Skip to content

Auto-scaling client for Apple Push Notification HTTP/2 service done "the Go way".

License

Notifications You must be signed in to change notification settings

baobabus/go-apns

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

51 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Go APNS

⚠️ WORK IN PROGRESS ⚠️

Go APNS is a client for Apple Push Notification service over HTTP/2 protocol done "the Go way".

Build Status GoDoc

Features

  • Designed to communicate with Apple Push Notification service over HTTP/2 protocol
  • Processes push requests asynchronously: just write your requests to a channel
  • Notifies about push results asynchronously: just receive on the callback channel
  • Automatically scales up processing pipeline as the load increases
  • Allows full control of the scaling process and connection handling
  • Effects back pressure as needed to ensure full awareness by the up-stream
  • Supports Go 1.7 and later

Processing Flow

Each APNS Client handles all aspects of communication with APN service, including management of HTTP/2 connections and controlling concurrent HTTP/2 streams.

Processing Flow Diagram

(Control channels and service goroutines are not shown)

  1. Submitter picks up push requests from the processing queue
  2. Submitter forwards requests to internal dispatch channel
  3. Each streamer maintains a single HTTP/2 connection to APN service
  4. One of the streamers picks up a push request from the dispatch queue
  5. The streamer allocates a stream in its HTTP/2 connection
  6. The streamer spins up a single-flight round-tripper goroutine
  7. The round-tripper synchronously POSTs a request to APN service over its streamer's HTTP/2 connection
  8. APN server's response is written to the callback channel
  9. Governor collects metrics for dispatch and callback channel blockages, evaluates processing throughput and spins up new streamers as needed

Scaling up

During each poll interval governor collects stats on inbound and outbound channel blockages, on the number of notifications pushed and the size of data sent out. It then evalutes the stats against its scaling configuration and spins up new streamers if and when is appropriate.

In the following illustrative scenario:

  • PollInterval = 0.2sec
  • MinSustain = 1sec
  • SettlePeriod = 2sec

Scaling sequence

(Processing rate and bandwidth stats are not shown)

  1. Blockages on outbound channel prevent blockages on inbound channel from being counted
  2. Minimum time of sustained blockage on inbound channel is encountered
    • new streamers are spun up asynchronously
    • sustained blockages on inboud channel have no effect while new streamers are starting
  3. All new streamers have completed their initialization
    • settle period begins
    • sustained blockages on inboud channel have no effect during settle period
  4. Settle period ends
    • since there's been minimum time of sustained blockage on inbound channel, more streamers are spun up
    • sustained blockages on inboud channel have no effect while new streamers are starting
  5. All new streamers have completed their initialization
    • settle period begins
    • sustained blockages on inboud channel have no effect during settle period
  6. Blockages on inbound channel end - no more scaling up is needed.

Example

Fire-and-forget example sends a notification to three recipients. It uses provider token authentication method and does not check for outcome, just making sure the push is complete.

package main

import (
	"log"

	"github.com/baobabus/go-apns/apns2"
	"github.com/baobabus/go-apns/cryptox"
)

func main() {

	// Load and parse out token signing key
	signingKey, err := cryptox.PKCS8PrivateKeyFromFile("token_signing_pk.p8")
	if err != nil {
		log.Fatal("Token signing key error: ", err)
	}

	// Set up our client
	client := &apns2.Client{
		Gateway:  apns2.Gateway.Production,
		Signer:   &apns2.JWTSigner{
			KeyID: "ABC123DEFG", // Your key ID
			TeamID: "DEF123GHIJ", // Your team ID
			SigningKey: signingKey,
		},
		CommsCfg: apns2.CommsFast,
		ProcCfg:  apns2.UnlimitedProcConfig,
	}

	// Start processing
	err = client.Start(nil)
	if err != nil {
		log.Fatal("Client start error: ", err)
	}

	// Mock motification and recipients
	header := &apns2.Header{ Topic: "com.example.Alert" }
	payload := &apns2.Payload{ APS: &apns2.APS{Alert: "Ping!"} }
	recipients := []string{
		"00fc13adff785122b4ad28809a3420982341241421348097878e577c991de8f0",
		"10fc13adff785122b4ad28809a3420982341241421348097878e577c991de8f0",
		"20fc13adff785122b4ad28809a3420982341241421348097878e577c991de8f0",
	}

	// Push to all recipients
	for _, rcpt := range recipients {
		notif := &apns2.Notification{
			Recipient: rcpt,
			Header:    header,
			Payload:   payload,
		}
		err := client.Push(notif, apns2.DefaultSigner, apns2.NoContext, apns2.DefaultCallback)
		if err != nil {
			log.Fatal("Push error: ", err)
		}
	}

	// Perform soft shutdown allowing the processing to complete.
	client.Stop()
}

Configuration Settings and Customization

Communication Settings

Following communication settings are supported:

DialTimeout

DialTimeout is the maximum amount of time a dial will wait for a connect to complete.

MinDialBackOff

MinDialBackOff is the minimum amount of time by which dial attempts should be delayed after encountering a refused connection. Actual back-off time will grow exponentially until a connection attempt is successful.

MaxDialBackOff

MaxDialBackOff is the maximum amount of time by which dial attempts should be delayed after encountering a refused connection.

DialBackOffJitter

DialBackOffJitter is used to calculate the ramdom amount to appy to each back-off time calculation.

RequestTimeout

RequestTimeout specifies a time limit for requests made by the HTTPClient. The timeout includes connection time, any redirects, and reading the response body.

KeepAlive

KeepAlive specifies the keep-alive period for an active network connection. If zero, keep-alives are not enabled. Apple recommends not closing connections to APN service at all, but a sinsibly long duration is acceptable.

MaxConcurrentStreams

MaxConcurrentStreams is the maximum allowed number of concurrent streams per HTTP/2 connection. If connection's MAX_CONCURRENT_STREAMS option is invoked by the remote side with a lower value, the remote request will be honored if possible. (See AllowHTTP2Incursion processing option.)

CommsCfg example:

CommsCfg{
	DialTimeout:          1 * time.Second,
	MinDialBackOff:       4 * time.Second,
	MaxDialBackOff:       10 * time.Minute,
	DialBackOffJitter:    10 * funit.Percent,
	RequestTimeout:       2 * time.Second,
	KeepAlive:            10 * time.Hour,
	MaxConcurrentStreams: 500,
}

Processing Settings

Following processing settings are supported:

MaxRetries

MaxRetries is the maximum number of times a failed notification push should be reattempted. This only applies to "retriable" failures.

RetryEval

RetryEval is the function that is called when a push attempt fails and retry eligibility needs to be determined.

MinConns

MinConns is minimum number of concurrent connections to APN servers that should be kept open. When a client is started it immeditely attempts to open the specified number of connections.

MaxConns

MaxConns is maximum allowed number of concurrent connections to APN service.

MaxRate

MaxRate is the throughput cap specified in notifications per second. It is not strictly enforced as would be the case with a true rate limiter. Instead it only prevents additional scaling from taking place once the specified rate is reached.

For clarity it is best expressed in idiomatic way:

MaxRate = 10000 / funit.Second
MaxBandwidth

MaxBandwidth is the throughput cap specified in bits per second. It is not strictly enforced as would be the case with a true rate limiter. Instead it only prevents additional scaling from taking place once the specified rate is reached.

For clarity it is best expressed in idiomatic way:

MaxBandwidth = 10 * funit.Kilobit / funit.Second
Scale

Scale specifies the manner of scaling up and winding down. Three scaling modes come prefefined: Incremental, Exponential and Constant.

Scale = scale.Incremental(2) // Add two new connections each time
MinSustain

MinSustain is the minimum duration of time over which the processing has to experience blocking before a scale-up attemp is made. It is also the minimum amount of time over which non-blocking processing has to take place before a wind-down attemp is made.

PollInterval

PollInterval is the time between performance metrics sampling attempts.

SettlePeriod

SettlePeriod is the amount of time given to the processing for it to settle down at the new rate after successful scaling up or winding down attempt. Sustained performance analysis is ignored during this time and no new scaling attempt is made.

AllowHTTP2Incursion

AllowHTTP2Incursion controls whether it is OK to perform reflection-based probing of HTTP/2 layer. When enabled, scaler may access certain private properties in x/net/http2 package if needed for more precise performance analysis.

UsePreciseHTTP2Metrics

UsePreciseHTTP2Metrics, if set to true, instructs the scaler to query HTTP/2 layer parameters on every call that requires the data. Set this to false if you wish to eliminate any additional overhead that this may introduce.

HTTP2MetricsRefreshPeriod

HTTP2MetricsRefreshPeriod, if set to a positive value, controls the frequency of "imprecise" metrics updates. Under this approach any relevant fields that are private to x/net/http2 packaged are only queried periodically. This reduces the overhead of any required reflection calls, but it also introduces the risk of potentially relying on some stale metrics. In most realistic situations, however, this can be easily tolerated given frequent enough refresh period.

HTTP2MetricsRefreshPeriod value is ignored and periodic updates are turned off if UsePreciseHTTP2Metrics is set to true. Setting HTTP2MetricsRefreshPeriod to 0 or negative value disables metrics refresh even if UsePreciseMetrics is false.

ProcCfg example:

ProcCfg{
	MaxRetries:                0,
	RetryEval:                 nil,
	MinConns:                  1,
	MaxConns:                  100,
	MaxRate:                   100000 / funit.Second,
	MaxBandwidth:              10 * funit.Kilobit / funit.Second,
	Scale:                     scale.Incremental(2),
	MinSustain:                2 * time.Second,
	PollInterval:              200 * time.Millisecond,
	SettlePeriod:              5 * time.Second,
	AllowHTTP2Incursion:       true,
	UsePreciseHTTP2Metrics:    false,
	HTTP2MetricsRefreshPeriod: 200 * time.Millisecond,
}

License

The MIT License (MIT)

Copyright (c) 2017 Aleksey Blinov

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

About

Auto-scaling client for Apple Push Notification HTTP/2 service done "the Go way".

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages