This is an AMP version of the article, its original content can be found here.
Synchronization Between Nodes
When two or more software modules are accessing the same resource, they have to be synchronized. This means that only one module at a time should be working with the resource. Without such synchronization, there will be collisions and conflicts. This is especially true when we're talking about "resources" that do not support atomic transactions.
To solve this issue and prevent conflicts, we have to introduce one more element into the picture. All software modules, before accessing the resource, should capture a lock from a centralized server. Once the manipulations with the resource are complete, the module should release the lock. While the lock is being captured by one module, no other modules will be able to capture it. The approach is very simple and well-known. However, I didn't find any cloud services that would provide such a locking and unlocking service over a RESTful API. So I decided to create one—stateful.co.
Here is a practical example. I have a Java web app that is hosted
at Heroku. There are three servers (a.k.a. "dynos") running the same
application. Why three? Because the web traffic is rather active, and one
server is not powerful enough. So I have to have three of them. They all
run exactly the same applications.
Each web app works with a table in Amazon DynamoDB. It updates the table, puts new items into it, deletes some items sometimes, and selects them. So far, so good, but conflicts are inevitable. Here is an example of a typical interaction scenario between the web app and DynamoDB (I'm using jcabi-dynamo):
The logic is obvious here. First, I retrieve an item from the table
then read its
salary, and then modify it according to my recalculation
algorithm. The problem is that another module may start to do the same while
I'm recalculating. It will read the same initial value from the table
and will start exactly the same recalculation. Then it will save a new
value, and I will save one too. We will end up having Jeff's salary modified
only once, while users will expect a double modification since two of them
initiated two transactions with two different web apps.
The right approach here is to "lock" the DynamoDB table first, even before reading the salary. Then do the modifications and eventually unlock it. Here is how stateful.co helps me. All I need to do is create a new named lock in the stateful.co web panel, get my authentication keys, and modify my Java code:
As you see, I wrap that critical transaction into
Callable, which will
be executed in isolation. This approach, obviously, doesn't
guarantee atomicity of transaction—if part of the transaction fails,
there won't be any automatic rollbacks and the DynamoDB table will be left
in a "broken" state.
Locks from stateful.co guarantee isolation in resource usage, and you can use any type of resources, including NoSQL tables, files, S3 objects, embedded software, etc.
I should not forget to add this dependency to my
Of course, you can do the same; the service is absolutely free of charge. And you can use any other languages, not just Java. BTW, if interested, contribute with your own SDK in your preferred language; I'll add it to the GitHub collection.