Serverless Applications

FaaS

• Inspired by functional programming
• Stateless
• Short-lived
• Single-function(ality) microservices
• Implied pricing model: memory + per-invocation

Programming model

• Heterogenous functions with specific signature (params, return values)
• order of invocation can expressed by workflows

Deployment model

• upload source or compiled binaries
• configure entry point, memory, limits

Execution model

• time and space limits
• local state limits (by time or no state at all)
• pay-per-use microbilling
• highly parallel autoscaled exeuction

Pricing

• Determined by provider
• Determined by application through service descriptions or annotations (emerging topic)

Service model

Control plane

• Deployment
• Configuration
• Authorisation
• Logging
• Billing

Computing plane

• Platform
• Compute
• Storage
• Network

Execution

Event -> trigger -> rules -> actions -> results

Events

• Monitoring event
• Sensor data
• Log entry
• …

Triggers

• HTTP
• XMPP
• AMQP

Rules: e.g. max 1 per hour Actions: Function invocation Result: e.g. JSON return value

Cloud Functions

Structure

• fixed or registered name
• parameter or via context
• single return value

Language

• in theory any
• in practice dominance of Node and Python

Configuration

• Runtime language, entry file
• Runtime parameters, memory, max duration, env variables

12factor alignment: declarative setup, config in environment, backing services for state handling, disposability / fast startup & graceful shutdown

System view

• Function: Developer-supplied code (app bundle)
• Function Encapsulation: Provider-supplied container image (runtime, libs)
• Function Execution: Provider-supplied infrastructure (hidden specs)

Configuration

• runtime environment (from list)
• memory 128-512 MB; default 256
• timeout 0.1-300 s; default 60
• authoring: template or blank document