fix(realtime): bound the debug-log sink and stop blocking on its writes

Ported onto the split provider layout: the route now lives in
server/providers/openai/debug-log.js and its state belongs to the handler
the factory returns rather than to module scope.

/api/realtime/debug-log is the one route in its family left unhardened.
Both siblings enforce a limiter through openai/rate-limit.js, and this was
the only handler still writing with fs.appendFileSync on the request path.

- Always-on per-IP limiter (120/min, 400 global). Deliberately not the
  opt-in openAiRateLimiter(): that one defaults to unlimited so throttling
  cannot silently change what a user spends, and this sink spends disk and
  event-loop time rather than provider quota.
- Serialized async appends with a 32 MB rotation ceiling and one retained
  generation. REALTIME_DEBUG_LOG_MAX_BYTES caps a single request body; it
  never bounded the file those bodies accumulate into, so any local page
  could grow .gev-logs/realtime-conversations.jsonl for as long as the dev
  server ran. Serializing also closes a read-then-write race on the size
  check, where concurrent callers all stat under the ceiling and none
  rotates.
- Generic 400/500 answers separating a malformed record or oversized body
  from a failed write, instead of echoing error.message — an errno and an
  absolute path for an fs failure.
- readRequestBody marks an oversized body with code 'BODY_TOO_LARGE', the
  marker its sibling readRequestBodyCapped already sets, so the two cases
  are distinguishable without matching on text.

Coverage extends src/tooling/localServices.test.mjs, which already drives
the real handler over a temporary sourceRoot. Both new tests fail on the
previous implementation and pass with this one.

Refs #23. Does not add authentication: these proxies have no auth
mechanism today, so introducing one is a maintainer design decision.
This commit is contained in:
Pedro Lobato committed 2026-09-14 21:24:12 -05:00
1 parent 96a6b49477
commit 44817e1286
5 files changed
+154 -16

No files matched your search

+70 -15
View File
@@ -1,16 +1,61 @@
import { defaultSourceRoot } from '../common/source-root.js';
import path from 'node:path';
import { readRequestBody } from '../common/request.js';
import fs from 'node:fs';
import { promises as fsp } from 'node:fs';
import { makeRateLimiter, clientKey } from '../common/rate-limit.js';
/** Cap on one request body. */
const REALTIME_DEBUG_LOG_MAX_BYTES = 8 * 1024 * 1024;
/**
* Ceiling for the live log before it rotates. One generation is kept
* (`…jsonl.1`), so the sink is bounded at twice this however long a session
* runs. REALTIME_DEBUG_LOG_MAX_BYTES caps a single request; it never bounded
* the file those requests accumulate into.
*/
const REALTIME_DEBUG_LOG_MAX_FILE_BYTES = 32 * 1024 * 1024;
/**
* Per-IP write ceiling. Always on, unlike the opt-in limiter the cost-bearing
* OpenAI routes share: throttling those by default would change what a user
* spends, while this sink spends disk and event-loop time. A voice session
* writes far below this, so the cap needs no configuration to stay invisible.
*/
const REALTIME_DEBUG_LOG_MAX_PER_MIN = 120;
function createDebugLogHandler({ sourceRoot = defaultSourceRoot } = {}) {
const REALTIME_DEBUG_LOG_DIR = path.join(sourceRoot, '.gev-logs');
const REALTIME_DEBUG_LOG_FILE = path.join(
REALTIME_DEBUG_LOG_DIR,
'realtime-conversations.jsonl',
);
const logDir = path.join(sourceRoot, '.gev-logs');
const logFile = path.join(logDir, 'realtime-conversations.jsonl');
const allow = makeRateLimiter({
windowMs: 60_000,
max: REALTIME_DEBUG_LOG_MAX_PER_MIN,
globalMax: 400,
});
// Appends are chained rather than fired concurrently because the rotation
// check is a read-then-write on file size: run in parallel, every caller
// stats a file still under the ceiling, none of them rotates, and the log
// grows past it. The records themselves are O_APPEND and never interleave.
let queue = Promise.resolve();
function append(line) {
const done = queue.then(async () => {
await fsp.mkdir(logDir, { recursive: true });
const size = await fsp.stat(logFile).then(
(stat) => stat.size,
() => 0,
);
if (size + Buffer.byteLength(line) > REALTIME_DEBUG_LOG_MAX_FILE_BYTES) {
await fsp.rename(logFile, `${logFile}.1`);
}
await fsp.appendFile(logFile, line);
});
// The caller still sees its own rejection, but one failed append must not
// poison the chain for every request queued behind it.
queue = done.catch(() => {});
return done;
}
return async (req, res) => {
if (req.method !== 'POST') {
res.statusCode = 405;
@@ -19,12 +64,19 @@ function createDebugLogHandler({ sourceRoot = defaultSourceRoot } = {}) {
return;
}
if (!allow(clientKey(req))) {
res.writeHead(429, {
'Content-Type': 'application/json',
'Retry-After': '5',
});
res.end(JSON.stringify({ error: 'Rate limit exceeded' }));
return;
}
try {
const body = await readRequestBody(req, REALTIME_DEBUG_LOG_MAX_BYTES);
const record = JSON.parse(body || '{}');
fs.mkdirSync(REALTIME_DEBUG_LOG_DIR, { recursive: true });
fs.appendFileSync(
REALTIME_DEBUG_LOG_FILE,
await append(
`${JSON.stringify({
loggedAt: new Date().toISOString(),
...record,
@@ -33,13 +85,16 @@ function createDebugLogHandler({ sourceRoot = defaultSourceRoot } = {}) {
res.statusCode = 204;
res.end();
} catch (error) {
res.statusCode = 400;
// A malformed record or an oversized body is the caller's fault; a failed
// write is ours, and only the second is worth a log line. Neither answer
// carries the error text, which for an fs failure is an errno and an
// absolute path.
const callerFault =
error instanceof SyntaxError || error?.code === 'BODY_TOO_LARGE';
if (!callerFault) console.warn('[realtime-debug-log] append failed');
res.statusCode = callerFault ? 400 : 500;
res.setHeader('Content-Type', 'application/json');
res.end(
JSON.stringify({
error: error?.message || 'Failed to write Realtime debug log',
}),
);
res.end(JSON.stringify({ error: 'Failed to write Realtime debug log' }));
}
};
}