Awesome

Fission Workflows

fission.io @fissionio

Fission Workflows is a workflow-based serverless function composition framework built on top of the Fission Function-as-a-Service (FaaS) platform.

:warning: Fission Workflows is currently in maintenance mode due to a time constraints of the core Fission team. If you are interested in contributing or helping maintain this project, contact the Fission team on the Fission Slack. :warning:

Highlights

• Fault-Tolerant: Fission Workflows engine keeps track of state, re-tries, handling of errors, etc. By internally utilizing event sourcing, it allows the engine to recover from failures and continue exactly were it left off.
• Scalable: Other than a backing data store, the workflow system is stateless and can easily be scaled. The independent nature of workflows allows for relatively straightforward sharding of workloads over multiple workflow engines.
• High Performance: In contrast to existing workflow engines targeting other domains, Fission Workflows is designed from the ground up for low overhead, low latency workflow executions.
• Extensible: All main aspects of the engine are extensible. For example, you can even define your own control flow constructs.
• Lightweight: With just the need for a single data store (NATS Streaming) and a FaaS platform (Fission), the engine consumes minimal resources.

Philosophy

The Fission Functions-as-a-Service framework provides simplicity and quick time-to-value for functions on any infrastructure using Kubernetes.

Functions tend to do one logically separate task, and they're usually short-lived. For many relatively simple applications this is good enough. But a more complex application that uses serverless functions must, in some way, compose functions together.

There are several ways to compose functions. A function could invoke another function using the Fission API or HTTP. But this requires the calling function to handle serialization, networking, etc.

You could also set up functions to be invoked using message queue topics. This requires less boilerplate within each function, but the structure of the application is not explicit; dependencies are buried inside mappings of message queue topics to functions.

In addition, both these approaches are operationally difficult, in terms of error analysis, performance debugging, upgrades, etc.

Workflows have been popular in other domains, such as data processing and scientific applications, and recently got introduced into the serverless model by AWS Step Functions and Azure Logic Apps.

Fission Workflows is an open-source alternative to these workflow systems. It allows users to compose Fission functions in powerful ways. Users can define their own control flow constructs, which in most other workflow engines are part of the internal API.

Concepts

Workflows can generally be represented in as a Directed Acyclic Graph (DAG). Consider the following example, a common pattern, the diamond-shaped workflow:

In this graph there is a single starting task A, a scatter task B triggering parallel execution of two branches with tasks C and D, followed by a synchronizing task E collecting the outputs of the parallel tasks.

Finally the graph concludes once final task F completes.

Although Fission Workflows has more functionality such as conditional branches and advanced control flow options, it fundamentally executes a dependency graph.

apiVersion: 1
output: WhaleWithFortune
tasks:
  GenerateFortune:
    run: fortune
    inputs: "{$.Invocation.Inputs.default}"

  WhaleWithFortune:
    run: whalesay
    inputs: "{$.Tasks.GenerateFortune.Output}"
    requires:
    - GenerateFortune

Task (also called a function here) is an atomic task, the 'building block' of a workflows.

Currently there are two options for executing tasks. First, Fission is used as the main function execution runtime, using fission functions as tasks. Second, for very small tasks, such as flow control constructs, internal functions execute within the workflow engine to minimize the network overhead.

A workflow execution is called a (Workflow) Invocation. The Fission Workflows engine assigns an UID invocation and stores it, persistently, in the data-store. This allows users to reference the invocation during and after the execution, such as to view the progress so far.

Finally, selectors and data transformations are inline functions which you can use to manipulate data without having to create a task for it. These inline functions consist of commonly used transformations, such as getting the length of an array or string. Additionally, selectors allow users to only pass through certain properties of data. In the example workflow, the JSONPath-like selector selects the default input of the invocation:

See the docs for a more extensive, in-depth overview of the system.

Usage

#
# Add binary environment and create two test function on your Fission setup 
#
fission env create --name binary --image fission/binary-env
fission function create --name whalesay --env binary --deploy examples/whales/whalesay.sh
fission function create --name fortune --env binary --deploy examples/whales/fortune.sh

#
# Create a workflow that uses those two functions; a workflow
# is just a function that uses the special "workflow" environment.
#
fission function create --name fortunewhale --env workflow --src examples/whales/fortunewhale.wf.yaml

#
# Map a HTTP GET to your new workflow function
#
fission route create --method GET --url /fortunewhale --function fortunewhale

#
# Invoke the workflow with an HTTP request
#
curl $FISSION_ROUTER/fortunewhale

See examples for other workflow examples.

Installation

See the installation guide.

Compiling

See the compilation guide.

Status and Roadmap

This is an early release for community participation and user feedback. It is under active development; none of the interfaces are stable yet. It should not yet be used in production!

Contributions are welcome in whatever form, including testing, examples, use cases, or adding features. For an overview of current issues, checkout the roadmap or browse through the open issues on Github.

Finally, we're looking for early developer feedback -- if you do use Fission or Fission Workflows, we'd love to hear how it's working for you, what parts in particular you'd like to see improved, and so on.

Talk to us on slack or twitter.