Borrows heavily from Temporal (and since it's a fork also Cadence) as well as DTFx.
See also:
- https://cschleiden.dev/blog/2022-02-13-go-workflows-part1/
- https://cschleiden.dev/blog/2022-05-02-go-workflows-part2/
On Go support: the current version of the library requires Go 1.18 or later. There is a version that doesn't require generics and relies more on interface{} instead, but I think the improved type safety is worth not supporting a version of Go before 1.18 for now.
See http://cschleiden.github.io/go-workflows for the current version of the documentation.
Workflows are written in Go code. The only exception is they must not use any of Go's non-deterministic features (select, iteration over a map, etc.). Inputs and outputs for workflows and activities have to be serializable:
func Workflow1(ctx workflow.Context, input string) error {
r1, err := workflow.ExecuteActivity[int](ctx, workflow.DefaultActivityOptions, Activity1, 35, 12).Get(ctx)
if err != nil {
panic("error getting activity 1 result")
}
log.Println("A1 result:", r1)
r2, err := workflow.ExecuteActivity[int](ctx, workflow.DefaultActivityOptions, Activity2).Get(ctx)
if err != nil {
panic("error getting activity 1 result")
}
log.Println("A2 result:", r2)
return nil
}Activities can have side-effects and don't have to be deterministic. They will be executed only once and the result is persisted:
func Activity1(ctx context.Context, a, b int) (int, error) {
return a + b, nil
}
func Activity2(ctx context.Context) (int, error) {
return 12, nil
}The worker is responsible for executing Workflows and Activities, both need to be registered with it.
func runWorker(ctx context.Context, mb backend.Backend) {
w := worker.New(mb, nil)
w.RegisterWorkflow(Workflow1)
w.RegisterActivity(Activity1)
w.RegisterActivity(Activity2)
if err := w.Start(ctx); err != nil {
panic("could not start worker")
}
}The backend is responsible for persisting the workflow events. Currently there is an in-memory backend implementation for testing, one using SQLite, one using MySql, and one using Redis.
b := sqlite.NewSqliteBackend("simple.sqlite")We can start workflows from the same process the worker runs in -- or they can be separate. Here we use the SQLite backend, spawn a single worker (which then executes both Workflows and Activities), and then start a single instance of our workflow
func main() {
ctx := context.Background()
b := sqlite.NewSqliteBackend("simple.sqlite")
go runWorker(ctx, b)
c := client.New(b)
wf, err := c.CreateWorkflowInstance(ctx, client.WorkflowInstanceOptions{
InstanceID: uuid.NewString(),
}, Workflow1, "input-for-workflow")
if err != nil {
panic("could not start workflow")
}
c2 := make(chan os.Signal, 1)
signal.Notify(c2, os.Interrupt)
<-c2
}