A simple go library for micro-batching.
Microbatching is a technique commonly used in stream processing pipelines of grouping events into small batches to reduce the overhead of processing events individually, and for fault tolerance. It can also be used to batch together tasks like network calls or writes to disk. There are good examples of microbatching in the GraphQL ecosystem with tools like dataloader.
To start microbatching, call Start() with the necessary config object. This includes a BatchProcessor and ResultHandler which need to be defined.
type batchProcessor struct{}
type resultHandler struct{}
func (batchProcessor) Run(batch []Job) JobResult {
// domain specific logic for handling a batch
}
func (resultHandler) Run(result JobResult) {
// domain specific logic for handling any results
}
config := microbatcher.Config[Job, JobResult]{
BatchProcessor: batchProcessor{}, // for processing each batch
ResultHandler: resultHandler{}, // for processing each batch result
Frequency: 1000, // number of milliseconds between each batch
MaxSize: 30, // if this limit is reached, we'll batch early
}
mb := microbatcher.Start(config)
Then submit jobs with mb.SubmitJob(job). Jobs are domain specific, called Job in this example but can be of any type.
Finally, call mb.WaitForResults() to wait for results to come back. ResultHandler will be called with the results of each batch.
BatchProcessor will be run in a new goroutine for each batch. The ResultHandler is called by WaitForResults(), so should be a safe place to pass data back to your application.
At the end of each cycle (according to Frequency) any new jobs will be sent to the BatchProcessor. Additionally, if the MaxSize limit is reached, jobs will be sent early. Note that this doesn't affect the time until the next cycle. Each batch is processed in it's own goroutine so that long running batches don't block each other. Here is a simple visualisation:
To run the example, cd into ./exampleclient and run go run .. See here for more information about the example application.