Risk of data corruption when using Node v22.7.0 and the MongoDB Node.js driver

MongoDB issued the alert titled “Node.js v22.7.0 runtime defect can lead to data encoding issues” on September 10, 2024 If you need detection or remediation scripts, you can find them there.

Within days of the Node v22.7.0 release, users were already reporting that UTF-8 encodings were broken. The issue results from the introduction of an incorrect optimization for buffer.write which can result in strings being encoded using ISO-8859-1 rather than UTF-8.

Though the use of the fast API for buffer.write will be disabled with Node v22.8.0, developers using MongoDB’s Node.js driver could experience data corruption with Node v22.7.0.

Mitigating the Issue

To avoid the possibility of data corruption due to this bug it is recommended that Node v22.7.0 is not used at all.

MongoDB recommends only using Node runtime versions documented as compatible in production environments. At the time of writing, Node v22.x is not considered a compatible runtime for use with the MongoDB Node.js driver.

How it Occurs

To illustrate how this can occur, consider the following reproduction:

import { MongoClient } from "mongodb";

const client = new MongoClient("mongodb://...");
const value = 'bébé';

async function run() {
  try {
    console.log(`Running Node ${process.versions.node}`);
    const coll = client.db("test").collection("foo");
    await coll.drop();

    let i = 0;
    while (Buffer.from(value).length === 6 && i < 20000) { i++ }

    await coll.insertOne({ _id: 1, message: value });
    const doc = await coll.findOne({ _id: 1 });
    console.log(`Found doc ${JSON.stringify(doc)}`);
  } finally {
    await client.close();
  }
}
run().catch(console.dir);

When run using a previous version of Node, the Buffer length is consistently evaluated for 20K iterations, a document is inserted into a MongoDB collection then successfully retrieved.

Running Node 22.6.0
Found doc {"_id":1,"message":"bébé"}

When the same reproduction is run using Node v22.7.0 however, invalid UTF-8 string data can be produced, which would then be inserted into the MongoDB collection, resulting in subsequent retrieval attempts failing.

Running Node 22.7.0
BSONError: Invalid UTF-8 string in BSON document
    at parseUtf8 (/Users/alex/temp/test-node/node_modules/bson/lib/bson.cjs:148:19)
    at Object.toUTF8 (/Users/alex/temp/test-node/node_modules/bson/lib/bson.cjs:273:21)
    ... 6 lines matching cause stack trace ...
    at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
    at async Collection.findOne (/Users/alex/temp/test-node/node_modules/mongodb/lib/collection.js:274:21) {
  [cause]: TypeError: The encoded data was not valid for encoding utf-8
      at TextDecoder.decode (node:internal/encoding:443:16)
      at parseUtf8 (/Users/alex/temp/test-node/node_modules/bson/lib/bson.cjs:145:37)
      at Object.toUTF8 (/Users/alex/temp/test-node/node_modules/bson/lib/bson.cjs:273:21)
      at deserializeObject (/Users/alex/temp/test-node/node_modules/bson/lib/bson.cjs:2952:31)
      at internalDeserialize (/Users/alex/temp/test-node/node_modules/bson/lib/bson.cjs:2863:12)
      at Object.deserialize (/Users/alex/temp/test-node/node_modules/bson/lib/bson.cjs:4335:12)
      at OnDemandDocument.toObject (/Users/alex/temp/test-node/node_modules/mongodb/lib/cmap/wire_protocol/on_demand/document.js:208:28)
      at CursorResponse.shift (/Users/alex/temp/test-node/node_modules/mongodb/lib/cmap/wire_protocol/responses.js:207:35)
      at FindCursor.next (/Users/alex/temp/test-node/node_modules/mongodb/lib/cursor/abstract_cursor.js:222:41)
      at process.processTicksAndRejections (node:internal/process/task_queues:105:5) {
    code: 'ERR_ENCODING_INVALID_ENCODED_DATA'
  }
}

Though MongoDB’s Node.js driver supports UTF-8 validation, that feature applies to decoding BSON strings that are being received from the MongoDB server. As the bug in Node v22.7.0 occurs when encoding strings as UTF-8, the invalid data can still be serialized to BSON and written to the database.