] of Object.entries(segments)) {
if (segment.kind !== "public") continue;
const v4 = segment.cidr.find((c) => !c.includes(":"));
if (v4) return { name, cidr: v4 };
}
return null;
}
/**
* Raise a registry inside the scenario and load the stocked images into it.
*
* Scenery, in the same sense the transit router is: nothing under test runs on it, it holds no
* identity, and no assertion is made about its internals. It exists so that a machine can fetch
* an image the way a real one does — over the network, from a registry, by digest.
*/
export async function raiseRegistry(
scenario: { segments: Record<string, { kind: string; cidr: string[] }> },
instanceId: string,
stock: Stock,
log: (message: string) => void = () => {},
/**
* The copy-on-write pool the scenario's machines were placed on. The registry goes on it too —
* NOT the profile's default `dir` pool — so a sized root disk is thin (paid for as it fills)
* rather than a full allocation on the host's own filesystem. Absent, it falls back to the
* profile default, which is the historical behaviour for a small registry.
*/
pool?: string,
): Promise<RaisedRegistry | null> {
if (stock.images.length === 0) return null;
const segment = registrySegment(scenario.segments);
if (!segment) {
throw new RegistryError(
`this scenario declares images and has no public IPv4 segment to serve them from.\n` +
` The registry stands in for the outside world, so it sits on a public segment.`,
);
}
const address = registryAddress(segment.cidr);
const name = `mlab-${instanceId}-registry`;
const prefix = segment.cidr.slice(segment.cidr.lastIndexOf("/"));
if (!(await succeeds(["config", "show", name], 15_000))) {
// The registry holds the WHOLE `images:` union on its own root disk, and the base image's
// default is only ~10GiB. A single-node bed stocks a handful of images and fits; a broad bed
// — and especially the full segmented mesh, whose union is both server sets at once (~28GiB)
// — overflows it, and the raise dies "no space left on device" while pushing blobs into the
// registry. So the registry gets a sized root disk. Thin on a copy-on-write pool, so a small
// bed pays only for what it actually stocks; MESH_LAB_REGISTRY_DISK overrides for a giant one.
const registryDisk = process.env["MESH_LAB_REGISTRY_DISK"] ?? "80GiB";
await incus([
"init", BASE_IMAGE_ALIAS, name, "--vm",
"-c", "security.secureboot=false",
"-c", "limits.memory=1GiB",
...(pool ? ["-s", pool] : []),
"-d", `root,size=${registryDisk}`,
"-c", `user.mesh-lab.instance=${instanceId}`,
// Tagged as a machine as well, so `destroy` finds it with one query — a router that
// carried only its own tag was left behind and held its networks open.
"-c", "user.mesh-lab.machine=registry",
"-c", "user.mesh-lab.registry=true",
], 300_000);
await succeeds(["config", "device", "remove", name, "eth0"], 15_000);
await incus([
"config", "device", "add", name, "eth0", "nic",
"nictype=bridged",
`parent=${networkName(instanceId, segment.name)}`,
`hwaddr=${macFor(instanceId, "registry", 0)}`,
]);
}
await succeeds(["start", name], 60_000);
await waitForAgent(name);
// Addressed the way every other machine is: a systemd-networkd unit matching the MAC.
//
// **This used to be `ip addr add`, and it stalled the lab.** An address set by hand leaves
// networkd waiting to configure a link it was never told about, so the link sits at
// `configuring`, `systemd-networkd-wait-online` never returns — its timeout is `infinity` —
// and `network-online.target` is never reached. Docker is ordered after that target, so
// `docker load` two lines below blocked on a socket whose daemon was queued behind a target
// that would never come.
//
// Matching on MAC and not on interface name is still the rule: a machine with a container
// runtime has a `docker0` that sorts before `enp5s0`, and naive selection configures that.
await addressLink(name, {
device: "eth0",
mac: macFor(instanceId, "registry", 0),
addresses: [`${address}${prefix}`],
// The registry takes the segment's default. It carried no MTU before this and still does
// not: what a scenario sets an MTU for is the path under test, and this is scenery.
mtu: undefined,
});
log(` registry on ${segment.name} at ${address}`);
// The registry's own image, placed by tag — an archive keeps a tag and cannot keep a digest,
// which is the whole reason this machine exists.
// Logged, not silenced. This is the step a stall sat in for thirty-five minutes while the
// caller had passed it a callback that threw everything away (novox/hq 04-ISSUES/024).
await placeImage(name, "registry", REGISTRY_IMAGE, log);
// The destination must EXIST before a recursive push, or incus copies the source's contents
// rather than the source — the data lands one directory too shallow, the registry finds
// nothing where it looks, and every pull fails with `not found`.
await incus(["exec", name, "--", "mkdir", "-p", "/srv/registry"], 60_000);
await incus(["file", "push", "-r", `${stock.dataDir}/docker`, `${name}/srv/registry/`], 900_000);
await incus(["exec", name, "--", "docker", "run", "-d",
"--name", "registry", "--restart", "unless-stopped",
"-p", `${REGISTRY_PORT}:5000`,
"-v", "/srv/registry:/var/lib/registry",
REGISTRY_IMAGE], 300_000);
// Read back that each image is SERVED, by asking for its manifest by digest — which is
// exactly what a machine will do.
//
// Not that the catalog endpoint answers: `{"repositories":[]}` contains the word
// `repositories`, so checking for that passed on a registry holding nothing at all, and the
// failure surfaced much later as a container that could not be pulled.
let answered = false;
for (let i = 0; i < 20 && !answered; i++) {
const ping = await incusOk(["exec", name, "--", "curl", "-s", "-o", "/dev/null",
"-w", "%{http_code}", "--max-time", "3",
`http://localhost:${REGISTRY_PORT}/v2/`], 30_000);
answered = ping?.trim() === "200";
if (!answered) await new Promise((r) => setTimeout(r, 2_000));
}
if (!answered) {
throw new RegistryError(
`the registry on ${name} started and never answered. Machines in this scenario cannot ` +
`fetch an image, so nothing that declares a container will work.`,
);
}
const pinned: string[] = [];
for (const image of stock.images) {
const code = await incusOk(["exec", name, "--", "curl", "-s", "-o", "/dev/null",
"-w", "%{http_code}", "--max-time", "5",
"-H", "Accept: application/vnd.docker.distribution.manifest.v2+json",
`http://localhost:${REGISTRY_PORT}/v2/${image.repository}/manifests/${image.digest}`,
], 60_000);
if (code?.trim() !== "200") {
throw new RegistryError(
`the registry on ${name} is running and does not serve ${image.repository}@${image.digest} ` +
`(it answered ${code?.trim() || "nothing"}).\n` +
` The images were stocked on this workstation and did not arrive intact, so a ` +
`machine declaring that image would fail to pull it.`,
);
}
const reference = pinnedReference(address, image);
pinned.push(reference);
log(` serving ${reference}`);
}
return { machine: name, segment: segment.name, address, pinned };
}
/**
* Confirm the registry serves the MACHINES, not just itself.
*
* **`raiseRegistry` proves the wrong thing, and the difference cost a whole raise.** It curls
* `localhost:5000` from inside the registry's own machine — which says the registry process is up
* and holds the blobs, and says nothing at all about the path every other machine actually uses:
* across a segment, and for the home nodes through a NAT gateway whose route is applied two steps
* LATER. So `raise` could return "serving" with the anchor unable to reach the registry at all, the
* substrate apply's first pull would fail, no node could enrol, and the instance was left a bare
* shell — VMs and a registry and nothing else.
*
* A raise is not finished while that is still possible. This is the check that makes "the registry
* is serving" mean what the next step needs it to mean: from each machine, on the address it will
* pin, over the network it will use, once its route and its firewall are in place.
*
* **What it proves and what it does not.** It asks for `/v2/` and then for one stocked manifest BY
* DIGEST — the same two requests a pull begins with, from the same place. It does not fetch layers:
* every digest was already read back inside the registry machine, so what is in question here is
* the path, not the content, and a probe per machine keeps the check to seconds rather than the
* many minutes a full pull of a hundred images would take.
*/
export async function confirmRegistryServes(
registry: RaisedRegistry,
machines: string[],
stock: Stock,
log: (message: string) => void = () => {},
waitSeconds = 180,
): Promise<void> {
const probe = stock.images[0];
if (!probe) return;
const base = `http://${registry.address}:${REGISTRY_PORT}`;
const manifest =
`-H "Accept: application/vnd.docker.distribution.manifest.v2+json" ` +
`${base}/v2/${probe.repository}/manifests/${probe.digest}`;
for (const machine of machines) {
const deadline = Date.now() + waitSeconds * 1_000;
let last = "";
let served = false;
while (!served && Date.now() < deadline) {
// One shell, two requests: the machine can reach the registry AND the registry answers for
// an image by the digest a declaration pins. Either alone passes on a registry serving
// nothing, which is the failure this whole function exists to stop reporting as success.
const said = await incusOk(["exec", machine, "--", "sh", "-c",
`printf '%s %s' ` +
`"$(curl -s -o /dev/null -w '%{http_code}' --max-time 5 ${base}/v2/)" ` +
`"$(curl -s -o /dev/null -w '%{http_code}' --max-time 10 ${manifest})"`,
], 40_000);
last = said?.trim() ?? "";
served = last === "200 200";
if (!served) await new Promise((r) => setTimeout(r, 3_000));
}
if (!served) {
throw new RegistryError(
`${machine} cannot pull from the registry at ${registry.address}:${REGISTRY_PORT} after ` +
`${waitSeconds}s (it got "${last || "nothing"}" for /v2/ and for ` +
`${probe.repository}@${probe.digest}).\n` +
` The registry answers on its own machine, so this is the PATH: this machine's route, ` +
`its gateway, or its firewall. Every image this scenario declares is unreachable from ` +
`here, so anything applied to it would fail on its first pull.`,
);
}
log(` ${machine} can pull from ${registry.address}:${REGISTRY_PORT}`);
}
}
async function waitForAgent(name: string): Promise<void> {
for (let i = 0; i < 90; i++) {
if (await succeeds(["exec", name, "--", "true"], 10_000)) return;
await new Promise((r) => setTimeout(r, 2_000));
}
throw new RegistryError(`${name} started and its agent never answered.`);
}
src/pinning.ts
+74 -28
View File
@@ -1,55 +1,101 @@
/**
* Rewriting an image reference to the one a scenario's own registry serves.
* Naming the mesh's own images by what they are.
*
* **A digest is not knowable until something is built** (novox/hq 04-ISSUES/025). A manifest in a
* repository can pin a third-party image, because somebody can ask a registry what a tag points
* at. It cannot pin an image the mesh builds itself: that image does not exist yet, and when it
* does its digest belongs to whichever registry served it.
* at. It cannot pin an image the mesh builds itself: mesh-control, mesh-builder, mesh-route-proxy,
* the per-module runtimes and the provisioners exist in no registry, so there is no manifest
* digest to write down. The catalogue ships sixty-four zeros for them, which parses, resolves,
* composes — and stops on the machine.
*
* The bundle has always had this problem and solves it by rewriting references once the scenario's
* registry is up and its digests are known. Modules have exactly the same problem and were solving
* it by shipping sixty-four zeros, which parses, resolves, composes — and stops on the machine.
* The lab used to answer that with a registry of its own: raise one inside the scenario, push
* everything into it, and rewrite every reference — third-party ones included — to the digest it
* assigned. **That registry does not exist in production, so the lab was testing a fiction**, and
* the fiction hid the bootstrap problems it was supposed to find.
*
* So the rewriting is shared rather than copied, and matches on the **repository**, because that
* is the part a person writes and the only part that survives being served somewhere else.
* What is true instead is two things:
*
* - **Third-party images are pulled from the internet.** They are left exactly as written, and
* the machine fetches them over its `egress` uplink the way any machine does.
* - **The mesh's own images are built and handed over.** They are loaded onto the machine from
* the workstation that built them, and named by the digest of their own image configuration —
* a bare `sha256:…`, which mesh-host accepts as "an image this machine already holds"
* (mesh-host, *an image may be named by the digest of its own configuration*).
*
* So the rewriting that remains is only the second kind, and it matches on the **repository**,
* because that is the part a person writes and the only part a placeholder digest does not say.
*/
/** `192.0.2.250:5000/ghcr.io/mailu/admin@sha256:…` → `ghcr.io/mailu/admin` */
export function repositoryOf(pinned: string): string {
const at = pinned.indexOf("@");
const body = at === -1 ? pinned : pinned.slice(0, at);
const slash = body.indexOf("/");
// Everything after the registry. A reference with no slash at all is its own repository.
return slash === -1 ? body : body.slice(slash + 1);
/** What the lab loaded onto a machine, and what a declaration should call it. */
export interface HeldImage {
/** As the scenario asked for it, a tag this workstation holds: `mesh-runtime-postgres:development`. */
requested: string;
/** What a manifest names it by, with no tag and no digest: `mesh-runtime-postgres`. */
repository: string;
/**
* The reference a declaration uses: a bare `sha256:<64 hex>`.
*
* The image's own ID — the digest of its configuration — which `docker load` preserves, so the
* name is the same on the workstation that built it and on every machine it was handed to.
*/
reference: string;
}
/** `alpine:3.20` and `alpine` both mean `alpine`; `foo/bar:1` means `foo/bar`. */
export function repositoryOf(reference: string): string {
const withoutDigest = reference.split("@")[0] ?? reference;
const lastColon = withoutDigest.lastIndexOf(":");
const lastSlash = withoutDigest.lastIndexOf("/");
return lastColon > lastSlash ? withoutDigest.slice(0, lastColon) : withoutDigest;
}
/**
* Replace every reference to a stocked repository with the reference this scenario serves.
* Whether a reference names an image the mesh builds for itself.
*
* Matching is on the repository and ignores whatever registry and digest were written down —
* a file may name `postgres@sha256:7456…` or `mesh-provision-postgres@sha256:0000…` and both mean
* *the postgres this scenario has*. That is the whole point: the text says which image, the
* scenario says which copy.
* **Derived from the shape the build produces, not from a list of names.** `make image
* builder-image provisioner-image objectstore-image redis-provisioner-image proxy-image` in
* mesh-control and `scripts/build-module-runtime.sh` here both tag their output `mesh-<something>`
* with no registry host and no upstream organisation — that is what "built here, published
* nowhere" looks like, and a hardcoded list would go stale the first time a module is added.
*
* A repository the scenario did not stock is left alone rather than blanked. It may be reachable
* some other way, and silently emptying a reference would produce the exact failure this exists to
* prevent.
* The absence of a slash carries the weight: `ghcr.io/mailu/admin` and
* `registry.example/novox/www` both say where they are fetched from, and a bare
* `mesh-runtime-plex` says there is nowhere.
*/
export function pinnedInto(text: string, served: string[]): string {
export function isMeshBuilt(reference: string): boolean {
const repository = repositoryOf(reference);
return !repository.includes("/") && repository.startsWith("mesh-");
}
/**
* Replace every reference to one of the mesh's own images with the image the machine holds.
*
* Matching is on the repository and ignores whatever digest was written down — a manifest says
* `mesh-runtime-postgres@sha256:0000…` and means *the runtime this machine was given*.
*
* **Everything else is left exactly as it is.** `postgres@sha256:7456…`, `gitea/gitea@sha256:…`
* and `ghcr.io/mailu/admin@sha256:…` are pulled from the internet over the machine's uplink, which
* is what a real machine does and the reason the lab's own registry is gone.
*/
export function pinnedInto(text: string, held: HeldImage[]): string {
let out = text;
for (const pinned of served) {
const repository = repositoryOf(pinned);
const escaped = repository.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
for (const image of held) {
const escaped = image.repository.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
// Optionally a registry, then the repository, then any digest. Anchored on a quote or
// whitespace so a longer repository ending in a shorter one is not half-replaced.
out = out.replaceAll(
new RegExp(`(?<=^|["\\s])(?:[A-Za-z0-9_.:-]+\\/)*${escaped}@sha256:[0-9a-f]{64}`, "g"),
pinned,
image.reference,
);
}
return out;
}
/** The reference for one repository, or undefined if this scenario loaded no such image. */
export function referenceFor(held: HeldImage[], repository: string): string | undefined {
return held.find((image) => image.repository === repository)?.reference;
}
/** Whether anything is still pinned to a placeholder, which would fail on the machine. */
export function stillUnpinned(text: string): string[] {
return [...text.matchAll(/([A-Za-z0-9_.:/-]+)@sha256:0{64}/g)].map((m) => m[1]!);
src/suite.ts
+3 -3
View File
@@ -53,9 +53,9 @@ export async function runSuite(args: string[]): Promise<number> {
// the long run reaches the end of that path in about 160 seconds.
//
// So it cost a whole scenario, every passing run, to save about 45 seconds on a failing one.
// A scenario is three machines including a registry that boots a kernel to serve files, which
// is where the two minutes went. `test/integration/canary.test.ts` is still there and still
// runs when it is named; it is no longer raised on the way to everything else.
// A scenario is machines that each boot a kernel, which is where the two minutes went.
// `test/integration/canary.test.ts` is still there and still runs when it is named; it is no
// longer raised on the way to everything else.
const { code, seen } = await runFiles(files);
console.log("\n" + reportOn(counted(seen), (p, f) => record(p, f, files, process.env, against)));
src/warm.ts
+11 -10
View File
@@ -23,6 +23,7 @@ import { homedir } from "node:os";
import { list, restore, snapshot, snapshots, destroy } from "./lifecycle/operate.ts";
import type { Against } from "./lastrun.ts";
import { whatWasTested } from "./lastrun.ts";
import type { HeldImage } from "./pinning.ts";
/** The state a warm instance is kept at. One label, because a second is a state nobody named. */
export const label = "warm";
@@ -31,13 +32,13 @@ export interface Warm {
scenario: string;
instanceId: string;
/**
* The image references the scenario's registry serves, pinned by digest.
* The mesh's own images, as loaded onto this instance's machines, by the ID each is held under.
*
* Kept because they are worked out while raising and a restored instance never raises. Without
* them a warm run knows nothing about what it can pull, and every test naming an image fails
* for a reason that has nothing to do with what it was testing.
* them a warm run knows nothing about what its machines hold, and every test naming one of our
* images fails for a reason that has nothing to do with what it was testing.
*/
images: string[];
images: HeldImage[];
/** The commit each repository was at when this was brought to its state. */
against: Against;
at: string;
@@ -187,23 +188,23 @@ export async function cool(): Promise<string | null> {
}
/**
* What a raised scenario stocked, held until it is kept.
* What a raised scenario loaded onto its machines, held until it is kept.
*
* Raising works the images out and snapshotting happens later, so this carries them between the
* two without the caller having to hold them.
*/
const stock = new Map<string, string[]>();
const stock = new Map<string, HeldImage[]>();
export function rememberStock(instanceId: string, images: string[]): void {
export function rememberStock(instanceId: string, images: HeldImage[]): void {
stock.set(instanceId, images);
}
function stockOf(instanceId: string): string[] {
function stockOf(instanceId: string): HeldImage[] {
return stock.get(instanceId) ?? [];
}
/** What a restored instance's registry serves, from when it was warmed. */
export function warmStock(instanceId: string): { images: string[] } {
/** What a restored instance's machines hold, from when it was warmed. */
export function warmStock(instanceId: string): { images: HeldImage[] } {
const warm = remembered();
if (!warm || warm.instanceId !== instanceId) return { images: [] };
return { images: warm.images };