Serverless OpenWhisk Plugin

This plugin enables support for the OpenWhisk platform within the Serverless Framework.

Getting Started

Register account with OpenWhisk

Before you can deploy your service to OpenWhisk, you need to have an account registered with the platform.

• Want to run the platform locally? Please read the project's Quick Start guide for deploying it locally.
• Want to use a hosted provider? Please sign up for an account with IBM Bluemix and then follow the instructions for getting access to OpenWhisk on Bluemix.

Set up account credentials

Account credentials for OpenWhisk can be provided through a configuration file or environment variables. This plugin requires the API endpoint, namespace and authentication credentials.

Do you want to use a configuration file for storing these values? Please follow the instructions for setting up the OpenWhisk command-line utility. This tool stores account credentials in the .wskprops file in the user's home directory. The plugin automatically extracts credentials from this file at runtime. No further configuration is needed.

Do you want to use environment variables for credentials? Use the following environment variables to be pass in account credentials. These values override anything extracted from the configuration file.

• OW_APIHOST - Platform endpoint, e.g. openwhisk.ng.bluemix.net
• OW_AUTH - Authentication key, e.g. xxxxxx:yyyyy
• OW_APIGW_ACCESS_TOKEN - API gateway access token (optional)

Install Framework & Dependencies

Due to an outstanding issue with provider plugins, the OpenWhisk provider must be installed as a global module.

$ npm install --global serverless serverless-openwhisk

This framework plugin requires Node.js runtime version 6.0 or above.

Create Service From Template

Using the create command, you can create an example service from the following template.

serverless create --template openwhisk-nodejs --path my_service
cd my_service
npm install

More service examples are available in the serverless-examples repository.

Using a self-hosted version of the platform?

Ensure you set the ignore_certs option in the serverless.yaml prior to deployment.

provider:
  name: openwhisk
  ignore_certs: true

Deploy Service

The sample service from the template can be deployed without modification.

serverless deploy

If the deployment succeeds, the following messages will be printed to the console.

$ serverless deploy
Serverless: Packaging service...
Serverless: Compiling Functions...
Serverless: Compiling API Gateway definitions...
Serverless: Compiling Rules...
Serverless: Compiling Triggers & Feeds...
Serverless: Deploying Functions...
Serverless: Deployment successful!
 
Service Information
platform:	openwhisk.ng.bluemix.net
namespace:	_
service:	my_service
 
actions:
my_service-dev-hello
 
triggers:
**no triggers deployed***
 
rules:
**no rules deployed**
 
endpoints:
**no routes deployed**
 
web-actions:
**no web actions deployed**

Test Service

Use the invoke command to test your newly deployed service.

$ serverless invoke --function hello
{
    "payload": "Hello, World!"
}
$ serverless invoke --function hello --data '{"name": "OpenWhisk"}'
{
    "payload": "Hello, OpenWhisk!"
}

Writing Functions - Node.js

Here's an index.js file containing an example handler function.

function main(params) {
  const name = params.name || 'World';
  return {payload:  'Hello, ' + name + '!'};
};
 
exports.main = main;

Modules should return the function handler as a custom property on the global exports object.

In the serverless.yaml file, the handler property is used to denote the source file and module property containing the serverless function.

functions:
  my_function:
    handler: index.main

Request Properties

OpenWhisk executes the handler function for each request. This function is called with a single argument, an object containing the request properties.

function main(params) {
  const parameter = params.parameter_name;
  ...
};

Function Return Values

The handler must return an object from the function call. Returning undefined or null will result in an error. If the handler is carrying out an asynchronous task, it can return a Promise.

// synchronous return 
function main () {
  return { payload: "..." }
}
 
// asychronous return 
function main(args) {
  return new Promise(function(resolve, reject) {
    setTimeout(function() {
      resolve({ done: true });
     }, 2000);
  })
}

If you want to return an error message, return an object with an error property with the message. Promise values that are rejected will be interpreted as runtime errors.

// synchronous return 
function main () {
  return { error: "..." }
}
 
// asychronous return 
function main(args) {
  return new Promise(function(resolve, reject) {
    setTimeout(function() {
      reject("error message");
     }, 2000);
  })
}

Using NPM Modules

NPM modules must be installed locally in the node_modules directory before deployment. This directory will be packaged up in the deployment artefact. Any dependencies included in node_modules will be available through require() in the runtime environment.

OpenWhisk provides a number of popular NPM modules in the runtime environment. Using these modules doesn't require them to be included in the deployment package. See this list for full details of which modules are available.

const leftPad = require("left-pad")
 
function pad_lines(args) {
    const lines = args.lines || [];
    return { padded: lines.map(l => leftPad(l, 30, ".")) }
};
 
exports.handler = pad_lines;

Writing Functions - Python

Here's an index.py file containing an example handler function.

def endpoint(params):
    name = params.get("name", "stranger")
    greeting = "Hello " + name + "!"
    print(greeting)
    return {"greeting": greeting}

In the serverless.yaml file, the handler property is used to denote the source file and module property containing the serverless function.

functions:
  my_function:
    handler: index.endpoint
    runtime: python

Request Properties

OpenWhisk executes the handler function for each request. This function is called with a single argument, a dictionary containing the request properties.

def endpoint(params):
    name = params.get("name", "stranger")
    ...

Function Return Values

The handler must return a dictionary from the function call.

def endpoint(params):
    ...
    return {"foo": "bar"}

If you want to return an error message, return an object with an error property with the message.

Writing Functions - Swift

Here's an index.swift file containing an example handler function.

func main(args: [String:Any]) -> [String:Any] {
    if let name = args["name"] as? String {
      return [ "greeting" : "Hello \(name)!" ]
    } else {
      return [ "greeting" : "Hello stranger!" ]
    }
}

In the serverless.yaml file, the handler property is used to denote the source file and module property containing the serverless function.

functions:
  my_function:
    handler: index.main
    runtime: swift

Request Properties

OpenWhisk executes the handler function for each request. This function is called with a single argument, a dictionary containing the request properties.

func main(args: [String:Any]) -> [String:Any] {
    let prop = args["prop"] as? String
}

Function Return Values

The handler must return a dictionary from the function call.

func main(args: [String:Any]) -> [String:Any] {
    ...
    return ["foo": "bar"]
}

If you want to return an error message, return an object with an error property with the message.

Writing Functions - Pre-Compiled Swift Binaries

OpenWhisk supports creating Swift actions from a pre-compiled binary. This reduces startup time for Swift actions by removing the need for a dynamic compilation step.

In the serverless.yaml file, the handler property can refer to the compiled binary file produced by the build.

functions:
  hello:
    handler: .build/release/Hello

This configuration will generate the deployment package for that function containing only this binary file. It will not include other local files, e.g. Swift source files.

Pre-compiled Swift actions must be compatible with the platform runtime and architecture. There is an open-source Swift package (OpenWhiskAction) that handles wrapping functions within a shim to support runtime execution.

import OpenWhiskAction

func hello(args: [String:Any]) -> [String:Any] {
    if let name = args["name"] as? String {
      return [ "greeting" : "Hello \(name)!" ]
    } else {
      return [ "greeting" : "Hello stranger!" ]
    }
}

OpenWhiskAction(main: hello)

Binaries produced by the Swift build process must be generated for the correct platform architecture. This Docker command will compile Swift sources files using the relevant Swift environment.

docker run --rm -it -v $(pwd):/swift-package openwhisk/swift3action bash -e -c "cd /swift-package && swift build -v -c release"

Writing Functions - Binary

OpenWhisk supports executing a compiled binary for the function handler. Using a Python wrapper, the file will be invoked within the openwhisk/dockerskeleton Docker container.

The binary must be compiled for the correct platform architecture and only link to shared libraries installed in the openwhisk/dockerskeleton runtime.

In the serverless.yaml file, the handler property is used to denote the binary file to upload.

functions:
  my_function:
    handler: bin_file
    runtime: binary

Request Properties

OpenWhisk executes the binary file for each request. Event parameters are streamed to stdio as a JSON object string.

Function Return Values

The handler must write a JSON object string with the response parameters to stdout before exiting.

If you want to return an error message, return an object with an error property with the message.

Writing Functions - Docker

OpenWhisk supports custom runtimes using public images on Docker Hub. These images are expected to support the platform API used to instantiate and invoke serverless functions.

All necessary files for execution must be provided within the image. Local source files will not be uploaded to the runtime environment.

In the serverless.yaml file, the handler property is used to denote the image label.

functions:
  my_function:
    handler: repo/image_name
    runtime: docker

Runtime Configuration Properties

The following OpenWhisk configuration properties are supported for functions defined in the serverless.yaml file.

functions:
  my_function:
    handler: file_name.handler_func
    runtime: 'runtime_label' // defaults to nodejs:default
    namespace: "..." // defaults to user-provided credentials
    memory: 256 // 128 to 512 (MB).
    timeout: 60 // 0.1 to 600 (seconds)
    properties:
      foo: bar // default parameters
    anotations:
      foo: bar // action annotations

Writing Sequences

OpenWhisk supports a special type of serverless function called sequences.

These functions are defined from a list of other serverless functions. Upon invocation, the platform executes each function in series. Request parameters are passed into the first function in the list. Each subsequent function call is passed the output from the previous step as input parameters. The last function's return value is returned as the response result.

Here's an example of the configuration to define a sequence function, composed of three other functions.

functions:
  my_function:
    sequence:
      - parse_input
      - do_some_algorithm
      - construct_output

Sequence functions do not have a handler file defined. If you want to refer to functions not defined in the serverless project, use the fully qualified identifier e.g. /namespace/package/action_name

Connecting HTTP Endpoints

Functions can be bound to public URL endpoints using the API Gateway service. HTTP requests to configured endpoints will invoke functions on-demand. Requests parameters are passed as function arguments. Function return values are serialised as the JSON response body.

HTTP endpoints for functions can be configured through the serverless.yaml file.

functions:
  my_function:
    handler: index.main
    events:
      - http: GET /api/greeting

API Gateway hosts serving the API endpoints will be shown during deployment.

$ serverless deploy
...
endpoints:
GET https://xxx-gws.api-gw.mybluemix.net/service_name/api/path --> service_name-dev-my_function

Calling the configured API endpoints will execute the deployed functions.

$ http get https://xxx-gws.api-gw.mybluemix.net/api/greeting?user="James Thomas"
HTTP/1.1 200 OK
Content-Type: application/json; charset=UTF-8
Date: Mon, 19 Dec 2016 15:47:53 GMT
 
{
    "message": "Hello James Thomas!"
}

IMPORTANT: API Gateway support is currently experimental and may be subject to breaking changes.

Exporting Web Actions

Functions can be turned into "web actions" which return HTTP content without use of an API Gateway. This feature is enabled by setting an annotation (web-export) in the configuration file.

functions:
  my_function:
    handler: index.main
    annotations:
      web-export: true

Functions with this annotation can be invoked through a URL template with the following parameters.

https://{APIHOST}/api/v1/experimental/web/{USER_NAMESPACE}/{PACKAGE}/{ACTION_NAME}.{TYPE}

• APIHOST - platform endpoint e.g. openwhisk.ng.bluemix.net.
• USER_NAMESPACE - this must be an explicit namespace and cannot use the default namespace (_).
• PACKAGE - action package or default.
• ACTION_NAME - default form ${servicename}-${space}-${name}.
• TYPE - .json, .html, .text or .http.

Return values from the function are used to construct the HTTP response. The following parameters are supported.

1. headers: a JSON object where the keys are header-names and the values are string values for those headers (default is no headers).
2. code: a valid HTTP status code (default is 200 OK).
3. body: a string which is either plain text or a base64 encoded string (for binary data).

Here is an example of returning HTML content:

function main(args) {
    var msg = "you didn't tell me who you are."
    if (args.name) {
        msg = `hello ${args.name}!`
    }
    return {body:
       `<html><body><h3><center>${msg}</center></h3></body></html>`}
}

Here is an example of returning binary data:

function main() {
   let png = <base 64 encoded string>
   return { 
      headers: { "Content-Type": "image/png" },
      body: png };
}

Functions can access request parameters using the following environment variables.

1. **__ow_meta_verb:** the HTTP method of the request.
2. **__ow_meta_headers:** the request headers.
3. **__ow_meta_path:** the unmatched path of the request.

Full details on this new feature are available in this blog post.

IMPORTANT: Web Actions is currently experimental and may be subject to breaking changes.

Scheduled Invocations

Functions can be set up to fire automatically using the alarm package. This allows you to invoke functions with preset parameters at specific times (12:00 each day) or according to a schedule (every ten minutes).

Scheduled invocation for functions can be configured through the serverless.yaml file.

The schedule event configuration is controlled by a string, based on the UNIX crontab syntax, in the format cron(X X X X X). This can either be passed in as a native string or through the rate parameter.

functions:
  my_function:
    handler: index.main
    events: 
      - schedule: cron(* * * * *) // fires each minute.

This above example generates a new trigger (${service}_crawl_schedule_trigger) and rule (${service}_crawl_schedule_rule) during deployment.

Other schedule event parameters can be manually configured, e.g trigger or rule names.

functions:
  aggregate:
    handler: statistics.handler
    events:
      - schedule:
          rate: cron(0 * * * *) // call once an hour
          trigger: triggerName
          rule: ruleName
          max: 10000 // max invocations, default: 1000, max: 10000
          params: // event params for invocation
            hello: world

IBM Message Hub Events

IBM Bluemix provides an "Apache Kafka"-as-a-Service called IBM Message Hub. Functions can be connected to fire when messages arrive on Kafka topics.

IBM Message Hub instances can be provisioned through the IBM Bluemix platform. OpenWhisk on Bluemix will export Message Hub service credentials bound to a package with the following name:

/${BLUEMIX_ORG}_${BLUEMIX_SPACE}/Bluemix_${SERVICE_NAME}_Credentials-1

Rather than having to manually define all the properties needed by the Message Hub trigger feed, you can reference a package to use instead. Credentials from the referenced package will be used when executing the trigger feed.

Developers only need to add the topic to listen to for each trigger.

# serverless.yaml 
functions:
    index:
        handler: users.main
        events:
            - message_hub: 
                package: /${BLUEMIX_ORG}_${BLUEMIX_SPACE}/Bluemix_${SERVICE_NAME}_Credentials-1
                topic: my_kafka_topic
 

The plugin will create a trigger called ${serviceName}_${fnName}_messagehub_${topic} and a rule called ${serviceName}_${fnName}_messagehub_${topic}_rule to bind the function to the message hub events.

The trigger and rule names created can be set explicitly using the trigger andrule parameters.

Other functions can bind to the same trigger using the inline trigger event referencing this trigger name.

# serverless.yaml 
functions:
    index:
        handler: users.main
        events:
            - message_hub: 
                package: /${BLUEMIX_ORG}_${BLUEMIX_SPACE}/Bluemix_${SERVICE_NAME}_Credentials-1
                topic: my_kafka_topic
                trigger: log_events
                rule: connect_index_to_kafka 
     another:
        handler: users.another
        events:
            - trigger: log_events 

Using Manual Parameters

Parameters for the Message Hub event source can be defined explicitly, rather than using pulling credentials from a package.

# serverless.yaml 
functions:
    index:
        handler: users.main
        events:
            - message_hub: 
                topic: my_kafka_topic
                brokers: afka01-prod01.messagehub.services.us-south.bluemix.net:9093
                user: USERNAME
                password: PASSWORD
                admin_url:  https://kafka-admin-prod01.messagehub.services.us-south.bluemix.net:443
                json: true
                binary_key: true
                binary_value: true

topic, brokers, user, password and admin_url are mandatory parameters.

Cloudant DB Events

IBM Cloudant provides a hosted NoSQL database, based upon CouchDB, running on IBM Bluemix. Functions can be connected to events fired when the database is updated. These events use the CouchDB changes feed to follow database modifications.

IBM Cloudant instances can be provisioned through the IBM Bluemix platform. OpenWhisk on Bluemix will export Cloudant service credentials bound to a package with the following name:

/${BLUEMIX_ORG}_${BLUEMIX_SPACE}/Bluemix_${SERVICE_NAME}_Credentials-1

Rather than having to manually define all the properties needed by the Cloudant trigger feed, you can reference a package to use instead. Credentials from the referenced package will be used when executing the trigger feed.

Developers only need to add the database name to follow for modifications.

# serverless.yaml 
functions:
    index:
        handler: users.main
        events:
            - cloudant: 
                package: /${BLUEMIX_ORG}_${BLUEMIX_SPACE}/Bluemix_${SERVICE_NAME}_Credentials-1
                db: my_db_name
 

The plugin will create a trigger called ${serviceName}_${fnName}_cloudant_${topic} and a rule called ${serviceName}_${fnName}_cloudant_${topic}_rule to bind the function to the Cloudant update events.

The trigger and rule names created can be set explicitly using the trigger andrule parameters.

Other functions can bind to the same trigger using the inline trigger event referencing this trigger name.

Using Manual Parameters

Parameters for the Cloudant event source can be defined explicitly, rather than using pulling credentials from a package.

# serverless.yaml 
functions:
    index:
        handler: users.main
        events:
            - cloudant: 
                host: xxx-yyy-zzz-bluemix.cloudant.com
                username: USERNAME
                password: PASSWORD
                db: db_name

Custom Event Triggers

Functions are connected to event sources in OpenWhisk using triggers and rules. Triggers create a named event stream within the system. Triggers can be fired manually or connected to external data sources, like databases or message queues.

Rules set up a binding between triggers and serverless functions. With an active rule, each time a trigger is fired, the function will be executed with the trigger payload.

Event binding for functions can be configured through the serverless.yaml file.

functions:
  my_function:
    handler: index.main
    events:
      - trigger: my_trigger

This configuration will create a trigger called servicename-my_trigger with an active rule binding my_function to this event stream.

Customising Rules

Rule names default to the following format servicename-trigger-to-action. These names be explicitly set through configuration.

functions:
  my_function:
    handler: index.main
    events:
      - trigger:
        name: "my_trigger"
        rule: "rule_name"

Customing Triggers

Triggers can be defined as separate resources in the serverless.yaml file. This allows you to set up trigger properties like default parameters.

functions:
  my_function:
    handler: index.main
    events:
      - trigger: my_trigger
      
resources:
  triggers:
    my_trigger:
      parameters: 
        hello: world            

Trigger Feeds

Triggers can be bound to external event sources using the feed property. OpenWhisk provides a catalogue of third-party event sources bundled as packages.

This example demonstrates setting up a trigger which uses the /whisk.system/alarms/alarm feed. The alarm feed will fire a trigger according to a user-supplied cron schedule.

resources:
  triggers:
    alarm_trigger:
      parameters: 
        hello: world
      feed: /whisk.system/alarms/alarm
      feed_parameters: 
        cron: '*/8 * * * * *'

Commands

The following serverless commands are currently implemented for the OpenWhisk provider.

• deploy - Deploy functions, triggers and rules for service.
• invoke- Invoke deployed serverless function and show result.
• invokeLocal- Invoke serverless functions locally and show result.
• remove - Remove functions, triggers and rules for service.
• logs - Display activation logs for deployed function.
• info - Display details on deployed functions, triggers and rules.