Skip to content
/ batch Public

Simple Go package for handling incoming requests in batches.

License

Notifications You must be signed in to change notification settings

elgopher/batch

Repository files navigation

Build Go Reference Go Report Card codecov Project Status: Active – The project has reached a stable, usable state and is being actively developed.

What it can be used for?

To increase database-driven web application throughput without sacrificing data consistency and data durability or making source code and architecture complex.

The batch package simplifies writing Go applications that process incoming requests (HTTP, GRPC etc.) in a batch manner: instead of processing each request separately, they group incoming requests to a batch and run whole group at once. This method of processing can significantly speed up the application and reduce the consumption of disk, network or CPU.

The batch package can be used to write any type of servers that handle thousands of requests per second. Thanks to this small library, you can create relatively simple code without the need to use low-level data structures.

Why batch processing improves performance?

Normally a web application is using following pattern to modify data in the database:

  1. Load resource from database. Resource is some portion of data such as set of records from relational database, document from Document-oriented database or value from KV store (in Domain-Driven Design terms it is called an aggregate). Lock the entire resource optimistically by reading version number.
  2. Apply change to data in plain Go
  3. Save resource to database. Release the lock by running atomic update with version check.

But such architecture does not scale well if the number of requests for a single resource is very high (meaning hundreds or thousands of requests per second). The lock contention in such case is very high and database is significantly overloaded. Also, round-trips between application server and database add latency. Practically, the number of concurrent requests is severely limited.

One solution to this problem is to reduce the number of costly operations. Because a single resource is loaded and saved thousands of times per second we can instead:

  1. Load the resource once (let's say once per second)
  2. Execute all the requests from this period of time on an already loaded resource. Run them all sequentially to keep things simple and data consistent.
  3. Save the resource and send responses to all clients if data was stored successfully.

Such solution could improve the performance by a factor of 1000. And resource is still stored in a consistent state.

The batch package does exactly that. You configure the duration of window, provide functions to load and save resource and once the request comes in - you run a function:

// Set up the batch processor:
processor := batch.StartProcessor(
    batch.Options[*YourResource]{ // YourResource is your own Go struct
        MinDuration:  100 * time.Millisecond,
        LoadResource: func(ctx context.Context, resourceKey string) (*YourResource, error){
            // resourceKey uniquely identifies the resource
            ...
        },
        SaveResource: ...,
    },
)

// And use the processor inside http/grpc handler or technology-agnostic service.
// ctx is a standard context.Context and resourceKey can be taken from request parameter
err := processor.Run(ctx, resourceKey, func(r *YourResource) {
    // Here you put the code which will executed sequentially inside batch  
})

For real-life example see example web application.

Installation

# Add batch to your Go module:
go get github.com/elgopher/batch

Please note that at least Go 1.18 is required. The package is using generics, which was added in 1.18.

Scaling out

Single Go http server is able to handle up to tens of thousands of requests per second on a commodity hardware. This is a lot, but very often you also need:

  • high availability (if one server goes down you want other to handle the traffic)
  • you want to handle hundred-thousands or millions of requests per second

For both cases you need to deploy multiple servers and put a load balancer in front of them. Please note though, that you have to carefully configure the load balancing algorithm. Round-robin is not an option here, because sooner or later you will have problems with locking (multiple server instances will run batches on the same resource). Ideal solution is to route requests based on URL path or query string parameters. For example some http query string parameter could have a resource key. You can instruct load balancer to calculate hash on this parameter and always route requests with the same key to the same backend. If backend will be no longer available the load balancer should route request to a different server.