Skip to main content
Version: v1.10

Concurrency model

Transactions commited to the ledger are fully atomic and serialized, supported by two separate and ordered concurrency-control mechanisms preventing effectively race-conditions to happen:

  1. Pre-commit in-memory or Redis-based locking
  2. Optimistic locking

Transaction commit flow

flow

Recommendation for multi-instances deployments

Due to the sequential nature of the ledger, multiple instances deployments should be carefully configured.

We recommended to use the Redis-based shared pre-commit lock using the redis flags. While the optimistic lock will ultimately be there to prevent race-conditions on commit, using the shared lock will reduce such commit attempts in the first place, yielding better performance for write heavy workloads.

Also, you can find a complete docker-compose example using 3 instances of the ledger, and a simple reverse proxy to handle traffic there.

Should the optimistic locking prevent a conflict on commit, it will surface it to the API consumer with a (Response 409) - it is the responsibility of the client to retry the transaction in this case.