distributed-locks-mongodb

MongoDB Algorithm for lock guarantees

We highly depend on the fact the mongo by default uses the _id as the primary unique key for the collection, and the fact that mongo-db guarantees atomicity on a document level

Perquisites

Make sure your mongo connection has majority for write concern and read from primary for read preference to avoid any inconsistences

Acquiring Lock

We first try to get the document in mongo that has the same value for the key (checking if someone else obtained a lock on the same critical section). Plus we project the current date in mongo

db.collection('collection')
    .findOne({ _id: 'key' }, {
        projection: {
          _id: 1,
          value: 1,
          ttl: 1,
          obtained_at: 1,
          current_date: '$$NOW',
        },
    });

If returned value is null (no one has a lock on the critical section), we try to upsert a document in mongo, we use upsert since in between this step and the previous one someone else might've obtained a lock on the same critical section and it could've even expired. the fields are:

• _id: Represents the critical section
• value: Represents the lock that locked the critical section
• ttl: Time to live in seconds
• obtained_at: The time this lock was successfully acquired

this.db.collection('collection').updateOne({
          _id: 'key',
          value: 'lock-value',
        }, {
          $set: {
            _id: 'key',
            value: 'lock-value',
            ttl: 10,
          },
          $currentDate: {
            obtained_at: { $type: 'date' },
          },
        }, { upsert: true })

If this was successful then we obtained the lock on the critical section

If the returned value from step one was not null (someone else obtained a lock for the same critical section), then we can only insert a new document if that returned lock has expired.

We do a simple check from the data returned from step one

if(data.current_date > data.obtained_at + data.ttl) {
    // Lock expired try to obtain a lock for this critical section
} 
else {
    // Lock still active can't lock the same critical section
}

If the if condition was true we do the following upsert statement

await this.db?.collection<Omit<MongoDocument, 'created_at'>>(this.collectionName).updateOne({
          _id: 'key',
          value: data.value,
          obtained_at: data.obtained_at,
        }, {
          $set: {
            _id: 'key',
            value: 'lock-value',
            ttl: 10,
          },
          $currentDate: {
            obtained_at: { $type: 'date' },
          },
        }, { upsert: true });

Notice that we don't only use the _id field to upsert the document we also use the the current lock's value and it's obtained_at as again some one else might have obtained the lock on the same critical section (same _id value but different value and obtained_at). If operation was successful then we obtained the lock

Releasing Lock

A simple delete query using the _id and the lock value

this.db.collection('collection')
      .deleteOne({
        _id: 'key',
        value: 'lock-value',
      });