jschoubben
c98ee7a82d
Broken with a package that does not exist, so the failure is real and fixable. The mesh reports it failed; the resources that could be applied were, because one broken thing no longer blocks the rest; `push --behind` names that machine and not the one that is fine; the module is corrected; and the machine recovers with nobody naming it. And with nothing behind, it says so rather than doing nothing quietly.
387 lines
20 KiB
TypeScript
387 lines
20 KiB
TypeScript
/**
|
|
* A mesh, raised from nothing, joined by two machines, delivering a credential neither the mesh
|
|
* nor the broker can read.
|
|
*
|
|
* Everything before this proves a part. This proves the parts meet — which is the thing the
|
|
* project keeps saying cannot be checked any other way (novox/hq ADR 0001: every fault of
|
|
* 2026-08-22 was found in production because nothing could be stood up locally).
|
|
*
|
|
* It needs a host binary and the substrate bundle:
|
|
*
|
|
* MESH_LAB_HOST_BINARY=.../mesh-host
|
|
* MESH_LAB_BUNDLE=.../examples/substrate-first-node.lock
|
|
*
|
|
* The bundle's image references are rewritten to the ones this scenario's own registry serves.
|
|
* A digest belongs to whatever registry serves it, so a committed bundle names a registry that is
|
|
* not this one — rewriting is what makes it applicable rather than a placeholder to tidy away.
|
|
*/
|
|
|
|
import { test, before, after } from "node:test";
|
|
import assert from "node:assert/strict";
|
|
import { existsSync, readFileSync } from "node:fs";
|
|
import { loadScenario } from "../../src/declaration/parse.ts";
|
|
import { raise } from "../../src/lifecycle/raise.ts";
|
|
import { destroy, exec } from "../../src/lifecycle/operate.ts";
|
|
import { hostBinaryPath, HOST_PATH } from "../../src/lifecycle/place.ts";
|
|
import { labIsUsable, destroyAll } from "./harness.ts";
|
|
|
|
const capability = await labIsUsable();
|
|
const binary = hostBinaryPath();
|
|
const bundle = process.env["MESH_LAB_BUNDLE"] ?? "";
|
|
|
|
const skip = !capability.usable
|
|
? `lab not usable: ${capability.why}`
|
|
: !binary || !existsSync(binary)
|
|
? "MESH_LAB_HOST_BINARY is not set to a built mesh-host"
|
|
: !bundle || !existsSync(bundle)
|
|
? "MESH_LAB_BUNDLE is not set to a substrate bundle (mesh-host examples/)"
|
|
: false;
|
|
|
|
const SCENARIO = "two-nodes";
|
|
let instanceId = "";
|
|
|
|
function quote(s: string): string {
|
|
return `'${s.replaceAll("'", `'\\''`)}'`;
|
|
}
|
|
|
|
async function on(machine: string, command: string): Promise<{ out: string; ok: boolean }> {
|
|
const { stdout } = await exec(instanceId, machine, [
|
|
"sh", "-c", `${command} 2>&1; echo "__exit=$?"`,
|
|
]);
|
|
const marker = stdout.lastIndexOf("__exit=");
|
|
return { out: stdout.slice(0, marker), ok: Number(stdout.slice(marker + 7).trim()) === 0 };
|
|
}
|
|
|
|
async function must(machine: string, command: string): Promise<string> {
|
|
const { out, ok } = await on(machine, command);
|
|
if (!ok) throw new Error(`${machine}: ${command}\n${out}`);
|
|
return out;
|
|
}
|
|
|
|
/** The control plane, which runs in a container on the first node. */
|
|
async function mesh(command: string): Promise<string> {
|
|
return must("anchor", `docker exec mesh-control /mesh-control ${command}`);
|
|
}
|
|
|
|
/**
|
|
* The bundle, with every image reference pointed at this scenario's registry.
|
|
*
|
|
* Matched by repository rather than by the whole reference, because the address and the digest
|
|
* both differ from whatever the committed bundle names — and a bundle that names the wrong
|
|
* registry is not wrong, it is built for a different target.
|
|
*/
|
|
function bundleFor(images: string[]): string {
|
|
let text = readFileSync(bundle, "utf8");
|
|
for (const pinned of images) {
|
|
const repository = pinned.slice(pinned.indexOf("/") + 1, pinned.indexOf("@"));
|
|
const escaped = repository.replaceAll("/", "\\/").replaceAll(".", "\\.");
|
|
text = text.replaceAll(
|
|
new RegExp(`[A-Za-z0-9_.:-]+\\/${escaped}@sha256:[0-9a-f]+`, "g"),
|
|
pinned,
|
|
);
|
|
}
|
|
return text;
|
|
}
|
|
|
|
/** Take a token out of what `token issue` printed. It is the one base64url blob on its own line. */
|
|
function tokenFrom(said: string): string {
|
|
const found = said.split("\n").map((l) => l.trim()).find((l) => l.length > 100 && !l.includes(" "));
|
|
assert.ok(found, `no token in:\n${said}`);
|
|
return found;
|
|
}
|
|
|
|
before(async () => {
|
|
if (skip) return;
|
|
const raised = await raise(loadScenario(`scenarios/${SCENARIO}.yml`), {});
|
|
instanceId = raised.instanceId;
|
|
|
|
// The first node raises everything from a file rather than from a bundle built into the binary,
|
|
// because the digests are this registry's and are not known until it is up.
|
|
await must("anchor", `cat > /tmp/substrate.lock <<'MESHBUNDLE'\n${bundleFor(raised.images)}\nMESHBUNDLE`);
|
|
await must("anchor", `${HOST_PATH} apply /tmp/substrate.lock`);
|
|
}, { timeout: 1_800_000 });
|
|
|
|
after(async () => {
|
|
if (instanceId) await destroy(instanceId);
|
|
await destroyAll(`${SCENARIO}-`);
|
|
}, { timeout: 600_000 });
|
|
|
|
test("a bare machine becomes a mesh", { skip, timeout: 600_000 }, async () => {
|
|
const running = await must("anchor", `docker ps --format '{{.Names}}'`);
|
|
for (const container of ["mesh-store", "mesh-broker", "mesh-control"]) {
|
|
assert.match(running, new RegExp(container), `${container} is not running`);
|
|
}
|
|
// Answering, not merely up. A container that is running is not a control plane that replies —
|
|
// a distinction this project has already paid for once.
|
|
assert.ok((await mesh("status")).length > 0);
|
|
});
|
|
|
|
test("both machines join it, and the token is all they need", { skip, timeout: 900_000 }, async () => {
|
|
for (const [machine, node] of [["anchor", "anchor"], ["laptop", "laptop"]] as const) {
|
|
await mesh(`node add ${node}`);
|
|
const token = tokenFrom(await mesh(`token issue --node ${node}`));
|
|
// No --name. The token says what the mesh calls the machine, which is the fault this walk
|
|
// found the first time it was run.
|
|
const said = await must(machine, `${HOST_PATH} enrol --token ${quote(token)}`);
|
|
assert.match(said, new RegExp(`enrolled as ${node}`), said);
|
|
assert.match(said, /sealing key/, "no sealing key was generated");
|
|
}
|
|
|
|
const recorded = await must("anchor",
|
|
`docker exec mesh-store psql -U postgres -d inventory -qAt ` +
|
|
`-c "select name from node where sealing_key is not null order by name"`);
|
|
assert.equal(recorded.trim().split("\n").map((l) => l.trim()).sort().join(","), "anchor,laptop",
|
|
"the mesh did not record a sealing key for both machines");
|
|
});
|
|
|
|
test("a credential reaches both ends and the mesh holds neither", { skip, timeout: 900_000 }, async () => {
|
|
// The whole argument, on real machines: the two ends must hold the SAME password, and it must
|
|
// appear nowhere the mesh or the broker could read it.
|
|
await must("anchor", `printf %s '{"module":"postgres","version":"1",` +
|
|
`"provides":[{"name":"database","scope":"mesh"}],"serves":{"database":{"port":5432}},` +
|
|
`"grants":{"database":"/var/lib/mesh-host/grants"},` +
|
|
`"receives":{"database":"/var/lib/mesh-host/grants/mesh.json"},"resources":[]}' > /tmp/pg.json`);
|
|
await must("anchor", `printf %s '{"module":"meshboard","version":"1",` +
|
|
`"requires":["database"],"contributes":{"database":{"name":"meshboard"}},` +
|
|
`"binds":{"database":"/etc/meshboard/database.json"},` +
|
|
`"secrets":{"database":"/etc/meshboard/database.password"},"resources":[]}' > /tmp/app.json`);
|
|
await must("anchor", `docker cp /tmp/pg.json mesh-control:/pg.json`);
|
|
await must("anchor", `docker cp /tmp/app.json mesh-control:/app.json`);
|
|
await mesh("module add /pg.json");
|
|
await mesh("module add /app.json");
|
|
|
|
await mesh("overlay place anchor --hub --endpoint 192.0.2.10:51820 --site lab");
|
|
await mesh("overlay place laptop --site lab");
|
|
for (const node of ["anchor", "laptop"]) await mesh(`assign ${node} networking`);
|
|
await mesh("assign anchor postgres");
|
|
await mesh("assign laptop meshboard");
|
|
|
|
for (const machine of ["anchor", "laptop"]) {
|
|
await must(machine, `nohup ${HOST_PATH} run > /var/log/mesh-host.log 2>&1 & sleep 3`);
|
|
}
|
|
await mesh("push");
|
|
await new Promise((r) => setTimeout(r, 8000));
|
|
|
|
const onConsumer = (await must("laptop", `cat /etc/meshboard/database.password`)).trim();
|
|
const onProvider = (await must("anchor", `cat /var/lib/mesh-host/grants/laptop.secret`)).trim();
|
|
assert.ok(onConsumer.length >= 40, `the consumer's credential is ${onConsumer.length} characters`);
|
|
assert.equal(onConsumer, onProvider,
|
|
"the two ends hold different passwords, so nothing could ever authenticate");
|
|
|
|
// Only the machine it is for may read it.
|
|
assert.match(await must("laptop", `stat -c %a /etc/meshboard/database.password`), /^600/);
|
|
|
|
// And it is nowhere it could have been read on the way. The declaration crossed the broker; the
|
|
// database is the control plane's; the state is what the node reported back.
|
|
for (const [machine, where] of [
|
|
["laptop", "/var/lib/mesh-host/declared.json"],
|
|
["laptop", "/var/lib/mesh-host/state.json"],
|
|
["anchor", "/var/lib/mesh-host/declared.json"],
|
|
] as const) {
|
|
const { out } = await on(machine, `grep -c ${quote(onConsumer)} ${where}`);
|
|
assert.equal(out.trim(), "0", `the password is in ${where} on ${machine}`);
|
|
}
|
|
const inTheMesh = await must("anchor",
|
|
`docker exec mesh-store psql -U postgres -d inventory -qAt ` +
|
|
`-c "select count(*) from secret where for_consumer like '%${onConsumer}%' ` +
|
|
`or for_provider like '%${onConsumer}%'"`);
|
|
assert.equal(inTheMesh.trim(), "0", "the control plane's database holds the password in the clear");
|
|
});
|
|
|
|
test("the consumer is also told where its database is", { skip, timeout: 300_000 }, async () => {
|
|
// A password with no address is not a connection. This is the readable half, which stays
|
|
// readable on purpose — it is the secret half that could not be composed, not this one.
|
|
const told = JSON.parse(await must("laptop", `cat /etc/meshboard/database.json`));
|
|
assert.equal(told.from, "anchor");
|
|
assert.equal(told.at, "anchor.internal");
|
|
assert.equal(told.serves.port, 5432);
|
|
// And the name it resolves to was written by the mesh as well, on this machine.
|
|
assert.match(await must("laptop", `grep anchor.internal /etc/hosts`), /10\.42\.0\.\d+/);
|
|
});
|
|
|
|
test("when a machine cannot do what it was told, the mesh says which and why", { skip, timeout: 900_000 }, async () => {
|
|
// Demonstrated with inserted rows first, which proves the query and not the path. This sends a
|
|
// real machine something it will genuinely fail at, and asks the mesh afterwards.
|
|
//
|
|
// A package that does not exist, because that is a failure of the ordinary kind: the host tries,
|
|
// the package manager says no, and some of the declaration is applied and some is not — which
|
|
// is the situation `status` exists to distinguish from a machine that refused everything.
|
|
await must("anchor", `printf %s '{"module":"impossible","version":"1","resources":[` +
|
|
`{"id":"nothing","type":"package","package":"a-package-that-does-not-exist"}]}' > /tmp/imp.json`);
|
|
await must("anchor", `docker cp /tmp/imp.json mesh-control:/imp.json`);
|
|
await mesh("module add /imp.json");
|
|
await mesh("assign laptop impossible");
|
|
await mesh("push laptop");
|
|
await new Promise((r) => setTimeout(r, 10_000));
|
|
|
|
const said = await mesh("status");
|
|
assert.match(said, /not doing what they were told/, said);
|
|
assert.match(said, /laptop/, said);
|
|
// The machine's own words about the resource that failed, not a summary written at this end.
|
|
assert.match(said, /impossible\.nothing/, `the failing resource is not named:\n${said}`);
|
|
|
|
// And the distinction survives: this machine FAILED, it did not refuse. Refused means it is
|
|
// exactly as it was; failed means it is in a state nobody declared, and they are fixed in
|
|
// different places.
|
|
assert.match(said, /laptop\s+failed/, said);
|
|
|
|
// The other machine is not implicated.
|
|
assert.doesNotMatch(said.split("not heard from")[0] ?? said, /anchor\s+(failed|refused)/,
|
|
"a machine that did as it was told is listed as wrong");
|
|
});
|
|
|
|
test("a declaration waits for a machine that is switched off", { skip, timeout: 900_000 }, async () => {
|
|
// A machine is disconnected as an ordinary situation, not an exception (novox/hq ADR 0004), so
|
|
// a push to one that is not listening must wait rather than vanish. The queue is durable and the
|
|
// message persistent, which ought to be enough — but a lost declaration is silent, and "ought to
|
|
// be" is not a property.
|
|
//
|
|
// The machine is not merely idle here: its host is stopped, so nothing is consuming its queue.
|
|
|
|
// Nothing this test asserts should depend on what another left behind. The machine still has a
|
|
// deliberately-impossible module from the test above, and while that is assigned the mesh never
|
|
// updates its account of what the machine holds — a partial report is not an account, on
|
|
// purpose (novox/hq 04-ISSUES/010).
|
|
await mesh("unassign laptop impossible");
|
|
|
|
// Stop listening, and prove it stopped — a test that pushed to a machine that was still running
|
|
// would pass having checked nothing.
|
|
//
|
|
// By process name, never by matching the command line: `pkill -f` matches the shell running it
|
|
// too, which kills the connection carrying the command and hangs the caller waiting for a reply
|
|
// that will never come. Cost an hour once, in this file.
|
|
await must("laptop", `pkill -x mesh-host || true; sleep 1`);
|
|
const listening = await on("laptop", `pgrep -x mesh-host`);
|
|
assert.equal(listening.ok, false, "the host is still running, so this proves nothing");
|
|
|
|
await must("anchor", `printf %s '{"module":"while-away","version":"1","resources":[` +
|
|
`{"id":"note","type":"file","path":"/etc/mesh-while-away","content":"waited"}]}' > /tmp/away.json`);
|
|
await must("anchor", `docker cp /tmp/away.json mesh-control:/away.json`);
|
|
await mesh("module add /away.json");
|
|
await mesh("assign laptop while-away");
|
|
await mesh("push laptop");
|
|
|
|
// Nothing has happened on the machine, because nothing is there to do it.
|
|
const before = await on("laptop", `test -f /etc/mesh-while-away`);
|
|
assert.equal(before.ok, false, "a machine with no host applied a declaration");
|
|
|
|
// And now it listens again. No second push, and nobody says anything.
|
|
await must("laptop", `nohup ${HOST_PATH} run > /var/log/mesh-host.log 2>&1 & sleep 8`);
|
|
let arrived = false;
|
|
for (let i = 0; i < 20 && !arrived; i++) {
|
|
arrived = (await on("laptop", `test -f /etc/mesh-while-away`)).ok;
|
|
if (!arrived) await new Promise((r) => setTimeout(r, 2000));
|
|
}
|
|
if (!arrived) {
|
|
// Everything needed to tell "the message was never queued" from "the host never read it".
|
|
const log = await on("laptop", `tail -20 /var/log/mesh-host.log`);
|
|
const queues = await on("anchor",
|
|
`docker exec mesh-broker lavinmqctl list_queues name messages 2>&1 | head -10`);
|
|
const owned = await on("anchor",
|
|
`docker exec mesh-store psql -U postgres -d inventory -qAt -c "select name, outcome from node_report r join node n on n.id=r.node"`);
|
|
assert.fail(`a declaration sent to a switched-off machine was lost\n` +
|
|
`--- the host's log ---\n${log.out}\n--- the broker's queues ---\n${queues.out}\n` +
|
|
`--- what each machine last did ---\n${owned.out}`);
|
|
}
|
|
assert.equal((await must("laptop", `cat /etc/mesh-while-away`)).trim(), "waited");
|
|
|
|
// And the mesh's account of what that machine holds catches up too, or a later declaration
|
|
// would tell it to remove what it has just been given.
|
|
await new Promise((r) => setTimeout(r, 4000));
|
|
const owned = await must("anchor",
|
|
`docker exec mesh-store psql -U postgres -d inventory -qAt ` +
|
|
`-c "select owned from node where name = 'laptop'"`);
|
|
assert.match(owned, /while-away\.note/, `the mesh does not know the machine holds it: ${owned}`);
|
|
});
|
|
|
|
test("unassigning takes away exactly what it should", { skip, timeout: 900_000 }, async () => {
|
|
// Removal is the half nobody tests. The mesh takes away what IT declared and no longer declares,
|
|
// and never what the machine raised for itself from its bundle — which is the fault that
|
|
// destroyed a substrate once (novox/hq 04-ISSUES/010).
|
|
//
|
|
// Two modules, so the test can tell "removed the right one" from "removed everything".
|
|
for (const [name, path] of [["kept", "/etc/mesh-kept"], ["going", "/etc/mesh-going"]] as const) {
|
|
await must("anchor", `printf %s '{"module":"${name}","version":"1","resources":[` +
|
|
`{"id":"note","type":"file","path":"${path}","content":"${name}"}]}' > /tmp/${name}.json`);
|
|
await must("anchor", `docker cp /tmp/${name}.json mesh-control:/${name}.json`);
|
|
await mesh(`module add /${name}.json`);
|
|
await mesh(`assign anchor ${name}`);
|
|
}
|
|
await mesh("push anchor");
|
|
await new Promise((r) => setTimeout(r, 6000));
|
|
assert.ok((await on("anchor", `test -f /etc/mesh-kept`)).ok, "the first module did not arrive");
|
|
assert.ok((await on("anchor", `test -f /etc/mesh-going`)).ok, "the second module did not arrive");
|
|
|
|
await mesh("unassign anchor going");
|
|
await mesh("push anchor");
|
|
await new Promise((r) => setTimeout(r, 6000));
|
|
|
|
assert.equal((await on("anchor", `test -f /etc/mesh-going`)).ok, false,
|
|
"an unassigned module's file is still there");
|
|
assert.ok((await on("anchor", `test -f /etc/mesh-kept`)).ok,
|
|
"unassigning one module took another one's file with it");
|
|
|
|
// And the substrate this machine raised from its own bundle is untouched. It was not declared by
|
|
// the mesh, so the mesh must never remove it — the machine would take its own control plane
|
|
// away, which is exactly what happened before origins existed.
|
|
const running = await must("anchor", `docker ps --format '{{.Names}}'`);
|
|
for (const container of ["mesh-store", "mesh-broker", "mesh-control"]) {
|
|
assert.match(running, new RegExp(container),
|
|
`${container} was removed by a declaration that never declared it`);
|
|
}
|
|
});
|
|
|
|
test("a machine keeps what it was given when the mesh says nothing about it", { skip, timeout: 600_000 }, async () => {
|
|
// The other direction of the same rule. A node that is sent a declaration mentioning none of its
|
|
// private network must not lose it: the network came from a module that is still assigned, and
|
|
// "not in this message" is not "no longer wanted".
|
|
assert.ok((await on("laptop", `test -f /etc/wireguard/mesh0.conf`)).ok,
|
|
"the private network's configuration is gone");
|
|
assert.ok((await on("laptop", `grep -q anchor.internal /etc/hosts`)).ok,
|
|
"the mesh's names are gone");
|
|
});
|
|
|
|
test("a machine that fell behind catches up without being named", { skip, timeout: 900_000 }, async () => {
|
|
// `status` says which machines are not doing what they were told; something has to act on it.
|
|
// `push --behind` is that something, and it is a command rather than a timer to begin with —
|
|
// a scheduler is then a scheduler over this, rather than a second path to the same act.
|
|
|
|
// Break one machine, in a way that is fixable: a package that does not exist yet.
|
|
await must("anchor", `printf %s '{"module":"fixable","version":"1","resources":[` +
|
|
`{"id":"pkg","type":"package","package":"a-package-that-does-not-exist-yet"},` +
|
|
`{"id":"note","type":"file","path":"/etc/mesh-fixable","content":"here"}]}' > /tmp/fix.json`);
|
|
await must("anchor", `docker cp /tmp/fix.json mesh-control:/fix.json`);
|
|
await mesh("module add /fix.json");
|
|
await mesh("assign laptop fixable");
|
|
await mesh("push laptop");
|
|
await new Promise((r) => setTimeout(r, 8000));
|
|
|
|
assert.match(await mesh("status"), /laptop\s+failed/, "the machine did not report a failure");
|
|
// And the resources that COULD be applied were — one broken thing no longer blocks the rest
|
|
// (novox/hq 04-ISSUES/011).
|
|
assert.ok((await on("laptop", `test -f /etc/mesh-fixable`)).ok,
|
|
"a resource after the failing one was never attempted");
|
|
|
|
// Nothing is behind on the other machine, so nothing is pushed to it.
|
|
const named = await mesh("push --behind");
|
|
assert.match(named, /laptop/, named);
|
|
assert.doesNotMatch(named, /sent anchor/, `a machine that was fine was pushed to:\n${named}`);
|
|
|
|
// Fix the cause, the way somebody would: the module stops asking for the impossible thing.
|
|
await must("anchor", `printf %s '{"module":"fixable","version":"1","resources":[` +
|
|
`{"id":"note","type":"file","path":"/etc/mesh-fixable","content":"here"}]}' > /tmp/fix.json`);
|
|
await must("anchor", `docker cp /tmp/fix.json mesh-control:/fix.json`);
|
|
await mesh("module add /fix.json");
|
|
|
|
// And nobody names the machine.
|
|
await mesh("push --behind");
|
|
await new Promise((r) => setTimeout(r, 8000));
|
|
|
|
const after = await mesh("status");
|
|
assert.doesNotMatch(after, /laptop\s+(failed|refused)/,
|
|
`the machine did not recover:\n${after}`);
|
|
|
|
// With nothing behind, it says so rather than doing nothing quietly.
|
|
assert.match(await mesh("push --behind"), /every machine is doing what it was told/);
|
|
});
|