如何使用Java Spring同步在码头集群中运行的微服务实例

Question

I am developing a Java 8 Spring 5 microservice that runs on a Jetty embedded server. I intend to spawn several instances of this microservice, so there will be several Jetty servers running concurrently on different ports, one for each spawned instance.

All instances share the same database which is a MongoDB database.

From within my microservice I invoke a third party REST Web Service. I am saving the data returned by this REST WS in my database, so all instances of the cluster can take advantage of the request made by one of the instances.

However, this data expires after some time, so one instance will have to invoke the external Web Service after some time.

I do not want two or more instances of my cluster to invoke this external web service at the same time, if one instance is invoking this web service, and another one requires the data, it should wait for the first instance to retrieve the data and then use it.

This is when i find the problem, because i do not know how to lock one instance of the jetty server when another one is invoking the web service.

I have been looking into MongoDBs 4.0 Transaction feature, to try and lock a document while the service is being invoked, but without success (still :) ).

Answer 1

You need a way to serialize access to this external service, that is, to ensure a single request is performed at any particular moment in time.

This is that thing the databases are really good at. With a specially crafted unique index you get exactly that. You don't even need transactions for this, because you will have only one collection (table).

The solution is to use optimistic locking. I haven't used myself in Java (only in PHP) but I found this question that could help you. Long story short, you use @org.springframework.data.annotation.Version on a property.

You should define a new Entity that represents the remote piece of data, having a state property with the following possible values: old (the default), fetching and fetched . When a local service needs the data, it loads the Entity and check the status:

• if the status is fetched then it uses it;
• if the status is old , then it tries to change it to fetching ; if the database rejects it because of concurrent modification then it should wait until the status becomes fetched ; if the saving succeeds then it calls the remote service and persist the result in the Entity as a property;
• if the status is fetching it should wait until it becomes fetched .