Pessimistic locking RavenDB
RavenDB always had optimistic concurrency, I consider this to be an important feature for building correct distributed and concurrent systems. However, RavenDB doesn’t implement pessimistic locking. At least, not explicitly. It turns out that we have all the components in place to support it. If you want to read more about what pessimistic locking actually is, this Stack Overflow answer has good coverage of the topic.
There are two types of pessimistic locking. Offline and online locking. In the online mode, the database server will take an actual lock when modifying a record. That model works for a conversation pattern with the database. Where you open a transaction and hold it open while you mutate the data. In today’s world, where most processing is handled using request / response (REST, RPC, etc), that kind of interaction is rare. Instead, you’ll typically want to use offline pessimistic lock. That is, a lock that can live longer than a single transaction. With RavenDB, we build this feature on top of the usual optimistic concurrency as well as the document expiration feature.
Let’s take the classic example of pessimistic locking. Reserving seats for a show. Once you have selected a seat, you have 15 minutes to complete the order, otherwise the seats will automatically be released. Here is the code to do this:
The key here is that we rely on the @expires feature to remove the seatLock document automatically. We use a well known document id to coordinate concurrent requests that try to get the same seat. The rest is just the usual RavenDB’s optimistic concurrency behavior.
You have 15 minutes before the expiration and then it goes poof. From the point of view of implementing this feature, you’ll spend most of your time writing the edge cases, because from the point of view of RavenDB, there is really not much here at all.
Comments
What happens when two persons request the same seat at the same time and one request goes to node A and the other one goes to node B? I guess none get a conflict, the conflict doen't occur until replication takes place. Am I right?
Jesus,
You are correct. If you are worried about this scenario, you can use the compare exchange feature and cluster wide transactions to ensure consistency between the nodes
Comment preview