Author SHA1 Message Date
jschoubben 81972a4995 A module answers the word the mesh asks: prepare
The runtime gains `prepare`, which brings this module's state to the shape this version needs and
exits (novox/hq ADR 0135). The entrypoints come from MESH_PREPARE, which a module's own image names
beside the entrypoints it already lists there — the module knows which of its files prepares its
state and nothing else could. No broker is connected: preparation runs before the version that would
use it. An empty list fails rather than passing quietly, because the mesh asks this only of a module
whose manifest says it prepares something, and exiting 0 would let that version serve against a
state nobody shaped.
2026-09-28 12:45:00 +02:00
mesh-admin 10e8191717 Merge pull request 'The runtime hears on its own inbox' (#17) from fix/the-runtime-hears-on-its-own-inbox into main 2026-09-28 02:30:24 +00:00
jschoubben d703cebff4 The runtime hears on its own inbox
Every user's inbox is private to it and the grant names it; a reply space the client invented was
refused, and with it every pull for the next message and every answer to a tool call.
2026-09-28 04:30:22 +02:00
mesh-admin 46b56d53a6 Merge pull request 'The pin is the only check: the runtime stops verifying the bus's name' (#16) from fix/the-pin-is-the-only-check into main 2026-09-28 02:03:46 +00:00
jschoubben d4a2802342 The pin is the only check: the runtime stops verifying the bus's name
Every module on the new runtime reached the handshake and failed on 'does not match
certificate's altnames': the bus's certificate names the seat, not the address a machine dials
it by, and pinning the exact certificate already decides everything a name check could. The
client's transport spreads the TLS options into Node's tls.connect, so the hostname check is
replaced with one that passes and the pinned certificate is the one authority accepted.
2026-09-28 04:03:43 +02:00
mesh-admin 8acfa7a07d Merge pull request 'One bus: the runtime pins the certificate after the server speaks, and the old transport goes' (#15) from feat/one-bus into main 2026-09-28 01:09:01 +00:00
3 changed files with 77 additions and 9 deletions
+16 -9
View File
@@ -19,7 +19,7 @@
import { createHash } from "node:crypto";
import net from "node:net";
import tls from "node:tls";
import { connect as natsConnect, headers as natsHeaders, StringCodec, type JsMsg, type Subscription } from "nats";
import { connect as natsConnect, headers as natsHeaders, StringCodec, type JsMsg, type Subscription, type TlsOptions } from "nats";
import type { Broker, Envelope, EventHeaders } from "@novox/mesh-sdk/messaging";
const sc = StringCodec();
@@ -85,6 +85,10 @@ export async function connectNats(
pass: cred.password,
name: `${cred.node ?? "?"}.${self}`,
tls: cred.fingerprint ? await pinnedTls(cred.url, cred.fingerprint) : undefined,
// Its own inbox, not a random one: every user's inbox is private to it (design 25 §4), and the
// grant names `_INBOX.<user>.>` — a reply space the client invented would be refused, and with
// it every pull for the next message and every answer to a tool call.
inboxPrefix: cred.user ? `_INBOX.${cred.user}` : undefined,
// Reconnect forever: the bus being restarted is an upgrade, not a reason for every module on
// the mesh to exit. `close()` stays the only thing that ends the connection.
maxReconnectAttempts: -1,
@@ -298,14 +302,15 @@ function normalizeFingerprint(fingerprint: string): string {
* A certificate authority is not consulted: the mesh issued this and knows its fingerprint,
* which is stronger than trusting whoever a machine's trust store happens to contain.
*
* **A constraint on the mesh, not a detail of this file.** Pinning the exact certificate makes
* hostname verification redundant in principle, but the NATS client exposes no hook to replace
* it — its TLS options are file paths and PEM strings, with no verify callback. So the
* certificate the mesh issues the bus **must carry a subject-alternative name matching the
* address nodes dial it by**. The fingerprint check below still happens and is still the real
* guarantee; what cannot be switched off is the check *beside* it.
* **The pin is the only check.** What comes back is handed to the client as its TLS options, and
* the client's transport spreads them into Node's own `tls.connect` — so the pinned certificate
* is the one authority the handshake accepts, and the hostname check beside it is replaced with
* one that always passes. Pinning the exact certificate makes verifying its name redundant, and
* the bus's certificate names the seat (`mesh-broker`), not the address a machine happens to
* dial it by: every module on the mesh met "does not match certificate's altnames" the first time
* it reached the handshake (2026-09-28).
*/
async function pinnedTls(rawUrl: string, fingerprint: string): Promise<{ ca: string }> {
async function pinnedTls(rawUrl: string, fingerprint: string): Promise<TlsOptions> {
const url = new URL(rawUrl.includes("://") ? rawUrl : `nats://${rawUrl}`);
const port = url.port ? Number(url.port) : 4222;
// **The bus speaks first, in the clear.** A NATS server sends its INFO line before TLS begins,
@@ -342,7 +347,9 @@ async function pinnedTls(rawUrl: string, fingerprint: string): Promise<{ ca: str
);
}
const pem = `-----BEGIN CERTIFICATE-----\n${certificate.raw.toString("base64").replace(/(.{64})/g, "$1\n")}\n-----END CERTIFICATE-----\n`;
return { ca: pem };
// Node's option, not the client's: the transport passes the whole object on. `undefined` from
// checkServerIdentity is "the name is fine"; the pin above already decided the rest.
return { ca: pem, checkServerIdentity: () => undefined } as TlsOptions;
}
/** The mesh's topic matching: `*` is one token, `#` the rest. This is the module's vocabulary —
+42
View File
@@ -5,6 +5,10 @@
// starts consuming as it is imported, so this also runs consumers.
// mesh-tools emit TYPE [JSON] emit one event onto the mesh and exit — an operable primitive,
// and what an events test uses to put a message on the wire.
// mesh-tools prepare bring this module's state to the shape this version needs and exit
// — the runtime's answer to the word the mesh asks every module
// (novox/hq ADR 0135). The entrypoints come from MESH_PREPARE, which
// the module's own image names beside MESH_TOOL_MODULES.
// mesh-tools run ENTRYPOINT run one compiled module entrypoint to completion and exit — the
// runtime side of a run-once step (novox/hq ADR 0052). It imports
// the given entrypoint, whose top-level code does its work — seed a
@@ -18,6 +22,7 @@
// amqps account scoped to this module. Preferred: a module holds its own.
// MESH_BROKER_URL a plain URL, for the bootstrap/admin case before a module has an account.
// MESH_TOOL_MODULES /path/a,/path/b,… compiled module entrypoints (serve mode)
// MESH_PREPARE /path/a,/path/b,… compiled entrypoints that prepare this module's state
// MESH_MODULE / MESH_NODE the identity stamped onto emitted events (ADR 0042)
import { readFileSync } from "node:fs";
@@ -177,12 +182,49 @@ async function runEntry(entrypoint: string): Promise<void> {
await import(pathToFileURL(entrypoint).href);
}
/**
* Bring this module's state to the shape this version needs, and exit — the runtime's answer to the
* one word the mesh asks every module (novox/hq ADR 0135).
*
* The entrypoints come from `MESH_PREPARE`, which a module's own image sets beside the entrypoints it
* already lists there: the module knows which of its files prepares its state, and nothing else could.
* Each is imported in the order given, to completion, with no broker — preparation runs before the
* version that would use it, so there is nothing yet to talk to.
*
* **An empty list is a failure, not a no-op.** The mesh only asks this of a module whose manifest says
* it prepares something; a module that says so and names nothing has been built wrong, and exiting 0
* would let that version serve against a state nobody shaped.
*/
async function prepareState(): Promise<void> {
const named = (process.env.MESH_PREPARE ?? "")
.split(",")
.map((entry) => entry.trim())
.filter((entry) => entry !== "");
if (named.length === 0) {
console.error(
"mesh-tools prepare: this module was asked to prepare its state and its image names nothing " +
"to do it with — set MESH_PREPARE to the compiled entrypoint(s) that prepare it, the way " +
"MESH_TOOL_MODULES names the ones it serves",
);
process.exit(1);
}
for (const entrypoint of named) {
console.log(`[mesh-tools] preparing with ${entrypoint}`);
await runEntry(entrypoint);
}
console.log(`[mesh-tools] prepared: ${named.length} entrypoint(s) ran to completion`);
}
async function main(): Promise<void> {
const [command, ...rest] = process.argv.slice(2);
if (command === "run") {
await runEntry(rest[0] ?? "");
return;
}
if (command === "prepare") {
await prepareState();
return;
}
if (command === "invoke") {
const [module, tool] = rest;
if (!module || !tool) {
+19
View File
@@ -1,3 +1,4 @@
import { spawn } from "node:child_process";
import { test } from "node:test";
import assert from "node:assert/strict";
import { fatalBrokerReason, PinMismatchError } from "../src/broker-nats.ts";
@@ -41,3 +42,21 @@ test("a non-Error value does not crash the classifier", () => {
assert.equal(fatalBrokerReason("just a string"), null);
assert.equal(fatalBrokerReason(undefined), null);
});
// **A module asked to prepare its state and naming nothing is a failure, not a no-op** (novox/hq
// ADR 0135). The mesh asks this only of a module whose manifest says it prepares something, so an
// image that names nothing was built wrong, and exiting 0 would let that version serve against a
// state nobody shaped.
test("preparing with nothing named fails rather than passing quietly", async () => {
const runtime = new URL("../dist/main.js", import.meta.url).pathname;
const ran = await new Promise<{ code: number | null; said: string }>((resolve) => {
const child = spawn(process.execPath, [runtime, "prepare"], {
env: { ...process.env, MESH_PREPARE: "" },
});
let said = "";
child.stderr.on("data", (chunk) => (said += String(chunk)));
child.on("close", (code) => resolve({ code, said }));
});
assert.notEqual(ran.code, 0, "a module that prepares nothing exited 0, so its version would serve");
assert.match(ran.said, /MESH_PREPARE/);
});