Serverless Development

Tools

• AWS Lambda
• Google Cloud Functions
• Azure Functions
• IBM OpenWhisk
• hook.io
• OVH Functions
• Snafu
• Podilizer, Termite (FaaSification of Java)
• Lambada (FaaSification of Python)

Overlays

PyWren

No explicit deployment prior to execution, deploy while executing

def my_fun(b):
    ...

pwex = pywren.default_executor()
res = pwex.map(my_fun, params)

Gee’s Lambada

Deployment with dependencies (requirements.txt references Lambada framework)

def my_fun(b):
    ...

tune = Lambada(role=..., region=..., memoery=128)

@tune.dancer
def my_fun_lambda(e, c):
    my_fun(e['stddev'])

FaaSification

Process of automated decomposition of software into a set of function-level services: code analysis + transformation + deployment + on-demand activation

Transformation rules

Entry points: no transformation of main function

Function definitions:

• Adapt to conventions: parameters, return value
• Scan recursively for function calls
• Export as function unit including dependencies

Function calls:

• if internal, rewire
• if I/O, replace
• otherwise, leave as is

Monads:

• Functional programming with side effects / side channels

Open research problems

• clustering
• code inspection (deep FaaSification)

Deployment

• Implicit: web IDE
• Explicit: provider tools

Provider specific

• Authorisation (e.g. function calling another function)
• Packaging

Abstraction

Abstraction approach: Serverless FW

serverless create --template spotinst-nodejs --path P

serverless config credentials --provider sptinst --token T --account A

serverless deploy function -f hello
serverless invoke -f hello

serverless.yml:

service: hello
provider:
    name: spotinst
    spotinst:
        environment:...
plugins:
    - serverless-spotinst-functions

functions:
    hello:
        handler: index.main
        runtime: nodejs8
        memory: 128
        timeout: 20
        access: public

Prototyping

Prototyping approach: Localstack

Emulation of commercial FaaS offerings and includion into CI

localstack start [--docker]

alias aws="aws --endpoint-url=http://localhost:4574"
aws lambda list-functions ...

Debugging

Problems

• Crash
• Missing dependency
• Slowness
• Undeterministic behaviour

Tools

• Insight
• Visualisation
• Debugging/Profiling tools (wskdebug for OpenWhisk)
• Feedback
• Autotuning

K-rules: Analyse -> Plan -> Execute -> Monitor

Analyse

• Stacktraces
• Profiles
• Anomalies over time

Plan

• Recombination of function clusters
• Memory reconfiguration

Execute

• Hot deployment
• Canary testing
• Step/Breakpoint execution

Monitor

• Invocation counts
• Invocation time
• Throttling
• Faults
• Variables

Feeds into:

• FaaS Debugging, Profiling and Autotuning Catalogue
• Management functions