feat(mosaic): add authenticated external lease broker (#836)
This commit was merged in pull request #836.
This commit is contained in:
650
packages/mosaic/src/lease-broker/lease-broker.acceptance.spec.ts
Normal file
650
packages/mosaic/src/lease-broker/lease-broker.acceptance.spec.ts
Normal file
@@ -0,0 +1,650 @@
|
||||
import { randomUUID } from 'node:crypto';
|
||||
import { chmod, mkdtemp, readFile, stat, symlink, writeFile } from 'node:fs/promises';
|
||||
import { createConnection, type Socket } from 'node:net';
|
||||
import { tmpdir } from 'node:os';
|
||||
import { join } from 'node:path';
|
||||
import { spawn, spawnSync, type ChildProcess } from 'node:child_process';
|
||||
|
||||
import { afterEach, describe, expect, test, vi } from 'vitest';
|
||||
|
||||
interface BrokerReply {
|
||||
ok: boolean;
|
||||
code?: string;
|
||||
session_id?: string;
|
||||
peer?: { pid: number; uid: number; gid: number; starttime: string };
|
||||
token?: string;
|
||||
}
|
||||
|
||||
const daemonPath = new URL('../../framework/tools/lease-broker/daemon.py', import.meta.url)
|
||||
.pathname;
|
||||
const children: ChildProcess[] = [];
|
||||
|
||||
async function withTimeout<T>(
|
||||
promise: Promise<T>,
|
||||
label: string,
|
||||
milliseconds = 3_000,
|
||||
): Promise<T> {
|
||||
let timer: NodeJS.Timeout | undefined;
|
||||
try {
|
||||
return await Promise.race([
|
||||
promise,
|
||||
new Promise<never>((_, reject) => {
|
||||
timer = setTimeout(() => reject(new Error(`${label} timed out`)), milliseconds);
|
||||
}),
|
||||
]);
|
||||
} finally {
|
||||
if (timer !== undefined) clearTimeout(timer);
|
||||
}
|
||||
}
|
||||
|
||||
async function rawRequest(
|
||||
socketPath: string,
|
||||
write: (socket: Socket) => void,
|
||||
): Promise<BrokerReply> {
|
||||
return await withTimeout(
|
||||
new Promise<BrokerReply>((resolve, reject) => {
|
||||
const socket = createConnection(socketPath);
|
||||
let response = '';
|
||||
socket.setEncoding('utf8');
|
||||
socket.once('error', reject);
|
||||
socket.on('data', (chunk: string) => {
|
||||
response += chunk;
|
||||
});
|
||||
socket.once('end', () => resolve(JSON.parse(response) as BrokerReply));
|
||||
socket.once('connect', () => write(socket));
|
||||
}),
|
||||
'raw broker request',
|
||||
);
|
||||
}
|
||||
|
||||
async function request(socketPath: string, requestValue: object): Promise<BrokerReply> {
|
||||
return await new Promise<BrokerReply>((resolve, reject) => {
|
||||
const socket = createConnection(socketPath);
|
||||
let response = '';
|
||||
socket.setEncoding('utf8');
|
||||
socket.once('error', reject);
|
||||
socket.on('data', (chunk: string) => {
|
||||
response += chunk;
|
||||
});
|
||||
socket.once('end', () => resolve(JSON.parse(response) as BrokerReply));
|
||||
socket.once('connect', () => socket.end(`${JSON.stringify(requestValue)}\n`));
|
||||
});
|
||||
}
|
||||
|
||||
async function startBroker(
|
||||
parentMode = 0o700,
|
||||
): Promise<{ root: string; socket: string; state: string }> {
|
||||
const root = await mkdtemp(join(tmpdir(), 'mosaic-lease-broker-'));
|
||||
await chmod(root, parentMode);
|
||||
const socket = join(root, 'broker.sock');
|
||||
const state = join(root, 'state.json');
|
||||
const child = spawn('python3', [daemonPath, '--socket', socket, '--state', state], {
|
||||
stdio: ['ignore', 'pipe', 'pipe'],
|
||||
});
|
||||
children.push(child);
|
||||
await new Promise<void>((resolve, reject) => {
|
||||
let stderr = '';
|
||||
child.stderr?.setEncoding('utf8');
|
||||
child.stderr?.on('data', (chunk: string) => (stderr += chunk));
|
||||
child.once('error', reject);
|
||||
child.once('exit', (code: number | null) =>
|
||||
reject(new Error(`broker exited ${code}: ${stderr}`)),
|
||||
);
|
||||
child.stdout?.once('data', () => resolve());
|
||||
});
|
||||
return { root, socket, state };
|
||||
}
|
||||
|
||||
async function startBrokerWithState(stateValue: string): Promise<string> {
|
||||
const root = await mkdtemp(join(tmpdir(), 'mosaic-lease-broker-'));
|
||||
await chmod(root, 0o700);
|
||||
const state = join(root, 'state.json');
|
||||
await writeFile(state, stateValue, { mode: 0o600 });
|
||||
const child = spawn('python3', [
|
||||
daemonPath,
|
||||
'--socket',
|
||||
join(root, 'broker.sock'),
|
||||
'--state',
|
||||
state,
|
||||
]);
|
||||
children.push(child);
|
||||
return await new Promise<string>((resolve) => {
|
||||
let raw = '';
|
||||
child.stderr?.on('data', (chunk: Buffer) => (raw += chunk.toString()));
|
||||
child.once('exit', () => resolve(raw));
|
||||
});
|
||||
}
|
||||
|
||||
afterEach(() => {
|
||||
for (const child of children.splice(0)) child.kill('SIGTERM');
|
||||
vi.restoreAllMocks();
|
||||
});
|
||||
|
||||
describe('authenticated external lease broker', () => {
|
||||
test('peercred returns true kernel (pid,starttime)', async () => {
|
||||
const getuid = process.getuid;
|
||||
const getgid = process.getgid;
|
||||
if (getuid === undefined || getgid === undefined) {
|
||||
throw new Error('Linux peer credentials require process.getuid() and process.getgid()');
|
||||
}
|
||||
|
||||
const { socket } = await startBroker();
|
||||
const reply = await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
const statText = await readFile(`/proc/${process.pid}/stat`, 'utf8');
|
||||
const fields = statText.slice(statText.lastIndexOf(')') + 2).split(' ');
|
||||
expect(reply).toMatchObject({
|
||||
ok: true,
|
||||
peer: {
|
||||
pid: process.pid,
|
||||
uid: getuid(),
|
||||
gid: getgid(),
|
||||
starttime: fields[19],
|
||||
},
|
||||
});
|
||||
});
|
||||
|
||||
test.each([null, '', 'chosen'])('caller-asserted session_id refused (%j)', async (session_id) => {
|
||||
const { socket } = await startBroker();
|
||||
const reply = await request(socket, {
|
||||
action: 'register_anchor',
|
||||
runtime_generation: 1,
|
||||
session_id,
|
||||
});
|
||||
expect(reply).toMatchObject({ ok: false, code: 'CALLER_SESSION_ID_REFUSED' });
|
||||
});
|
||||
|
||||
test('sibling-substitution rejected', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const launcher = spawn(
|
||||
process.execPath,
|
||||
[
|
||||
'-e',
|
||||
`const n=require('net');const s=n.connect(${JSON.stringify(socket)},()=>s.end(JSON.stringify({action:'register_anchor',runtime_generation:1})+'\\n'));s.on('data',d=>{process.send(JSON.parse(d));setInterval(()=>{},1000)})`,
|
||||
],
|
||||
{ stdio: ['ignore', 'ignore', 'ignore', 'ipc'] },
|
||||
);
|
||||
children.push(launcher);
|
||||
const registration = await new Promise<BrokerReply>((resolve) =>
|
||||
launcher.once('message', (message) => resolve(message as BrokerReply)),
|
||||
);
|
||||
const attacker = spawn(
|
||||
process.execPath,
|
||||
[
|
||||
'-e',
|
||||
`const n=require('net');const s=n.connect(${JSON.stringify(socket)},()=>s.end(JSON.stringify({action:'authenticate',session_id:${JSON.stringify(registration.session_id)},runtime_generation:1})+'\\n'));s.pipe(process.stdout)`,
|
||||
],
|
||||
{ stdio: ['ignore', 'pipe', 'ignore'] },
|
||||
);
|
||||
children.push(attacker);
|
||||
let raw = '';
|
||||
attacker.stdout?.setEncoding('utf8');
|
||||
attacker.stdout?.on('data', (chunk: string) => (raw += chunk));
|
||||
await new Promise<void>((resolve) => attacker.once('exit', () => resolve()));
|
||||
expect(JSON.parse(raw)).toMatchObject({ ok: false, code: 'ANCESTRY_MISMATCH' });
|
||||
});
|
||||
|
||||
test('generation bump revokes prior incarnation', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const registered = await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
expect(
|
||||
await request(socket, {
|
||||
action: 'authenticate',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: 2,
|
||||
}),
|
||||
).toMatchObject({ ok: true });
|
||||
expect(
|
||||
await request(socket, {
|
||||
action: 'authenticate',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: 1,
|
||||
}),
|
||||
).toMatchObject({ ok: false, code: 'STALE_GENERATION' });
|
||||
});
|
||||
|
||||
test('same anchor re-registration reuses its session and revokes the prior incarnation', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const first = await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
const binding = {
|
||||
compaction_epoch: 2,
|
||||
request_epoch: 3,
|
||||
h_source: 'a'.repeat(64),
|
||||
h_payload: 'b'.repeat(64),
|
||||
schema_version: 1,
|
||||
};
|
||||
const minted = await request(socket, {
|
||||
action: 'mint_token',
|
||||
session_id: first.session_id,
|
||||
runtime_generation: 1,
|
||||
binding,
|
||||
});
|
||||
|
||||
const bumped = await request(socket, { action: 'register_anchor', runtime_generation: 2 });
|
||||
const repeated = await request(socket, { action: 'register_anchor', runtime_generation: 2 });
|
||||
const lower = await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
|
||||
expect(bumped).toMatchObject({ ok: true, session_id: first.session_id });
|
||||
expect(repeated).toMatchObject({ ok: true, session_id: first.session_id });
|
||||
expect(lower).toMatchObject({ ok: false, code: 'STALE_GENERATION' });
|
||||
expect(
|
||||
await request(socket, {
|
||||
action: 'authenticate',
|
||||
session_id: first.session_id,
|
||||
runtime_generation: 1,
|
||||
}),
|
||||
).toMatchObject({ ok: false, code: 'STALE_GENERATION' });
|
||||
expect(
|
||||
await request(socket, {
|
||||
action: 'consume_token',
|
||||
session_id: first.session_id,
|
||||
runtime_generation: 2,
|
||||
token: minted.token,
|
||||
}),
|
||||
).toMatchObject({ ok: false, code: 'TOKEN_REPLAY' });
|
||||
});
|
||||
|
||||
test('crypto token path works when Math.random is poisoned', async () => {
|
||||
const { socket } = await startBroker();
|
||||
vi.spyOn(Math, 'random').mockImplementation(() => {
|
||||
throw new Error('Math.random forbidden');
|
||||
});
|
||||
const registered = await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
const binding = {
|
||||
compaction_epoch: 2,
|
||||
request_epoch: 3,
|
||||
h_source: 'a'.repeat(64),
|
||||
h_payload: 'b'.repeat(64),
|
||||
schema_version: 1,
|
||||
};
|
||||
const first = await request(socket, {
|
||||
action: 'mint_token',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: 1,
|
||||
binding,
|
||||
});
|
||||
const second = await request(socket, {
|
||||
action: 'mint_token',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: 1,
|
||||
binding,
|
||||
});
|
||||
expect(first.token).toMatch(/^[a-f0-9]{64}$/);
|
||||
expect(second.token).not.toBe(first.token);
|
||||
expect(
|
||||
await request(socket, {
|
||||
action: 'consume_token',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: 1,
|
||||
token: first.token,
|
||||
}),
|
||||
).toMatchObject({ ok: true });
|
||||
expect(
|
||||
await request(socket, {
|
||||
action: 'consume_token',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: 1,
|
||||
token: first.token,
|
||||
}),
|
||||
).toMatchObject({ ok: false, code: 'TOKEN_REPLAY' });
|
||||
});
|
||||
|
||||
test('socket parent 0700 and socket 0600 enforced', async () => {
|
||||
const { root, socket, state } = await startBroker();
|
||||
await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
expect((await stat(root)).mode & 0o777).toBe(0o700);
|
||||
expect((await stat(socket)).mode & 0o777).toBe(0o600);
|
||||
expect((await stat(state)).mode & 0o777).toBe(0o600);
|
||||
});
|
||||
|
||||
test('insecure existing posture refused', async () => {
|
||||
await expect(startBroker(0o755)).rejects.toThrow(/INSECURE_PARENT_MODE/);
|
||||
});
|
||||
|
||||
test('malformed and oversized frames fail closed without killing broker', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const malformed = await new Promise<string>((resolve, reject) => {
|
||||
const connection = createConnection(socket, () => connection.end('{nope}\n'));
|
||||
let raw = '';
|
||||
connection.on('data', (chunk: Buffer) => (raw += chunk.toString()));
|
||||
connection.once('end', () => resolve(raw));
|
||||
connection.once('error', reject);
|
||||
});
|
||||
expect(JSON.parse(malformed)).toMatchObject({ ok: false, code: 'MALFORMED_REQUEST' });
|
||||
const registered = await request(socket, {
|
||||
action: 'register_anchor',
|
||||
runtime_generation: 1,
|
||||
nonce: randomUUID(),
|
||||
});
|
||||
expect(registered.ok).toBe(true);
|
||||
});
|
||||
|
||||
test('silent connection deadline cannot prevent the next valid registration', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const silent = createConnection(socket);
|
||||
await new Promise<void>((resolve, reject) => {
|
||||
silent.once('connect', resolve);
|
||||
silent.once('error', reject);
|
||||
});
|
||||
const registered = await withTimeout(
|
||||
request(socket, { action: 'register_anchor', runtime_generation: 1 }),
|
||||
'registration behind silent connection',
|
||||
);
|
||||
expect(registered.ok).toBe(true);
|
||||
silent.destroy();
|
||||
});
|
||||
|
||||
test('queued silent peers cannot serialize the next valid registration', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const silentConnections = await Promise.all(
|
||||
Array.from(
|
||||
{ length: 4 },
|
||||
() =>
|
||||
new Promise<Socket>((resolve, reject) => {
|
||||
const connection = createConnection(socket);
|
||||
connection.once('connect', () => resolve(connection));
|
||||
connection.once('error', reject);
|
||||
}),
|
||||
),
|
||||
);
|
||||
|
||||
try {
|
||||
await new Promise((resolve) => setTimeout(resolve, 100));
|
||||
const started = performance.now();
|
||||
const registered = await withTimeout(
|
||||
request(socket, { action: 'register_anchor', runtime_generation: 1 }),
|
||||
'registration behind queued silent connections',
|
||||
6_000,
|
||||
);
|
||||
const elapsed = performance.now() - started;
|
||||
|
||||
expect(registered.ok).toBe(true);
|
||||
expect(elapsed).toBeLessThan(1_500);
|
||||
} finally {
|
||||
for (const connection of silentConnections) connection.destroy();
|
||||
}
|
||||
});
|
||||
|
||||
test('silent peers are reaped at the concurrency bound and their slots are reclaimed', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const concurrencyCap = 16;
|
||||
const peers = Array.from({ length: concurrencyCap }, () => {
|
||||
const connection = createConnection(socket);
|
||||
return {
|
||||
connection,
|
||||
connected: new Promise<void>((resolve, reject) => {
|
||||
connection.once('connect', resolve);
|
||||
connection.once('error', reject);
|
||||
}),
|
||||
closed: new Promise<void>((resolve) => connection.once('close', () => resolve())),
|
||||
};
|
||||
});
|
||||
|
||||
try {
|
||||
await Promise.all(peers.map(({ connected }) => connected));
|
||||
await new Promise((resolve) => setTimeout(resolve, 100));
|
||||
const started = performance.now();
|
||||
const registration = withTimeout(
|
||||
request(socket, { action: 'register_anchor', runtime_generation: 1 }),
|
||||
'registration while silent peers hold the concurrency bound',
|
||||
2_500,
|
||||
);
|
||||
const reaping = withTimeout(
|
||||
Promise.all(peers.map(({ closed }) => closed)),
|
||||
'silent peer deadline reaping',
|
||||
2_500,
|
||||
);
|
||||
const [registered] = await Promise.all([registration, reaping]);
|
||||
const elapsed = performance.now() - started;
|
||||
|
||||
expect(registered.ok).toBe(true);
|
||||
expect(elapsed).toBeGreaterThan(500);
|
||||
expect(elapsed).toBeLessThan(2_500);
|
||||
} finally {
|
||||
for (const { connection } of peers) connection.destroy();
|
||||
}
|
||||
});
|
||||
|
||||
test('newline-only client without half-close gets no success and cannot block next request', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const incomplete = createConnection(socket);
|
||||
let raw = '';
|
||||
incomplete.setEncoding('utf8');
|
||||
incomplete.on('data', (chunk: string) => (raw += chunk));
|
||||
await new Promise<void>((resolve, reject) => {
|
||||
incomplete.once('connect', () => {
|
||||
incomplete.write(
|
||||
`${JSON.stringify({ action: 'register_anchor', runtime_generation: 1 })}\n`,
|
||||
);
|
||||
resolve();
|
||||
});
|
||||
incomplete.once('error', reject);
|
||||
});
|
||||
await new Promise((resolve) => setTimeout(resolve, 1_100));
|
||||
expect(raw).not.toContain('"ok":true');
|
||||
const registered = await withTimeout(
|
||||
request(socket, { action: 'register_anchor', runtime_generation: 1 }),
|
||||
'registration after non-half-closed client',
|
||||
);
|
||||
expect(registered.ok).toBe(true);
|
||||
incomplete.destroy();
|
||||
});
|
||||
|
||||
test('client disconnect cannot prevent the next valid authentication', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const registered = await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
const reset = createConnection(socket);
|
||||
await new Promise<void>((resolve, reject) => {
|
||||
reset.once('connect', () => {
|
||||
reset.write(`${JSON.stringify({ action: 'register_anchor', runtime_generation: 1 })}\n`);
|
||||
reset.destroy();
|
||||
resolve();
|
||||
});
|
||||
reset.once('error', reject);
|
||||
});
|
||||
const authenticated = await withTimeout(
|
||||
request(socket, {
|
||||
action: 'authenticate',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: 1,
|
||||
}),
|
||||
'authentication after client disconnect',
|
||||
);
|
||||
expect(authenticated.ok).toBe(true);
|
||||
});
|
||||
|
||||
test('delayed second frame is rejected and the next request succeeds', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const reply = await rawRequest(socket, (connection) => {
|
||||
connection.write(`${JSON.stringify({ action: 'register_anchor', runtime_generation: 1 })}\n`);
|
||||
setTimeout(() => connection.end('{}\n'), 50);
|
||||
});
|
||||
expect(reply).toMatchObject({ ok: false, code: 'MALFORMED_REQUEST' });
|
||||
expect(
|
||||
await request(socket, { action: 'register_anchor', runtime_generation: 1 }),
|
||||
).toMatchObject({
|
||||
ok: true,
|
||||
});
|
||||
});
|
||||
|
||||
test('unterminated frame is rejected and the next request succeeds', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const reply = await rawRequest(socket, (connection) => connection.end('{}'));
|
||||
expect(reply).toMatchObject({ ok: false, code: 'MALFORMED_REQUEST' });
|
||||
expect(
|
||||
await request(socket, { action: 'register_anchor', runtime_generation: 1 }),
|
||||
).toMatchObject({
|
||||
ok: true,
|
||||
});
|
||||
});
|
||||
|
||||
test('genuinely oversized frame is rejected and the next request succeeds', async () => {
|
||||
const { socket } = await startBroker();
|
||||
const reply = await rawRequest(socket, (connection) =>
|
||||
connection.end(`${JSON.stringify({ padding: 'x'.repeat(64 * 1024) })}\n`),
|
||||
);
|
||||
expect(reply).toMatchObject({ ok: false, code: 'MALFORMED_REQUEST' });
|
||||
expect(
|
||||
await request(socket, { action: 'register_anchor', runtime_generation: 1 }),
|
||||
).toMatchObject({
|
||||
ok: true,
|
||||
});
|
||||
});
|
||||
|
||||
test('boolean runtime generations fail closed', async () => {
|
||||
const { socket } = await startBroker();
|
||||
expect(
|
||||
await request(socket, { action: 'register_anchor', runtime_generation: true }),
|
||||
).toMatchObject({ ok: false, code: 'INVALID_GENERATION' });
|
||||
const registered = await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
expect(
|
||||
await request(socket, {
|
||||
action: 'authenticate',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: false,
|
||||
}),
|
||||
).toMatchObject({ ok: false, code: 'INVALID_IDENTITY' });
|
||||
});
|
||||
|
||||
test.each([
|
||||
{ compaction_epoch: true, request_epoch: 0, schema_version: 1 },
|
||||
{ compaction_epoch: 0, request_epoch: -1, schema_version: 1 },
|
||||
{ compaction_epoch: 0, request_epoch: 0, schema_version: false },
|
||||
{ compaction_epoch: 0, request_epoch: 0, schema_version: -1 },
|
||||
{ compaction_epoch: 0, request_epoch: 0, schema_version: 1, h_source: 'A'.repeat(64) },
|
||||
{ compaction_epoch: 0, request_epoch: 0, schema_version: 1, h_payload: 'a'.repeat(63) },
|
||||
])('invalid cycle binding fails closed without persisting a token (%j)', async (override) => {
|
||||
const { socket, state } = await startBroker();
|
||||
const registered = await request(socket, { action: 'register_anchor', runtime_generation: 1 });
|
||||
const binding = Object.assign(
|
||||
{
|
||||
compaction_epoch: 0,
|
||||
request_epoch: 0,
|
||||
h_source: 'a'.repeat(64),
|
||||
h_payload: 'b'.repeat(64),
|
||||
schema_version: 1,
|
||||
},
|
||||
override,
|
||||
);
|
||||
expect(
|
||||
await request(socket, {
|
||||
action: 'mint_token',
|
||||
session_id: registered.session_id,
|
||||
runtime_generation: 1,
|
||||
binding,
|
||||
}),
|
||||
).toMatchObject({ ok: false, code: 'INVALID_BINDING' });
|
||||
const persisted = JSON.parse(await readFile(state, 'utf8')) as { tokens: object };
|
||||
expect(persisted.tokens).toEqual({});
|
||||
});
|
||||
|
||||
test('StateStore write-all unit path handles partial writes and cleans failed temp files', () => {
|
||||
const result = spawnSync('python3', [join(import.meta.dirname, 'state_store_unittest.py')], {
|
||||
encoding: 'utf8',
|
||||
});
|
||||
expect(result.status, result.stderr).toBe(0);
|
||||
});
|
||||
|
||||
test('persistence integrity failure refuses startup', async () => {
|
||||
expect(await startBrokerWithState('{corrupt')).toContain('STATE_INTEGRITY');
|
||||
});
|
||||
|
||||
test.each([
|
||||
{ version: 1, sessions: {}, tokens: {}, unexpected: true },
|
||||
{ version: 1, sessions: { bad: {} }, tokens: {} },
|
||||
{
|
||||
version: 1,
|
||||
sessions: {
|
||||
['a'.repeat(64)]: { anchor_pid: true, anchor_starttime: '1', runtime_generation: 0 },
|
||||
},
|
||||
tokens: {},
|
||||
},
|
||||
{
|
||||
version: 1,
|
||||
sessions: {
|
||||
['a'.repeat(64)]: { anchor_pid: 1, anchor_starttime: '01', runtime_generation: 0 },
|
||||
},
|
||||
tokens: {},
|
||||
},
|
||||
{
|
||||
version: 1,
|
||||
sessions: {
|
||||
['a'.repeat(64)]: { anchor_pid: 1, anchor_starttime: '1', runtime_generation: 0 },
|
||||
['b'.repeat(64)]: { anchor_pid: 1, anchor_starttime: '1', runtime_generation: 1 },
|
||||
},
|
||||
tokens: {},
|
||||
},
|
||||
{
|
||||
version: 1,
|
||||
sessions: {
|
||||
['a'.repeat(64)]: { anchor_pid: 1, anchor_starttime: '1', runtime_generation: 0 },
|
||||
},
|
||||
tokens: {
|
||||
['b'.repeat(64)]: {
|
||||
session_id: 'c'.repeat(64),
|
||||
runtime_generation: 0,
|
||||
binding: {
|
||||
compaction_epoch: 0,
|
||||
request_epoch: 0,
|
||||
h_source: 'd'.repeat(64),
|
||||
h_payload: 'e'.repeat(64),
|
||||
schema_version: 1,
|
||||
},
|
||||
consumed: false,
|
||||
},
|
||||
},
|
||||
},
|
||||
{
|
||||
version: 1,
|
||||
sessions: {
|
||||
['a'.repeat(64)]: { anchor_pid: 1, anchor_starttime: '1', runtime_generation: 1 },
|
||||
},
|
||||
tokens: {
|
||||
['b'.repeat(64)]: {
|
||||
session_id: 'a'.repeat(64),
|
||||
runtime_generation: 2,
|
||||
binding: {
|
||||
compaction_epoch: 0,
|
||||
request_epoch: 0,
|
||||
h_source: 'd'.repeat(64),
|
||||
h_payload: 'e'.repeat(64),
|
||||
schema_version: 1,
|
||||
},
|
||||
consumed: false,
|
||||
},
|
||||
},
|
||||
},
|
||||
])('nested corrupt state refuses startup (%#)', async (stateValue) => {
|
||||
expect(await startBrokerWithState(JSON.stringify(stateValue))).toContain('STATE_INTEGRITY');
|
||||
});
|
||||
|
||||
test('symlink state refuses startup', async () => {
|
||||
const root = await mkdtemp(join(tmpdir(), 'mosaic-lease-broker-'));
|
||||
await chmod(root, 0o700);
|
||||
const target = join(root, 'target.json');
|
||||
const state = join(root, 'state.json');
|
||||
await writeFile(target, JSON.stringify({ version: 1, sessions: {}, tokens: {} }), {
|
||||
mode: 0o600,
|
||||
});
|
||||
await symlink(target, state);
|
||||
const child = spawn('python3', [
|
||||
daemonPath,
|
||||
'--socket',
|
||||
join(root, 'broker.sock'),
|
||||
'--state',
|
||||
state,
|
||||
]);
|
||||
children.push(child);
|
||||
const stderr = await new Promise<string>((resolve) => {
|
||||
let raw = '';
|
||||
child.stderr?.on('data', (chunk: Buffer) => (raw += chunk.toString()));
|
||||
child.once('exit', () => resolve(raw));
|
||||
});
|
||||
expect(stderr).toContain('STATE_INTEGRITY');
|
||||
});
|
||||
|
||||
test('oversized state refuses startup', async () => {
|
||||
expect(await startBrokerWithState(' '.repeat(4 * 1024 * 1024 + 1))).toContain(
|
||||
'STATE_INTEGRITY',
|
||||
);
|
||||
});
|
||||
});
|
||||
333
packages/mosaic/src/lease-broker/state_store_unittest.py
Normal file
333
packages/mosaic/src/lease-broker/state_store_unittest.py
Normal file
@@ -0,0 +1,333 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Standard-library edge tests for lease-broker atomic state persistence."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import copy
|
||||
import importlib.util
|
||||
import json
|
||||
import os
|
||||
import tempfile
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
from unittest.mock import patch
|
||||
|
||||
|
||||
DAEMON_PATH = Path(__file__).parents[2] / "framework/tools/lease-broker/daemon.py"
|
||||
SPEC = importlib.util.spec_from_file_location("lease_broker_daemon", DAEMON_PATH)
|
||||
if SPEC is None or SPEC.loader is None:
|
||||
raise RuntimeError("unable to load lease broker daemon")
|
||||
DAEMON = importlib.util.module_from_spec(SPEC)
|
||||
SPEC.loader.exec_module(DAEMON)
|
||||
|
||||
|
||||
class StateStoreCommitTest(unittest.TestCase):
|
||||
def make_store(self, root: Path):
|
||||
os.chmod(root, 0o700)
|
||||
store = DAEMON.StateStore(root / "state.json")
|
||||
store.value["marker"] = "partial-write-proof"
|
||||
return store
|
||||
|
||||
def test_partial_writes_persist_the_complete_payload(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory:
|
||||
root = Path(directory)
|
||||
store = self.make_store(root)
|
||||
real_write = os.write
|
||||
|
||||
def partial_write(descriptor: int, payload: bytes) -> int:
|
||||
return real_write(descriptor, payload[: max(1, len(payload) // 3)])
|
||||
|
||||
with patch.object(DAEMON.os, "write", side_effect=partial_write):
|
||||
store.commit()
|
||||
|
||||
self.assertEqual(json.loads(store.path.read_text()), store.value)
|
||||
|
||||
def test_zero_progress_removes_owned_temporary_file(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory:
|
||||
root = Path(directory)
|
||||
store = self.make_store(root)
|
||||
with patch.object(DAEMON.os, "write", return_value=0):
|
||||
with self.assertRaises(OSError):
|
||||
store.commit()
|
||||
|
||||
self.assertFalse(store.path.exists())
|
||||
self.assertEqual(list(root.glob(".*.tmp")), [])
|
||||
|
||||
def test_oversized_payload_is_refused_before_replacing_state(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory:
|
||||
root = Path(directory)
|
||||
store = self.make_store(root)
|
||||
store.value.pop("marker")
|
||||
store.commit()
|
||||
durable = store.path.read_bytes()
|
||||
store.value["oversized"] = "x" * DAEMON.MAX_STATE
|
||||
|
||||
with patch.object(DAEMON.os, "open", wraps=os.open) as mocked_open:
|
||||
with self.assertRaisesRegex(DAEMON.BrokerFailure, "STATE_TOO_LARGE"):
|
||||
store.commit()
|
||||
|
||||
self.assertEqual(mocked_open.call_count, 0)
|
||||
self.assertEqual(store.path.read_bytes(), durable)
|
||||
self.assertEqual(list(root.glob(".*.tmp")), [])
|
||||
|
||||
|
||||
class StateStoreValidationTest(unittest.TestCase):
|
||||
@staticmethod
|
||||
def binding() -> dict[str, object]:
|
||||
return {
|
||||
"compaction_epoch": 0,
|
||||
"request_epoch": 0,
|
||||
"h_source": "a" * 64,
|
||||
"h_payload": "b" * 64,
|
||||
"schema_version": 1,
|
||||
}
|
||||
|
||||
def test_impossible_or_over_capacity_token_state_is_rejected(self) -> None:
|
||||
session_id = "1" * 64
|
||||
session = {
|
||||
"anchor_pid": 123,
|
||||
"anchor_starttime": "456",
|
||||
"runtime_generation": 2,
|
||||
}
|
||||
live_token = {
|
||||
"session_id": session_id,
|
||||
"runtime_generation": 2,
|
||||
"binding": self.binding(),
|
||||
"consumed": False,
|
||||
}
|
||||
cases = {
|
||||
"stale generation": {
|
||||
"2" * 64: {**live_token, "runtime_generation": 1},
|
||||
},
|
||||
"consumed token": {
|
||||
"2" * 64: {**live_token, "consumed": True},
|
||||
},
|
||||
"over capacity": {
|
||||
f"{index:064x}": copy.deepcopy(live_token)
|
||||
for index in range(DAEMON.MAX_PENDING_TOKENS + 1)
|
||||
},
|
||||
}
|
||||
|
||||
for label, tokens in cases.items():
|
||||
with self.subTest(label=label), tempfile.TemporaryDirectory() as directory:
|
||||
root = Path(directory)
|
||||
os.chmod(root, 0o700)
|
||||
state_path = root / "state.json"
|
||||
state_path.write_text(json.dumps({
|
||||
"version": 1,
|
||||
"sessions": {session_id: session},
|
||||
"tokens": tokens,
|
||||
}))
|
||||
os.chmod(state_path, 0o600)
|
||||
|
||||
with self.assertRaisesRegex(DAEMON.BrokerFailure, "STATE_INTEGRITY"):
|
||||
DAEMON.StateStore(state_path)
|
||||
|
||||
|
||||
class BrokerBehaviorTest(unittest.TestCase):
|
||||
def make_broker(self, root: Path):
|
||||
os.chmod(root, 0o700)
|
||||
return DAEMON.Broker(DAEMON.StateStore(root / "state.json"))
|
||||
|
||||
@staticmethod
|
||||
def binding() -> dict[str, object]:
|
||||
return {
|
||||
"compaction_epoch": 0,
|
||||
"request_epoch": 0,
|
||||
"h_source": "a" * 64,
|
||||
"h_payload": "b" * 64,
|
||||
"schema_version": 1,
|
||||
}
|
||||
|
||||
def register(self, broker, generation: int = 1) -> str:
|
||||
response = broker.handle((123, 1000, 1000), {
|
||||
"action": "register_anchor",
|
||||
"runtime_generation": generation,
|
||||
})
|
||||
return response["session_id"]
|
||||
|
||||
def mint(self, broker, session_id: str, generation: int = 1) -> str:
|
||||
response = broker.handle((123, 1000, 1000), {
|
||||
"action": "mint_token",
|
||||
"session_id": session_id,
|
||||
"runtime_generation": generation,
|
||||
"binding": self.binding(),
|
||||
})
|
||||
return response["token"]
|
||||
|
||||
def consume(self, broker, session_id: str, token: str, generation: int = 1):
|
||||
return broker.handle((123, 1000, 1000), {
|
||||
"action": "consume_token",
|
||||
"session_id": session_id,
|
||||
"runtime_generation": generation,
|
||||
"token": token,
|
||||
})
|
||||
|
||||
def test_anchor_generation_bump_reuses_session_and_revokes_token(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory:
|
||||
root = Path(directory)
|
||||
broker = self.make_broker(root)
|
||||
with (
|
||||
patch.object(
|
||||
DAEMON,
|
||||
"proc_node",
|
||||
return_value={"pid": 123, "ppid": 1, "starttime": "456"},
|
||||
),
|
||||
patch.object(DAEMON, "verified_ancestry", return_value=True),
|
||||
):
|
||||
first = broker.handle((123, 1000, 1000), {
|
||||
"action": "register_anchor",
|
||||
"runtime_generation": 1,
|
||||
})
|
||||
minted = broker.handle((123, 1000, 1000), {
|
||||
"action": "mint_token",
|
||||
"session_id": first["session_id"],
|
||||
"runtime_generation": 1,
|
||||
"binding": self.binding(),
|
||||
})
|
||||
bumped = broker.handle((123, 1000, 1000), {
|
||||
"action": "register_anchor",
|
||||
"runtime_generation": 2,
|
||||
})
|
||||
repeated = broker.handle((123, 1000, 1000), {
|
||||
"action": "register_anchor",
|
||||
"runtime_generation": 2,
|
||||
})
|
||||
|
||||
self.assertEqual(bumped["session_id"], first["session_id"])
|
||||
self.assertEqual(repeated["session_id"], first["session_id"])
|
||||
self.assertNotIn(minted["token"], broker.store.tokens())
|
||||
restarted = self.make_broker(root)
|
||||
self.assertEqual(restarted.store.tokens(), {})
|
||||
self.assertEqual(
|
||||
restarted.store.sessions()[first["session_id"]]["runtime_generation"], 2
|
||||
)
|
||||
with self.assertRaisesRegex(DAEMON.BrokerFailure, "STALE_GENERATION"):
|
||||
with patch.object(DAEMON, "proc_node", return_value={"starttime": "456"}):
|
||||
broker.handle((123, 1000, 1000), {
|
||||
"action": "register_anchor",
|
||||
"runtime_generation": 1,
|
||||
})
|
||||
|
||||
def test_successful_consume_deletes_token_and_replay_is_refused(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory, (
|
||||
patch.object(DAEMON, "proc_node", return_value={"pid": 123, "ppid": 1, "starttime": "456"})
|
||||
), patch.object(DAEMON, "verified_ancestry", return_value=True):
|
||||
broker = self.make_broker(Path(directory))
|
||||
session_id = self.register(broker)
|
||||
token = self.mint(broker, session_id)
|
||||
|
||||
self.assertEqual(self.consume(broker, session_id, token), {"ok": True})
|
||||
self.assertNotIn(token, broker.store.tokens())
|
||||
with self.assertRaisesRegex(DAEMON.BrokerFailure, "TOKEN_REPLAY"):
|
||||
self.consume(broker, session_id, token)
|
||||
|
||||
def test_normal_cycles_remain_bounded_and_restartable(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory, (
|
||||
patch.object(DAEMON, "proc_node", return_value={"pid": 123, "ppid": 1, "starttime": "456"})
|
||||
), patch.object(DAEMON, "verified_ancestry", return_value=True):
|
||||
root = Path(directory)
|
||||
broker = self.make_broker(root)
|
||||
session_id = self.register(broker)
|
||||
|
||||
for _ in range(DAEMON.MAX_PENDING_TOKENS * 3):
|
||||
self.consume(broker, session_id, self.mint(broker, session_id))
|
||||
|
||||
self.assertEqual(broker.store.tokens(), {})
|
||||
self.assertLess((root / "state.json").stat().st_size, DAEMON.MAX_STATE)
|
||||
restarted = self.make_broker(root)
|
||||
self.assertEqual(restarted.store.tokens(), {})
|
||||
self.assertIn(session_id, restarted.store.sessions())
|
||||
|
||||
def test_pending_token_capacity_refusal_does_not_mutate_state(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory, (
|
||||
patch.object(DAEMON, "proc_node", return_value={"pid": 123, "ppid": 1, "starttime": "456"})
|
||||
), patch.object(DAEMON, "verified_ancestry", return_value=True):
|
||||
root = Path(directory)
|
||||
broker = self.make_broker(root)
|
||||
session_id = self.register(broker)
|
||||
for _ in range(DAEMON.MAX_PENDING_TOKENS):
|
||||
self.mint(broker, session_id)
|
||||
before = copy.deepcopy(broker.store.value)
|
||||
durable = broker.store.path.read_bytes()
|
||||
|
||||
with self.assertRaisesRegex(DAEMON.BrokerFailure, "TOKEN_CAPACITY"):
|
||||
self.mint(broker, session_id)
|
||||
|
||||
self.assertEqual(broker.store.value, before)
|
||||
self.assertEqual(broker.store.path.read_bytes(), durable)
|
||||
|
||||
def test_directory_fsync_failure_poisoned_store_cannot_continue(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory, patch.object(
|
||||
DAEMON,
|
||||
"proc_node",
|
||||
return_value={"pid": 123, "ppid": 1, "starttime": "456"},
|
||||
):
|
||||
root = Path(directory)
|
||||
broker = self.make_broker(root)
|
||||
real_fsync = os.fsync
|
||||
|
||||
def fail_directory_fsync(descriptor: int) -> None:
|
||||
if os.path.isdir(f"/proc/self/fd/{descriptor}"):
|
||||
raise OSError("directory fsync failed")
|
||||
real_fsync(descriptor)
|
||||
|
||||
with patch.object(DAEMON.os, "fsync", side_effect=fail_directory_fsync):
|
||||
with self.assertRaisesRegex(
|
||||
DAEMON.StateCommitUncertain, "STATE_COMMIT_UNCERTAIN"
|
||||
):
|
||||
self.register(broker)
|
||||
|
||||
durable = json.loads(broker.store.path.read_text())
|
||||
self.assertEqual(broker.store.value, durable)
|
||||
self.assertTrue(broker.store.poisoned)
|
||||
before = copy.deepcopy(broker.store.value)
|
||||
with self.assertRaisesRegex(
|
||||
DAEMON.StateCommitUncertain, "STATE_COMMIT_UNCERTAIN"
|
||||
):
|
||||
broker.handle((123, 1000, 1000), {
|
||||
"action": "register_anchor",
|
||||
"runtime_generation": 2,
|
||||
})
|
||||
self.assertEqual(broker.store.value, before)
|
||||
|
||||
def test_commit_failures_before_replace_roll_back_every_broker_mutation(self) -> None:
|
||||
with tempfile.TemporaryDirectory() as directory, (
|
||||
patch.object(DAEMON, "proc_node", return_value={"pid": 123, "ppid": 1, "starttime": "456"})
|
||||
), patch.object(DAEMON, "verified_ancestry", return_value=True):
|
||||
root = Path(directory)
|
||||
broker = self.make_broker(root)
|
||||
session_id = self.register(broker)
|
||||
token = self.mint(broker, session_id)
|
||||
|
||||
def assert_rollback(request: dict[str, object]) -> None:
|
||||
before = copy.deepcopy(broker.store.value)
|
||||
durable = broker.store.path.read_bytes()
|
||||
with patch.object(broker.store, "commit", side_effect=OSError("fsync failed")):
|
||||
with self.assertRaisesRegex(OSError, "fsync failed"):
|
||||
broker.handle((123, 1000, 1000), request)
|
||||
self.assertEqual(broker.store.value, before)
|
||||
self.assertEqual(broker.store.path.read_bytes(), durable)
|
||||
|
||||
assert_rollback({"action": "register_anchor", "runtime_generation": 2})
|
||||
assert_rollback({
|
||||
"action": "mint_token", "session_id": session_id,
|
||||
"runtime_generation": 1, "binding": self.binding(),
|
||||
})
|
||||
assert_rollback({
|
||||
"action": "consume_token", "session_id": session_id,
|
||||
"runtime_generation": 1, "token": token,
|
||||
})
|
||||
|
||||
with tempfile.TemporaryDirectory() as second_directory:
|
||||
second = self.make_broker(Path(second_directory))
|
||||
with patch.object(second.store, "commit", side_effect=OSError("fsync failed")):
|
||||
with self.assertRaisesRegex(OSError, "fsync failed"):
|
||||
self.register(second)
|
||||
self.assertEqual(
|
||||
second.store.value, {"version": 1, "sessions": {}, "tokens": {}}
|
||||
)
|
||||
self.assertFalse(second.store.path.exists())
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user