mongochangestream

Mongo Change Stream

Sync a MongoDB collection to any database. Requires Redis for state management. An initial scan is performed while change stream events are handled. In order to prevent a potential race condition see the strategies section below.

If the inital scan doesn't complete for any reason (e.g., server restart) the scan will resume where it left off. This is deterministic since the collection scan is sorted by _id by default. Change streams will likewise resume from the last resume token upon server restarts. See the official MongoDB docs for more information on change stream resumption:

https://www.mongodb.com/docs/manual/changeStreams/#std-label-change-stream-resume

WARNING: If the Node process is stopped prior to receiving the initial change event for the collection there is a risk that changes to documents that took place while the server was restarting would be missed.

This library uses debug. To enable you can do something like:

DEBUG=mongochangestream node myfile.js

import { default as Redis } from 'ioredis'
import { initSync } from 'mongochangestream'
import { ChangeStreamDocument, MongoClient } from 'mongodb'

const redis = new Redis()

const mongoUrl = 'mongodb+srv://...'
const client = await MongoClient.connect(mongoUrl)
const db = client.db('someDb')
const coll = db.collection('someColl')

const processCSRecords = async (docs: ChangeStreamDocument[]) => {
  console.dir(docs, { depth: 10 })
}
const processRecords = async (docs: ChangeStreamInsertDocument[]) => {
  console.dir(docs, { depth: 10 })
}

// Sync collection
const sync = initSync(redis, coll)
const initialScan = await sync.runInitialScan(processRecords)
initialScan.start()
// Process change stream
const changeStream = await sync.processChangeStream(processCSRecords)
changeStream.start()
setTimeout(changeStream.stop, 30000)
// Detect schema changes and ignore metadata fields (i.e., title and description)
const schemaChange = await sync.detectSchemaChange(db, {
  shouldRemoveMetadata: true,
})
schemaChange.start()
sync.emitter.on('schemaChange', () => {
  initialScan.stop()
  changeStream.stop()
})

Below are the available methods.

The processChangeStream method will never complete, but runInitialScan will complete once it has scanned all documents in the collection. runInitialScan batches records for efficiency.

The reset method will delete all relevant keys for a given collection in Redis.

import { ChangeStreamDocument, Collection, Document } from 'mongodb'

export type ProcessChangeStreamRecords = (
  docs: ChangeStreamDocument[]
) => MaybePromise<void>

export type ProcessInitialScanRecords = (
  docs: ChangeStreamInsertDocument[]
) => MaybePromise<void>

const runInitialScan = async (
  processRecords: ProcessInitialScanRecords,
  options: QueueOptions & ScanOptions = {}
)

const processChangeStream = async (
  processRecords: ProcessChangeStreamRecords,
  options: QueueOptions & ChangeStreamOptions = {}
)

const detectSchemaChange = async (db: Db, options: ChangeOptions = {})

Maintaining Health

Look for the cursorError event and restart the process or resync as needed. See also the missingOplogEntry utility function that helps determine if an oplog entry is no longer present and resumption of a change stream from a previous point is not possible.

It is recommended that you run a periodic check (e.g., every minute) to determine the health of the destination database that data is being synced to. If an issue is detected use the pausable API to pause and resume all syncing operations.

Companion Libraries

This library is meant to be built on. To that end, the following libraries are currently implemented and maintained.

Sync MongoDB to MongoDB mongo2mongo

Sync MongoDB to Elasticsearch mongo2elastic

Sync MongoDB to CrateDB mongo2crate

Resilience

Both the initial scan and change stream processing are designed to handle and resume from failures. Here are some scenarios:

The syncing server goes down

In this scenario, processing will continue with the last recorded state when resumed.

The syncing server is being shutdown with a sigterm

In this scenario, calling stop for the initial scan and change stream will cleanly end processing.

The MongoDB primary goes down and a new primary is elected

In this scenario, you will need to subscribe to the cursorError event and restart the process or handle otherwise.

Limit throughput to prevent overloading the destination database

The initialScan and processChangeStream functions support throttling via the batchQueue options - maxItemsPerSec and maxBytesPerSec. See the prom-utils library for more details.

Change Stream Strategies

The idea behind these strategies is to prevent overwriting a document with an out-of-date version of the document. In order to prevent that scenario inserts must only succeed if the document doesn't already exist. Likewise, updates must be capable of inserting the full document if it doesn't already exist (i.e., perform a replace or an upsert).

The initial scan returns a simulated change event document with operationType set to insert. An actual update change event will include the field-level changes in addition to the full document after the change.

NOTE: Exceptions are not caught by this library. You must catch them in your processRecords callback and handle them accordingly. For example, an insert that fails due to a primary key already existing in the destination datastore.

Elasticsearch

Insert

POST /index/_create/id
...

Update

POST /index/_doc/id
...

Remove

DELETE /index/_doc/id

SQL (MySQL, CrateDB)

Insert

INSERT INTO table ...

Update

MySQL

INSERT INTO table ... ON DUPLICATE KEY UPDATE changedField = someValue

CrateDB

INSERT INTO table ... ON CONFLICT DO UPDATE SET changedField = someValue

Remove

DELETE FROM table WHERE id = someId

MongoDB

Insert

db.collection('someColl').insertOne(...)

Update

db.collection('someColl').replaceOne({_id: ObjectId(...)}, ..., {upsert: true})

Remove

db.collection('someColl').deleteOne({_id: ObjectId(...)})