Handling concurrency

When multiple users interact with a financial system simultaneously, ensuring accurate and consistent transaction processing becomes crucial. Blnk handles this challenge through a concurrency control system that combines two key mechanisms: transaction queuing and optimistic locking.

In this guide, you’ll learn how these mechanisms work together to streamline transaction processing, prevent data conflicts and maintain system integrity.

​

Architectural Insights

Blnk’s architecture incorporates queuing and optimistic locking to manage concurrency as illustrated below:

This diagram illustrates Blnk’s layered architecture where transactions enter through the API layer and are then routed to the appropriate service based on the details of the transaction.

Queuing transactions act as a buffer to manage the flow of transactions to ensure sequential processing for similar operations. The database layer, equipped with optimistic locking, is where the final check for version consistency occurs, ensuring that only transactions based on the current state of the data are applied.

​

Queuing transactions

The queuing system manages transaction flow through partition-based processing. When transactions enter the system, they are routed to specific queue partitions based on their characteristics. Similar transactions affecting the same balance are grouped in a single partition, ensuring sequential processing order.

Key aspects of the queuing system:

• Partition Assignment: Incoming transactions are assigned to available partitions. If one partition is busy, new transactions are routed to other free partitions.
• Worker Processing: Worker processes pull transactions from the queue partitions. The number of simultaneously processed transactions depends on available workers.
• Sequential Consistency: Transactions within the same partition are processed sequentially, maintaining order for related operations.

​

Idempotency in transaction processing

Blnk’s transaction system ensures idempotency through reference value tracking. When multiple instances of the same transaction occur due to network or system errors, the system:

1. Identifies duplicate transactions using their reference value
2. Processes only one instance of the transaction
3. Discards additional duplicate requests

This mechanism prevents double processing and ensures consistent ledger entries, eliminating the risk of duplicate debits or credits.

​

Optimistic locking for conflict Resolution

The optimistic locking mechanism complements the queuing system by providing conflict resolution at the data layer. Instead of locking resources preemptively, the system:

1. Allows transactions to proceed concurrently.
2. Performs conflict detection at commit time.
3. Checks if data has been modified since the transaction started.
4. Rolls back and retries transactions if conflicts are detected.

This approach maximizes throughput while maintaining data consistency. When conflicts occur, indicating that another transaction has modified the data since it was last read, the system initiates a rollback and retry process.

​

How optimistic locking works in Blnk

1. Version tracking: Each record in the database is associated with a version number. When a transaction reads a record, it notes the version number.
2. Update attempt: When the transaction attempts to update the record, it specifies the version number it originally read.
3. Conflict detection: If the current version in the database does not match the version number noted by the transaction in #2, a conflict is detected (indicating that another transaction has updated the record). If no conflict is detected, the transaction moves to queue for processing.
4. Resolution: Upon conflict detection, the transaction is rolled back. Blnk may automatically retry the transaction, depending on the scenario and the conflict’s nature.

​

Need help?

We are very happy to help you make the most of Blnk, regardless of whether it is your first time or you are switching from another tool.

To ask questions or discuss issues, please contact us or join our Discord community.