Go-Flow

A Golang based high performance, scalable and distributed workflow framework

It allows to programmatically author distributed workflow as Directed Acyclic Graph (DAG) of tasks. GoFlow executes your tasks on an array of Flow workers by uniformly distributing the loads

Install It

Install GoFlow

go mod init myflow
go get github.com/s8sg/goflow

Write First Flow

Library to Build Flow github.com/s8sg/goflow/flow

Make a flow.go file

package main

import (
	"fmt"
	"github.com/s8sg/goflow"
	flow "github.com/s8sg/goflow/flow"
)

// Workload function
func doSomething(data []byte, option map[string][]string) ([]byte, error) {
	return []byte(fmt.Sprintf("you said \"%s\"", string(data))), nil
}

// Define provide definition of the workflow
func DefineWorkflow(f *flow.Workflow, context *flow.Context) error {
	f.SyncNode().Apply("test", doSomething)
	return nil
}

func main() {
	fs := &goflow.FlowService{
		Port:                8080,
		RedisURL:            "localhost:6379",
		OpenTraceUrl:        "localhost:5775",
		WorkerConcurrency:   5,
	}
	fs.Register("myflow", DefineWorkflow)
	fs.Start()
}

Start() runs a HTTP Server that listen on the provided port and as a flow worker that handles the workload

Run It

Start redis

docker run --name redis -p 6379:6379 -d redis

Run the Flow

go build -o goflow
./goflow

Invoke It

curl -d hallo localhost:8080/myflow

Scale It

GoFlow scale horizontally, you can distribute the load by just adding more instances.

Worker Mode

Alternatively you can start your GoFlow in worker mode. As a worker GoFlow only handles the workload, and if required you can only scale the workers

fs := &goflow.FlowService{
    RedisURL:            "localhost:6379",
    OpenTraceUrl:        "localhost:5775",
    WorkerConcurrency:   5,
}
fs.Register("myflow", DefineWorkflow)
fs.StartWorker()

Register Multiple Flow

Register() allows user to bind multiple flows onto single flow service. This way a server and or a worker can be used for more than one flows

fs.Register("createUser", DefineCreateUserFlow)
fs.Register("deleteUser", DefineDeleteUserFlow)

Execute It

Using Client

Using the client you can requests the flow directly. The requests are always async and gets queued for the worker to pick up

fs := &goflow.FlowService{
    RedisURL: "localhost:6379",
}
fs.Execute("myflow", &goflow.Request{
    Body: []byte("hallo")
})

Using Redis

For testing, it is helpful to use the redis-cli program to insert jobs onto the Redis queue:

redis-cli -r 100 RPUSH goflow:queue:myflow '{"class":"GoFlow","args":["hallo"]}'

this will insert 100 jobs for the GoFlow worker onto the myflow queue

Currently redis queue based job only take one argument as string

Creating More Complex DAG

The initial example is a single vertex DAG. Single vertex DAG (referred as SyncNode) are great for synchronous task

Using GoFlow's DAG construct one can achieve more complex compositions with multiple vertexes and connect them using edges. A multi-vertex flow is always asynchronous in nature where each nodes gets distributed across the workers

Below is an example of a simple multi vertex flow to validate a KYC image of a user and mark the user according to the result. This is a asynchronous process consist of batch jobs

func DefineWorkflow(f *flow.Workflow, context *flow.Context) error {
    dag := f.Dag()
    dag.Node("get-kyc-image").Apply('load-profile', loadProfile)
       .Apply("get-image-url", getPresignedURLForImage)
    dag.Node("face-detect").Apply("face-detect", detectFace)
    dag.Node("mark-profile").Apply("mark-profile", markProfileBasedOnStatus)
    dag.Edge("get-kyc-image", "face-detect")
    dag.Edge("face-detect", "mark-profile")
    return nil
}

Branching

Branching are great for parallelizing independent workloads in separate branches

Branching can be achieved with simple vertex and edges. GoFlow provides a special operator Aggregator to aggregate result of multiple branch on a converging node

We are extending our earlier example to include a new requirement to match the face with existing data and we are performing the operation in parallel to reduce time

func DefineWorkflow(f *flow.Workflow, context *flow.Context) error {
    dag := f.Dag()
    dag.Node("get-kyc-image").Apply("load-profile", loadProfile)
       .Apply("get-image-url", getPresignedURLForImage)
    dag.Node("face-detect").Apply("face-detect", detectFace)
    dag.Node("face-match").Apply("face-match", matchFace)
    // Here mark-profile depends on the result from face-detect and face-match, 
    // we are using a aggregator to create unified results
    dag.Node("mark-profile", flow.Aggregator(func(responses map[string][]byte) ([]byte, error) {
       status := validateResults(responses["face-detect"],  responses["face-match"])
       return []byte(status), nil
    })).Apply("mark-profile", markProfileBasedOnStatus)
    dag.Edge("get-kyc-image", "face-detect")
    dag.Edge("get-kyc-image", "face-match")
    dag.Edge("face-detect", "mark-profile")
    dag.Edge("face-match", "mark-profile")
    return nil
}

Subdag

Subdag allows to reuse existing DAG by embedding it into DAG with wider functionality

SubDag is available as a GoFlow DAG construct which takes a separate DAG as an input and composite it within a vertex, where the vertex completion depends on the embedded DAG's completion

func (currentDag *Dag) SubDag(vertex string, dag *Dag)

Say we have a separate flow that needs the same set of steps to validate a user. With our earlier example we can separate out the validation process into subdag and put it in a library that can be shared across different flows

func KycImageValidationDag() *flow.Dag {
    dag := flow.NewDag()
    dag.Node("verify-url").Appply("verify-image-url", s3DocExists)
    dag.Node("face-detect").Apply("face-detect", detectFace)
    dag.Node("face-match").Apply("face-match", matchFace)
    dag.Node("generate-result", flow.Aggregator(func(responses map[string][]byte) ([]byte, error) {
           status := validateResults(responses["face-detect"],  responses["face-match"])
           status = "failure"
           if status {
              status = "success"
           }
           return []byte(status), nil
        })).Apply("generate-result", func(data []byte, option map[string][]string) ([]byte, error) {
           return data, nil
        })
    dag.Edge("verify-url", "face-detect")
    dag.Edge("verify-url", "face-match")
    dag.Edge("face-detect", "generate-result")
    dag.Edge("face-match", "generate-result")
    return dag
}

Our existing flow embeds the library DAG

func DefineWorkflow(f *flow.Workflow, context *flow.Context) error {
    dag := f.Dag()
    dag.Node("get-image").Apply("load-profile", loadProfile)
           .Apply("get-image-url", getPresignedURLForImage)
    dag.SubDag("verify-image", common.KycImageValidationDag)
    dag.Node("mark-profile").Apply("mark-profile", markProfileBasedOnStatus)
    dag.Edge("get-image", "verify-image")
    dag.Edge("verify-image", "mark-profile")
    return nil
}

Conditional Branching

Conditional branching is a great way to choose different execution path dynamically

GoFlow provides a DAG component called ConditionalBranch. ConditionalBranch creates a vertex that composites different conditional branches as an individual subdags, each identified with a unique key resemble the condition

func (currentDag *Dag) ConditionalBranch(vertex string, conditions []string, condition sdk.Condition,
    options ...BranchOption) (conditiondags map[string]*Dag)

Condition is a special handler that allows user to dynamically choose one or more execution path based on the result from earlier node and return a set of condition Keys

User gets the condition branches as a response where each branch specific dags are mapped against the specific condition. User can farther define each branch using the DAG constructs

Below is the updated example with a conditional Branch where we are trying to call face-match only when face-detect passes

func KycImageValidationDag() *flow.Dag {
    dag := flow.NewDag()
    dag.Node("verify-url").Apply("verify-image-url", s3DocExists)
    dag.Node("face-detect").Apply("face-detect", detectFace)
    // here face match happen only when face-detect is success
    branches = dag.ConditionalBranch("handle-face-detect-response", []string{"pass"}, func(response []byte) []string {
        response := ParseFaceDetectResponse(response)
        if response[0] == "pass" { return []string{"pass"}  }
        return []string{}
    })
    // On the pass branch we are performing the `face-match`
    // As defined condition `pass` is not matched execution of next node `generate-result` is continued
    branches["pass"].Node("face-match").Apply("face-match", matchFace)
  
    dag.Node("generate-result", generateResult)
    dag.Edge("verify-url", "face-detect")
    dag.Edge("face-detect", "handle-face-detect-response")
    dag.Edge("handle-face-detect-response", "generate-result")
    return dag
}

You can also have multiple conditional branch in a workflow and different nodes corresponding to each branch

Below is the updated example with two conditional Branches where we are trying to call face-match or create-user based on response from previous node

func KycImageValidationDag() *flow.Dag {
    dag := flow.NewDag()
    dag.Node("verify-url").Apply("verify-image-url", s3DocExists)
    dag.Node("face-detect").Apply("face-detect", detectFace)
    // here face match happen only when face-detect is success
    // otherwise create-user is called
    branches = dag.ConditionalBranch("handle-face-detect-response", []string{"pass", "fail"}, func(response []byte) []string {
        response := ParseFaceDetectResponse(response)
        if response[0] == "pass" { return []string{"pass"}  }
        return []string{"fail"}
    })
    // On the pass branch we are performing the `face-match`
    branches["pass"].Node("face-match").Apply("face-match", matchFace)
    // on the fail branch we are performing `create-user`
    branches["fail"].Node("create-user").Apply("create-user", createUser)
  
    dag.Node("generate-result", generateResult)
    dag.Edge("verify-url", "face-detect")
    dag.Edge("face-detect", "handle-face-detect-response")
    dag.Edge("handle-face-detect-response", "generate-result")
    return dag
}

Foreach Branching

Foreach branching allows user to iteratively perform a certain set of task for a range of values

GoFlow provides a DAG component called ForEachBranch. ForEachBranch creates a vertex composites of a subdag that defines the flow within the iteration

func (currentDag *Dag) ForEachBranch(vertex string, foreach sdk.ForEach, options ...BranchOption) (dag *Dag)

ForEach is a special handler that allows user to dynamically return a set of key and values. For each of the items in the returned set, the user defined dag will get executed

User gets the foreach branch as a response and can define the flow using the DAG constructs

We are updating our flow to execute over a set of user that has been listed for possible fraud

func DefineWorkflow(f *flow.Workflow, context *flow.Context) error {
    dag := f.Dag()
    dag.Node("get-users").Apply("get-listed-users", getListedUsers)
    verifyDag = dag.ForEachBranch("for-each-user-verify", func(data []byte) map[string][]byte {
       users := ParseUsersList(data)
       forEachSet := make(map[string][]byte)
       for _, user := range users {
           forEachSet[user.id] = []byte(user.GetKycImageUrl())
       }
       return forEachSet
    })
    verifyDag.SubDag("verify-image", KycImageValidationDag)
    verifyDag.Node("mark-profile").Apply("mark-profile", markProfileBasedOnStatus)
    verifyDag.Edge("verify-image", "mark-profile")

    dag.Edge("get-users", "for-each-user-verify")
    return nil
}