server: commit operations before enqueueing and answer seq_no races with PREV_HASH_MISMATCH
The operation and receipt now commit before the queue send. A Postgres 23505 on operations_agent_id_seq_no_key is answered with 400 PREV_HASH_MISMATCH carrying the stored chain hash so hook retries can resync. A failed enqueue after the commit is logged as operation.enqueue_failed and the receipt is still returned. Every producer now names its message: MessageQueue.send takes the durable message id, and the BullMQ adapter passes it as the job id, so the receipt's queue_message_id is the real queue identifier. Claude-Session: https://claude.ai/code/session_01QoUjdc7HGKTsgj5GhG8HvF
This commit is contained in:
1 parent
060a072f1f
commit
1bb9d4d49d
7 files changed
+158
-20
No files matched your search
@@ -56,6 +56,7 @@ This repository publishes the open-source server, console, integration catalog,
|
||||
- Keep production source files at or below 500 lines.
|
||||
- Keep functions focused on one ownership boundary.
|
||||
- Propagate unexpected errors to the CLI or request boundary.
|
||||
- Commit the operation and receipt before enqueueing. Log a failed enqueue as `operation.enqueue_failed` and still return the receipt; answer a `UNIQUE(agent_id, seq_no)` conflict with `PREV_HASH_MISMATCH` carrying the stored chain hash.
|
||||
- Keep technical claims executable through tests or generated validation.
|
||||
- Keep the mirrored Python package, distribution, and CLI version in `sdks/python/elydora/_version.py`; Setuptools metadata must read that literal dynamically.
|
||||
- Ship `sdks/python/elydora/py.typed` in every mirrored wheel and verify public annotations from an installed-wheel consumer.
|
||||
|
||||
@@ -9,7 +9,6 @@
|
||||
import { Queue as BullQueue } from 'bullmq';
|
||||
import type { ConnectionOptions } from 'bullmq';
|
||||
import type { MessageQueue } from './interfaces.js';
|
||||
import { generateUUIDv7 } from '../utils/uuid.js';
|
||||
|
||||
export const QUEUE_NAME = 'elydora-queue';
|
||||
|
||||
@@ -20,14 +19,14 @@ export class BullMQAdapter implements MessageQueue {
|
||||
this.queue = new BullQueue(QUEUE_NAME, { connection });
|
||||
}
|
||||
|
||||
async send(body: unknown): Promise<{ messageId: string }> {
|
||||
const job = await this.queue.add('message', body as Record<string, unknown>, {
|
||||
async send(messageId: string, body: unknown): Promise<void> {
|
||||
await this.queue.add('message', body as Record<string, unknown>, {
|
||||
jobId: messageId,
|
||||
attempts: 3,
|
||||
backoff: { type: 'exponential', delay: 1000 },
|
||||
removeOnComplete: { count: 100 },
|
||||
removeOnFail: { count: 500 },
|
||||
});
|
||||
return { messageId: job.id ?? generateUUIDv7() };
|
||||
}
|
||||
|
||||
async close(): Promise<void> {
|
||||
|
||||
@@ -117,6 +117,6 @@ export interface Cache {
|
||||
* Mirrors the subset of Cloudflare Queue methods used by Elydora services.
|
||||
*/
|
||||
export interface MessageQueue {
|
||||
/** Enqueue a message. Returns the assigned message ID. */
|
||||
send(body: unknown): Promise<{ messageId: string }>;
|
||||
/** Enqueue a message under the caller's durable message ID. */
|
||||
send(messageId: string, body: unknown): Promise<void>;
|
||||
}
|
||||
@@ -12,6 +12,7 @@ import { RedisCacheAdapter } from './adapters/redis-cache.js';
|
||||
import { BullMQAdapter } from './adapters/bullmq.js';
|
||||
import { createEpoch } from './services/epoch-service.js';
|
||||
import { DEFAULT_EPOCH_INTERVAL_MS } from './shared/index.js';
|
||||
import { generateUUIDv7 } from './utils/uuid.js';
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Initialize adapters
|
||||
@@ -65,7 +66,7 @@ async function runEpochCreation(): Promise<void> {
|
||||
|
||||
if (epoch) {
|
||||
try {
|
||||
await queue.send({
|
||||
await queue.send(generateUUIDv7(), {
|
||||
type: 'tsa',
|
||||
epoch_id: epoch.epoch_id,
|
||||
org_id: epoch.org_id,
|
||||
|
||||
@@ -99,7 +99,7 @@ export async function createExport(
|
||||
.run();
|
||||
|
||||
// Enqueue the export job
|
||||
await queue.send({
|
||||
await queue.send(generateUUIDv7(), {
|
||||
type: 'export',
|
||||
export_id: exportId,
|
||||
org_id: orgId,
|
||||
|
||||
@@ -201,20 +201,10 @@ export async function submitOperation(
|
||||
});
|
||||
|
||||
// ------------------------------------------------------------------
|
||||
// Step 10: Enqueue for async processing
|
||||
// Step 10: Reserve the queue message ID; the send follows the commit
|
||||
// ------------------------------------------------------------------
|
||||
const queueMessageResult = await queue.send({
|
||||
type: 'operation',
|
||||
operation_id: eor.operation_id,
|
||||
org_id: eor.org_id,
|
||||
agent_id: eor.agent_id,
|
||||
});
|
||||
|
||||
// D1 batch insert: operation + receipt
|
||||
const queueMessageId = generateUUIDv7();
|
||||
const receiptId = generateUUIDv7();
|
||||
const queueMessageId = typeof queueMessageResult === 'object' && queueMessageResult !== null
|
||||
? String((queueMessageResult as Record<string, unknown>).messageId ?? receiptId)
|
||||
: receiptId;
|
||||
|
||||
// ------------------------------------------------------------------
|
||||
// Step 11: Generate EAR receipt
|
||||
@@ -291,11 +281,45 @@ export async function submitOperation(
|
||||
.bind(receiptId, eor.operation_id, r2ReceiptKey, receivedAt),
|
||||
);
|
||||
|
||||
try {
|
||||
await db.batch(statements);
|
||||
} catch (error) {
|
||||
if (!isSeqNoConflict(error)) throw error;
|
||||
const winner = await db
|
||||
.prepare('SELECT chain_hash FROM operations WHERE agent_id = ? ORDER BY seq_no DESC LIMIT 1')
|
||||
.bind(eor.agent_id)
|
||||
.first<{ chain_hash: string }>();
|
||||
if (!winner) throw new Error('seq_no conflict without a stored operation.');
|
||||
throw new AppError(400, 'PREV_HASH_MISMATCH', {
|
||||
key: 'operation.prevHashMismatch',
|
||||
params: { expected: winner.chain_hash, actual: eor.prev_chain_hash },
|
||||
});
|
||||
}
|
||||
|
||||
try {
|
||||
await queue.send(queueMessageId, {
|
||||
type: 'operation',
|
||||
operation_id: eor.operation_id,
|
||||
org_id: eor.org_id,
|
||||
agent_id: eor.agent_id,
|
||||
});
|
||||
} catch (error) {
|
||||
console.error('operation.enqueue_failed', {
|
||||
operation_id: eor.operation_id,
|
||||
queue_message_id: queueMessageId,
|
||||
error: error instanceof Error ? error.message : String(error),
|
||||
});
|
||||
}
|
||||
|
||||
return { receipt: ear };
|
||||
}
|
||||
|
||||
function isSeqNoConflict(error: unknown): boolean {
|
||||
if (typeof error !== 'object' || error === null) return false;
|
||||
const { code, constraint } = error as { code?: unknown; constraint?: unknown };
|
||||
return code === '23505' && constraint === 'operations_agent_id_seq_no_key';
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Get operation
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
@@ -0,0 +1,113 @@
|
||||
import assert from 'node:assert/strict';
|
||||
import { generateKeyPairSync, randomUUID } from 'node:crypto';
|
||||
import test from 'node:test';
|
||||
|
||||
import { AppError } from '../dist/middleware/error-handler.js';
|
||||
import { submitOperation } from '../dist/services/operation-service.js';
|
||||
import { jcsCanonicalise, sha256Base64url, signEd25519 } from '../dist/utils/crypto.js';
|
||||
|
||||
const GENESIS = 'AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA';
|
||||
const WINNER = 'Rxlf4j36C3KvIQ3hWuOkX698BR5iDypUFuB70JjEuvM';
|
||||
|
||||
function keyPair() {
|
||||
const pair = generateKeyPairSync('ed25519');
|
||||
return {
|
||||
seed: pair.privateKey.export({ format: 'jwk' }).d,
|
||||
publicKey: pair.publicKey.export({ format: 'jwk' }).x,
|
||||
};
|
||||
}
|
||||
|
||||
async function signedEor(agentId, orgId, kid, seed) {
|
||||
const payload = { tool: 'Bash' };
|
||||
const unsigned = {
|
||||
op_version: '1.0',
|
||||
operation_id: randomUUID(),
|
||||
org_id: orgId,
|
||||
agent_id: agentId,
|
||||
issued_at: Date.now(),
|
||||
ttl_ms: 30_000,
|
||||
nonce: randomUUID(),
|
||||
operation_type: 'ai.tool_use',
|
||||
subject: { session_id: 'unit' },
|
||||
action: { tool: 'Bash' },
|
||||
payload,
|
||||
payload_hash: await sha256Base64url(jcsCanonicalise(payload)),
|
||||
prev_chain_hash: GENESIS,
|
||||
agent_pubkey_kid: kid,
|
||||
};
|
||||
const signature = await signEd25519(seed, new TextEncoder().encode(jcsCanonicalise(unsigned)));
|
||||
return { ...unsigned, signature };
|
||||
}
|
||||
|
||||
function pgConflict() {
|
||||
return Object.assign(new Error('duplicate key value violates unique constraint "operations_agent_id_seq_no_key"'), {
|
||||
code: '23505',
|
||||
constraint: 'operations_agent_id_seq_no_key',
|
||||
});
|
||||
}
|
||||
|
||||
function fakes(batchError, agentId, orgId, kid, publicKey) {
|
||||
const sent = [];
|
||||
const db = {
|
||||
prepare: (sql) => ({
|
||||
bind: () => ({
|
||||
first: async () => {
|
||||
if (sql.includes('FROM agents')) return { agent_id: agentId, org_id: orgId, status: 'active' };
|
||||
if (sql.includes('FROM agent_keys')) return { kid, agent_id: agentId, public_key: publicKey, status: 'active' };
|
||||
if (sql.includes('SELECT chain_hash, seq_no')) return null;
|
||||
if (sql.includes('SELECT chain_hash FROM operations')) return { chain_hash: WINNER };
|
||||
throw new Error(`unexpected query: ${sql}`);
|
||||
},
|
||||
}),
|
||||
}),
|
||||
batch: async () => {
|
||||
if (batchError) throw batchError;
|
||||
return [];
|
||||
},
|
||||
};
|
||||
const r2 = { put: async () => undefined };
|
||||
const kv = { get: async () => null, put: async () => undefined };
|
||||
const queue = { send: async (messageId, body) => { sent.push({ messageId, body }); } };
|
||||
return { db, r2, kv, queue, sent };
|
||||
}
|
||||
|
||||
test('a seq_no conflict on insert reports the winning chain hash as PREV_HASH_MISMATCH', async () => {
|
||||
const agent = keyPair();
|
||||
const server = keyPair();
|
||||
const eor = await signedEor('agent-conflict', 'org-conflict', 'agent-conflict-key-1', agent.seed);
|
||||
const { db, r2, kv, queue, sent } = fakes(pgConflict(), 'agent-conflict', 'org-conflict', 'agent-conflict-key-1', agent.publicKey);
|
||||
|
||||
await assert.rejects(
|
||||
submitOperation(db, r2, kv, queue, eor, server.seed),
|
||||
(error) => error instanceof AppError
|
||||
&& error.statusCode === 400
|
||||
&& error.errorCode === 'PREV_HASH_MISMATCH'
|
||||
&& error.message.includes(`Expected prev_chain_hash "${WINNER}"`),
|
||||
);
|
||||
assert.equal(sent.length, 0);
|
||||
});
|
||||
|
||||
test('other insert failures propagate unchanged and nothing is enqueued', async () => {
|
||||
const agent = keyPair();
|
||||
const server = keyPair();
|
||||
const eor = await signedEor('agent-conflict', 'org-conflict', 'agent-conflict-key-1', agent.seed);
|
||||
const { db, r2, kv, queue, sent } = fakes(new Error('connection reset'), 'agent-conflict', 'org-conflict', 'agent-conflict-key-1', agent.publicKey);
|
||||
|
||||
await assert.rejects(
|
||||
submitOperation(db, r2, kv, queue, eor, server.seed),
|
||||
(error) => error instanceof Error && !(error instanceof AppError) && error.message === 'connection reset',
|
||||
);
|
||||
assert.equal(sent.length, 0);
|
||||
});
|
||||
|
||||
test('a committed operation is enqueued with its reserved queue message id', async () => {
|
||||
const agent = keyPair();
|
||||
const server = keyPair();
|
||||
const eor = await signedEor('agent-conflict', 'org-conflict', 'agent-conflict-key-1', agent.seed);
|
||||
const { db, r2, kv, queue, sent } = fakes(null, 'agent-conflict', 'org-conflict', 'agent-conflict-key-1', agent.publicKey);
|
||||
|
||||
const { receipt } = await submitOperation(db, r2, kv, queue, eor, server.seed);
|
||||
assert.equal(sent.length, 1);
|
||||
assert.equal(sent[0].messageId, receipt.queue_message_id);
|
||||
assert.equal(sent[0].body.operation_id, eor.operation_id);
|
||||
});
|
||||
Reference in new issue
Block a user