http2

Documentation

Overview

http2 is a collection of functions meant to supplement the capabilities provided by the standard "net/http" package.

Index

• Constants
• func AcceptsGzipResponse(httpRequest *http.Request) bool
• func NewGzipResponseWriter(writer http.ResponseWriter, compressionLevel int) gzipResponseWriter
• type ConnectionParams
  • func DefaultPoolParams() ConnectionParams
• type DialError
• type LBPoolInstanceInfo
• type LBStrategy
• type LoadBalancedPool
  • func NewLoadBalancedPool(params ConnectionParams) *LoadBalancedPool
  • func (pool *LoadBalancedPool) Close()
  • func (pool *LoadBalancedPool) Do(req *http.Request) (*http.Response, error)
  • func (pool *LoadBalancedPool) Get() (*http.Client, error)
  • func (pool *LoadBalancedPool) GetInstancePool(instanceId int) (*SimplePool, error)
  • func (pool *LoadBalancedPool) SetStrategy(strategy LBStrategy)
  • func (pool *LoadBalancedPool) Update(instanceInfos []LBPoolInstanceInfo)
• type Pool
• type SimplePool
  • func NewSimplePool(addr string, params ConnectionParams) *SimplePool
  • func (pool *SimplePool) Close()
  • func (pool *SimplePool) Do(req *http.Request) (resp *http.Response, err error)
  • func (pool *SimplePool) Get() (*http.Client, error)

Constants

const (
	GzipEncoding          string = "gzip"
	AcceptEncodingHeader         = "Accept-Encoding"
	ContentTypeHeader            = "Content-Type"
	ContentEncodingHeader        = "Content-Encoding"
)

const (
	// Reasonable default for HTTP connect timeouts
	DefaultConnectTimeout = time.Second

	// Reasonable default for HTTP timeouts
	DefaultTimeout = 5 * time.Second

	// Reasonable default for maximum idle connections
	DefaultMaxIdle = 10
)

Functions

func AcceptsGzipResponse

func AcceptsGzipResponse(httpRequest *http.Request) bool

func NewGzipResponseWriter

func NewGzipResponseWriter(writer http.ResponseWriter, compressionLevel int) gzipResponseWriter

compressionLevel - one of the compression levels in the gzip package.

Types

type ConnectionParams

type ConnectionParams struct {
	// Number of idle HTTP clients we allow to remain in the pool
	MaxIdle int

	// Use SSL transport?
	UseSSL bool

	// Skip verification of server SSL certificates?
	SkipVerifySSL bool

	// Timeout for connection (includes DNS resolution)
	ConnectTimeout time.Duration

	// Timeout for waiting for an HTTP response header
	ResponseTimeout time.Duration

	// Host header to use instead of address.
	HostHeader *string

	// Dial function to use instead of the default
	Dial func(network, addr string) (net.Conn, error)
}

func DefaultPoolParams

func DefaultPoolParams() ConnectionParams

type DialError

type DialError struct {
	errors.DropboxError
}

type LBPoolInstanceInfo

type LBPoolInstanceInfo struct {
	InstanceId int
	Addr       string
}

type LBStrategy

type LBStrategy int

const (
	// In 'RoundRobin' load balancing strategy requests are sent to
	// different hosts in round robin fashion.
	LBRoundRobin LBStrategy = 0
	// In 'Fixed' load balancing strategy requests are routed to same host,
	// others are used only in case of failover.
	LBFixed LBStrategy = 1
)

type LoadBalancedPool

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

func NewLoadBalancedPool

func NewLoadBalancedPool(params ConnectionParams) *LoadBalancedPool

func (*LoadBalancedPool) Close

func (pool *LoadBalancedPool) Close()

func (*LoadBalancedPool) Do

func (pool *LoadBalancedPool) Do(req *http.Request) (*http.Response, error)

Issues an HTTP request, distributing more load to relatively unloaded instances.

func (*LoadBalancedPool) Get

func (pool *LoadBalancedPool) Get() (*http.Client, error)

Checks out an HTTP connection from an instance pool, favoring less loaded instances.

func (*LoadBalancedPool) GetInstancePool

func (pool *LoadBalancedPool) GetInstancePool(instanceId int) (*SimplePool, error)

Returns a SimplePool for given instanceId, or an error if it does not exist. TODO(zviad): right now this scans all instances, thus if there are a lot of instances per partition it can become very slow. If it becomes a problem, fix it!

func (*LoadBalancedPool) SetStrategy

func (pool *LoadBalancedPool) SetStrategy(strategy LBStrategy)

Sets Load Balancing strategy. Must be called before pool is actually put to use.

func (*LoadBalancedPool) Update

func (pool *LoadBalancedPool) Update(instanceInfos []LBPoolInstanceInfo)

type Pool

type Pool interface {
	// Similar interface as net/http.Client.Do()
	// Most important note is that: Callers should close resp.Body
	// when done reading from it. If resp.Body is not closed, the
	// Client's underlying RoundTripper (typically Transport) may not
	// be able to re-use a persistent TCP connection to the server
	// for a subsequent "keep-alive" request.
	Do(*http.Request) (*http.Response, error)

	// Returns http.Client to perform http requests with, preferable
	// to just use Do() function instead of this.
	Get() (*http.Client, error)

	// Closes underlying connection pool.
	Close()
}

A generic interface for HTTP connection pools

type SimplePool

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

Pool of persistent HTTP connections. The only limit is on the max # of idle connections we cache. Like Python's dropbox.curllib.CurlConnectionPool.

func NewSimplePool

func NewSimplePool(addr string, params ConnectionParams) *SimplePool

Creates a new HTTP connection pool using the given address and pool parameters.

'addr' is a net.Dial()-style 'host:port' destination for making the TCP connection for HTTP/HTTPS traffic. It will be used as the hostname by default for virtual hosting and SSL certificate validation; if you'd like to use a different hostname, set params.HostHeader.

func (*SimplePool) Close

func (pool *SimplePool) Close()

Closes all idle connections in this pool

func (*SimplePool) Do

func (pool *SimplePool) Do(req *http.Request) (resp *http.Response, err error)

Performs the HTTP request using our HTTP client

func (*SimplePool) Get

func (pool *SimplePool) Get() (*http.Client, error)

Returns the HTTP client, which is thread-safe.

Note that we use http.Client, rather than httputil.ClientConn, despite http.Client being higher- level. This is normally a liability for backend code, but it has more robust error handling and provides functionality that's more comparable to pycurl/curllib.