patron

README

patron

Patron is a framework for creating microservices.

Patron is french for template or pattern, but it means also boss which we found out later (no pun intended).

The entry point of the framework is the Service. The Service uses Components to handle the processing of sync and async requests. The Service starts by default a HTTP Component which hosts the debug, health and metric endpoints. Any other endpoints will be added to the default HTTP Component as Routes. The service set's up by default logging with zerolog, tracing and metrics with jaeger and prometheus.

Patron provides abstractions for the following functionality of the framework:

• service, which orchestrates everything
• components and processors, which provide a abstraction of adding processing functionality to the service
  • asynchronous message processing (RabbitMQ, Kafka)
  • synchronous processing (HTTP)
• metrics and tracing
• logging

Patron provides same defaults for making the usage as simple as possible.

Service

The Service has the role of glueing all of the above together, which are:

• setting up logging
• setting up termination by os signal
• starting and stopping components
• handling component errors
• setting up metrics and tracing

The service has some default settings which can be changed via environment variables:

• Service HTTP port, for setting the default HTTP components port to 50000 with PATRON_HTTP_DEFAULT_PORT
• Log level, for setting zerolog with INFO log level with PATRON_LOG_LEVEL
• Tracing, for setting up jaeger tracing with
  • agent address 0.0.0.0:6831 with PATRON_JAEGER_AGENT
  • sampler type probabilisticwith PATRON_JAEGER_SAMPLER_TYPE
  • sampler param 0.1 with PATRON_JAEGER_SAMPLER_PARAM

Component

A Component is a interface that exposes the following API:

type Component interface {
  Run(ctx context.Context) error
  Shutdown(ctx context.Context) error
}

The above API gives the Service the ability to start and gracefully shutdown a component. The framework divides the components in 2 categories:

• synchronous, which are components that follow the request/response pattern and
• asynchronous, which consume messages from a source but don't respond anything back

The following component implementations are available:

• HTTP (sync)
• RabbitMQ consumer (async)
• Kafka consumer (async)

Adding to the above list is as easy as implementing a Component and a Processor for that component.

Example

Setting up a new service with a HTTP Component is as easy as the following code:

  // Set up HTTP routes
  routes := make([]sync_http.Route, 0)
  routes = append(routes, sync_http.NewRoute("/", http.MethodGet, processor, true))
  
  srv, err := patron.New("test", patron.Routes(routes))
  if err != nil {
    log.Fatalf("failed to create service %v", err)
  }

  err = srv.Run()
  if err != nil {
    log.Fatalf("failed to create service %v", err)
  }

The above is pretty much self-explanatory. The processor follows the sync pattern.

Processors

Synchronous

The implementation of the processor is responsible to create a Request by providing everything that is needed (Headers, Fields, decoder, raw io.Reader) pass it to the implementation by invoking the Process method and handle the Response or the error returned by the processor.

The sync package contains only a function definition along with the models needed:

type ProcessorFunc func(context.Context, *Request) (*Response, error)

The Request model contains the following properties (which are provided when calling the "constructor" NewRequest)

• Fields, which may contain any fields associated with the request
• Raw, the raw request data (if any) in the form of a io.Reader
• decode, which is a function of type encoding.Decode that decodes the raw reader

A exported function exists for decoding the raw io.Reader in the form of

Decode(v interface{}) error

The Response model contains the following properties (which are provided when calling the "constructor" NewResponse)

• Payload, which may hold a struct of type interface{}

Asynchronous

The implementation of the async processor follows exactly the same principle as the sync processor. The main difference is that:

• The Request is the Message and contains only data as []byte
• There is no Response, so the processor may return a error

type ProcessorFunc func(context.Context, *Message) error

Everything else is exactly the same.

Metrics and Tracing

Tracing and metrics are provided by jaeger's implementation of the OpenTracing project. Every component has been integrated with the above library and produces traces and metrics. Metrics are provided with the default HTTP component at the /metrics route for Prometheus to scrape. Tracing will be send to a jaeger agent which can be setup though environment variables mentioned in the config section. Sane defaults are applied for making the use easy. We have included some clients inside the trace package which are instrumented and allow propagation of tracing to downstream systems. The tracing information is added to each implementations header. These clients are:

• HTTP
• AMQP
• Kafka

Logging

The log package is designed to be a leveled logger with field support.

The log package defines two interfaces (Logger and Factory) that have to be implemented in order to set up the logging in this framework. After implementing the two interfaces you can setup logging by doing the following:

  // instantiate the implemented factory and fields (map[string]interface{})
  err := log.Setup(factory, fields)
  // handle error

If the setup is omitted the package will not setup any logging!

In order to get a logger use the the following:

  logger := log.Create()

which returns a logger with all fields appended along with the source code file of the class where the logger has been created.

It is advisable to create one logger in every type/package once and use it to get information about the log origin!

The implementations should support following log levels:

• Debug, which should log the message with debug level
• Info, which should log the message with info level
• Warn, which should log the message with warn level
• Error, which should log the message with error level
• Panic, which should log the message with panic level and panics
• Fatal, which should log the message with fatal level and terminates the application

The first four (Debug, Info, Warn and Error) give the opportunity to differentiate the messages by severity. The last two (Panic and Fatal) do the same and do additional actions (panic and termination).

The package supports fields, which are logged along with the message, to augment the information further to ease querying in the log management system.

The following implementations are provided as sub-package:

• zerolog, which supports the excellent zerolog library and is set up by default

Logger

The logger interface defines the actual logger.

type Logger interface {
  Level() Level
  Fields() map[string]interface{}
  Fatal(...interface{})
  Fatalf(string, ...interface{})
  Panic(...interface{})
  Panicf(string, ...interface{})
  Error(...interface{})
  Errorf(string, ...interface{})
  Warn(...interface{})
  Warnf(string, ...interface{})
  Info(...interface{})
  Infof(string, ...interface{})
  Debug(...interface{})
  Debugf(string, ...interface{})
}

In order to be consistent with the design the implementation of the Fatal(f) have to terminate the application with an error and the Panic(f) need to panic.

Factory

The factory interface defines a factory for creating a logger.

type Factory interface {
  Create(map[string]interface{}) Logger
}

Two methods are supported:

• Create, which creates a logger with the specified fields (or nil)

Documentation

Overview

Package patron framework

Patron is a framework for creating microservices.

Patron is french for template or pattern, but it means also boss which we found out later (no pun intended).

The entry point of the framework is the Service. The Service uses Components to handle the processing of sync and async requests. The Service can setup as many Components it wants, even multiple HTTP components provided the port does not collide. The Service starts by default a HTTP component which hosts the debug, health and metric endpoints. Any other endpoints will be added to the default HTTP Component as Routes. The service set's up by default logging with zerolog, tracing and metrics with jaeger and prometheus.

Patron provides abstractions for the following functionality of the framework:

• service, which orchestrates everything
• components and processors, which provide a abstraction of adding processing functionality to the service
• asynchronous message processing (RabbitMQ, Kafka)
• synchronous processing (HTTP)
• metrics and tracing
• logging
• configuration management

Patron provides same defaults for making the usage as simple as possible. For more details please check out github repository https://github.com/mantzas/patron.

Index

• func SetupLogging(name, version string) error
• type Component
• type OptionFunc
  • func Components(cc ...Component) OptionFunc
  • func HealthCheck(hcf http.HealthCheckFunc) OptionFunc
  • func Routes(rr []http.Route) OptionFunc
• type Service
  • func New(name, version string, oo ...OptionFunc) (*Service, error)
  • func (s *Service) Run() error
  • func (s *Service) Shutdown() error

Functions

func SetupLogging

func SetupLogging(name, version string) error

SetupLogging set's up default logging.

Types

type Component

type Component interface {
	Run(ctx context.Context) error
	Shutdown(ctx context.Context) error
}

Component interface for implementing service components.

type OptionFunc

type OptionFunc func(*Service) error

OptionFunc definition for configuring the service in a functional way.

func Components

func Components(cc ...Component) OptionFunc

Components option for adding additional components to the service.

func HealthCheck

func HealthCheck(hcf http.HealthCheckFunc) OptionFunc

HealthCheck option for overriding the default health check of the default HTTP component.

func Routes

func Routes(rr []http.Route) OptionFunc

Routes option for adding routes to the default HTTP component.

type Service

type Service struct {
	// contains filtered or unexported fields
}

Service is responsible for managing and setting up everything. The service will start by default a HTTP component in order to host management endpoint.

func New

func New(name, version string, oo ...OptionFunc) (*Service, error)

New creates a new named service and allows for customization through functional options.

func (*Service) Run

func (s *Service) Run() error

Run starts up all service components and monitors for errors. If a component returns a error the service is responsible for shutting down all components and terminate itself.

func (*Service) Shutdown

func (s *Service) Shutdown() error

Shutdown all components gracefully with a predefined timeout.