@fastify/reply-from
Fastify plugin to forward the current HTTP request to another server.
HTTP2 to HTTP is supported too.
Install
npm i @fastify/reply-from
Compatibility with @fastify/multipart
@fastify/reply-from
and @fastify/multipart
should not be registered as sibling plugins nor should they be registered in plugins which have a parent-child relationship.<br>
The two plugins are incompatible, in the sense that the behavior of @fastify/reply-from
might not be the expected one when the above-mentioned conditions are not respected.<br>
This is due to the fact that @fastify/multipart
consumes the multipart content by parsing it, hence this content is not forwarded to the target service by @fastify/reply-from
.<br>
However, the two plugins may be used within the same fastify instance, at the condition that they belong to disjoint branches of the fastify plugins hierarchy tree.
Usage
The following example set up two Fastify servers and forward the request
from one to the other:
'use strict'
const Fastify = require('fastify')
const target = Fastify({
logger: true
})
target.get('/', (request, reply) => {
reply.send('hello world')
})
const proxy = Fastify({
logger: true
})
proxy.register(require('@fastify/reply-from'), {
base: 'http://localhost:3001/'
})
proxy.get('/', (request, reply) => {
reply.from('/')
})
target.listen({ port: 3001 }, (err) => {
if (err) {
throw err
}
proxy.listen({ port: 3000 }, (err) => {
if (err) {
throw err
}
})
})
API
Plugin options
base
Set the base URL for all the forwarded requests. Will be required if http2
is set to true
Note that every path will be discarded.
Custom URL protocols unix+http:
and unix+https:
can be used to forward requests to a unix
socket server by using querystring.escape(socketPath)
as the hostname. This is not supported
for http2 nor undici. To illustrate:
const socketPath = require('node:querystring').escape('/run/http-daemon.socket')
proxy.register(require('@fastify/reply-from'), {
base: 'unix+http://${socketPath}/'
});
undici
By default, undici will be used to perform the HTTP/1.1
requests. Enabling this flag should guarantee
20-50% more throughput.
This flag could controls the settings of the undici client, like so:
proxy.register(require('@fastify/reply-from'), {
base: 'http://localhost:3001/',
undici: {
connections: 128,
pipelining: 1,
keepAliveTimeout: 60 * 1000,
tls: {
rejectUnauthorized: false
}
}
})
See undici own options for more configurations.
You can also pass the plugin a custom instance:
proxy.register(require('@fastify/reply-from'), {
base: 'http://localhost:3001/',
undici: new undici.Pool('http://localhost:3001')
})
http
Set the http
option to an Object to use
Node's http.request
will be used if you do not enable http2
. To customize the request
,
you can pass in agentOptions
and
requestOptions
. To illustrate:
proxy.register(require('@fastify/reply-from'), {
base: 'http://localhost:3001/',
http: {
agentOptions: {
keepAliveMsecs: 10 * 60 * 1000
},
requestOptions: {
timeout: 5000
}
}
})
You can also pass custom HTTP agents. If you pass the agents, then the http.agentOptions will be ignored. To illustrate:
proxy.register(require('@fastify/reply-from'), {
base: 'http://localhost:3001/',
http: {
agents: {
'http:': new http.Agent({ keepAliveMsecs: 10 * 60 * 1000 }),
'https:': new https.Agent({ keepAliveMsecs: 10 * 60 * 1000 })
},
requestOptions: {
timeout: 5000
}
}
})
http2
You can either set http2
to true
or set the settings object to connect to a HTTP/2 server.
The http2
settings object has the shape of:
proxy.register(require('@fastify/reply-from'), {
base: 'http://localhost:3001/',
http2: {
sessionTimeout: 10000,
requestTimeout: 5000,
sessionOptions: {
rejectUnauthorized: true
},
requestOptions: {
endStream: true
}
}
})
disableRequestLogging
By default package will issue log messages when a request is received. By setting this option to true, these log messages will be disabled.
Default for disableRequestLogging
will be false
. To disable the log messages set disableRequestLogging
to true
.
proxy.register(require('@fastify/reply-from'), {
base: 'http://localhost:3001/',
disableRequestLogging: true
})
cacheURLs
The number of parsed URLs that will be cached. Default: 100
.
disableCache
This option will disable the URL caching.
This cache is dedicated to reduce the amount of URL object generation.
Generating URLs is a main bottleneck of this module, please disable this cache with caution.
contentTypesToEncode
An array of content types whose response body will be passed through JSON.stringify()
.
This only applies when a custom body
is not passed in. Defaults to:
[
'application/json'
]
retryMethods
On which methods should the connection be retried in case of socket hang up.
Be aware that setting here not idempotent method may lead to unexpected results on target.
By default: ['GET', 'HEAD', 'OPTIONS', 'TRACE']
This plugin will always retry on 503 errors, unless retryMethods
does not contain GET
.
globalAgent
Enables the possibility to explictly opt-in for global agents.
Usage for undici global agent:
import { setGlobalDispatcher, ProxyAgent } from 'undici'
const proxyAgent = new ProxyAgent('my.proxy.server')
setGlobalDispatcher(proxyAgent)
fastify.register(FastifyReplyFrom, {
base: 'http://localhost:3001/',
globalAgent: true
})
Usage for http/https global agent:
fastify.register(FastifyReplyFrom, {
base: 'http://localhost:3001/',
globalAgent: true,
http: {
}
})
destroyAgent
If set to true
, it will destroy all agents when the Fastify is closed.
If set to false
, it will not destroy the agents.
By Default: false
maxRetriesOn503
This plugin will always retry on GET
requests that returns 503 errors, unless retryMethods
does not contain GET
.
This option set the limit on how many times the plugin should retry the request, specifically for 503 errors.
By Default: 10
retryDelay
This plugin gives the client an option to pass their own retry callback to allow the client to define what retryDelay they would like on any retries
outside the scope of what is handled by default in fastify-reply-from. To see the default please refer to index.js getDefaultDelay()
If a handler
is passed to the retryDelay
object the onus is on the client to invoke the default retry logic in their callback otherwise default cases such as 500 will not be handled
err
is the error thrown by making a request using whichever agent is configuredreq
is the raw request details sent to the underlying agent. Note: this object is not a Fastify request object, but instead the low-level request for the agent.res
is the raw response returned by the underlying agent (if available) Note: this object is not a Fastify response, but instead the low-level response from the agent. This property may be null if no response was obtained at all, like from a connection reset or timeout.attempt
in the object callback refers to the current retriesAttempt number. You are given the freedom to use this in concert with the retryCount property set to handle retriesgetDefaultRetry
refers to the default retry handler. If this callback returns not null and you wish to handle those case of errors simply invoke it as done below.
Given example
const customRetryLogic = ({err, req, res, attempt, getDefaultRetry}) => {
const defaultDelay = getDefaultDelay();
if (defaultDelay) return defaultDelay();
if (res && res.statusCode === 500 && req.method === 'GET') {
return 300
}
if (err && err.code == "UND_ERR_SOCKET"){
return 600
}
return null
}
.......
fastify.register(FastifyReplyFrom, {
base: 'http://localhost:3001/',
retryDelay: customRetryLogic
})
Note the Typescript Equivalent
const customRetryLogic = ({req, res, err, getDefaultRetry}: RetryDetails) => {
...
}
...
reply.from(source, [opts])
The plugin decorates the
Reply
instance with a from
method, which will reply to the original request
from the desired source. The options allows to override any part of
the request or response being sent or received to/from the source.
Note: If base
is specified in plugin options, the source
here should not override the host/origin.
onResponse(request, reply, res)
Called when a HTTP response is received from the source.
The default behavior is reply.send(res)
, which will be disabled if the
option is specified.
When replying with a body of a different length it is necessary to remove
the content-length
header.
{
onResponse: (request, reply, res) => {
reply.removeHeader('content-length');
reply.send('New body of different length');
}
}
onError(reply, error)
Called when a HTTP response is received with error from the source.
The default behavior is reply.send(error)
, which will be disabled if the
option is specified.
It must reply the error.
Called to rewrite the headers of the response, before them being copied
over to the outer response.
Parameters are the original headers and the Fastify request.
It must return the new headers object.
Called to rewrite the headers of the request, before them being sent to the other server.
Parameters are the Fastify request and the original request headers.
It must return the new headers object.
getUpstream(request, base)
Called to get upstream destination, before the request is being sent. Useful when you want to decide which target server to call based on the request data.
Helpful for a gradual rollout of new services.
Parameters are the Fastify request and the base string from the plugin options.
It must return the upstream destination.
Only http1! As http2 uses one connection for the whole session only the base upstream is used. If you want to
have different upstreams based on the request you can add multiple Fastify.register's with different
ContraintStrategies.
e.g.:
Route grpc-web/http1 and grpc/http2 to different routes with a ContentType-ConstraintStrategy:
const contentTypeMatchContraintStrategy = {
// strategy name for referencing in the route handler `constraints` options
name: 'contentType',
// storage factory for storing routes in the find-my-way route tree
storage: function () {
let handlers = {}
return {
get: (type: any) => { return handlers[type] || null },
set: (type: any, store: any) => { handlers[type] = store }
}
},
// function to get the value of the constraint from each incoming request
deriveConstraint: (req: any, ctx: any) => {
return req.headers['content-type']
},
// optional flag marking if handlers without constraints can match requests that have a value for this constraint
mustMatchWhenDerived: true
}
server.addConstraintStrategy(contentTypeMatchContraintStrategy);
and then 2 different upstreams with different register's:
// grpc-web / http1
server.register(fastifyHttpProxy, {
// Although most browsers send with http2, nodejs cannot handle this http2 request
// therefore we have to transport to the grpc-web-proxy via http1
http2: false,
upstream: 'http://grpc-web-proxy',
constraints: { "contentType": "application/grpc-web+proto" }
});
// grpc / http2
server.register(fastifyHttpProxy, {
http2: true,
upstream: 'http://grpc.server',
constraints: { "contentType": "application/grpc+proto" }
});
queryString
or queryString(search, reqUrl)
Replaces the original querystring of the request with what is specified.
This will be passed to
querystring.stringify
.
object
: accepts an object that will be passed to querystring.stringify
function
: function that will return a string with the query parameters e.g. name=test&type=user
body
Replaces the original request body with what is specified. Unless
contentType
is specified, the content will be passed
through JSON.stringify()
.
Setting this option for GET, HEAD requests will throw an error "Rewriting the body when doing a {GET|HEAD} is not allowed".
Setting this option to null
will strip the body (and content-type
header) entirely from the proxied request.
method
Replaces the original request method with what is specified.
retriesCount
How many times it will try to pick another connection on socket hangup (ECONNRESET
error).
Useful when keeping the connection open (KeepAlive).
This number should be a function of the number of connections and the number of instances of a target.
By default: 0 (disabled)
contentType
Override the 'Content-Type'
header of the forwarded request, if we are
already overriding the body
.
formbody
expects the body to be returned as a string and not an object.
Use the contentTypesToEncode
option to pass in ['application/x-www-form-urlencoded']
HTTP & HTTP2 timeouts
This library has:
timeout
for http
set by default. The default value is 10 seconds (10000
).requestTimeout
& sessionTimeout
for http2
set by default.
- The default value for
requestTimeout
is 10 seconds (10000
). - The default value for
sessionTimeout
is 60 seconds (60000
).
When a timeout happens, 504 Gateway Timeout
will be returned to the client.
TODO
License
MIT