I would like to propose adding migration locks for MongoDB to ensure that migrations do not run concurrently across multiple instances of an application.

Since MongoDB is a NoSQL database, the locking approach differs from traditional RDBMS implementations. Below is a proposed approach.

1. Use a dedicated lock collection

   • Create a collection (e.g., migration_locks) that stores a lock document.

     • Example structure:

       {
         "_id": "migration_lock",
         "lockedAt": <timestamp>,
         "lockedBy": <instance-id>
       }

   • Before running migrations:

     • Attempt to insert or acquire the lock document.
     • If the lock already exists, the migration process should exit or wait.
     • If the lock is acquired successfully, proceed with the migration.

   • After migrations complete:

     • Delete the lock document to release the lock.

2. Use Atomic Operations

   • MongoDB operations such as:
     • findOneAndUpdate
     • insertOne with a unique _id
       can ensure atomic lock acquisition, preventing race conditions.

3. Optional: TTL Index for Safety

   • To avoid stale locks if a migration process crashes, we can add a TTL index on lockedAt.
     • Example:
       • Automatically expire lock documents after a configurable duration.

4. Transactions (If Available)

   • If MongoDB transactions are available (MongoDB ≥ 4.0 with replica sets):
   • Lock acquisition and migration metadata updates could run within a session transaction.
   • This would improve the reliability and atomicity of migration operations.

cc: @coolwednesday