README
¶
Note: This project is no longer actively maintained. Please refer to its spiritual successor rmq.
--
redismq
What is this
This is a fast, persistent, atomic message queue implementation that uses redis as its storage engine written in go. It uses atomic list commands to ensure that messages are delivered only once in the right order without being lost by crashing consumers.
Details can be found in the blog post about its initial design: http://big-elephants.com/2013-09/building-a-message-queue-using-redis-in-go/
A second article desribes the performance improvements of the current version: http://big-elephants.com/2013-10/tuning-redismq-how-to-use-redis-in-go/
What it's not
It's not a standalone server that you can use as a message queue, at least not for now. The implementation is done purely client side. All message queue commands are "translated" into redis commands and then executed via a redis client.
If you want to use this with any other language than go you have to translate all of the commands into your language of choice.
How to use it
All most all use cases are either covered in the examples or in the tests.
So the best idea is just to read those and figure it from there. But in any case:
Basics
To get started you need a running redis server. Since the tests run FlushDB() an otherwise unused database is highly recommended
The first step is to create a new queue:
package main
import (
"fmt"
"github.com/adjust/redismq"
)
func main() {
testQueue := redismq.CreateQueue("localhost", "6379", "", 9, "clicks")
...
}
To write into the queue you simply use Put():
...
testQueue := redismq.CreateQueue("localhost", "6379", "", 9, "clicks")
testQueue.Put("testpayload")
...
}
The payload can be any kind of string, yes even a 10MB one.
To get messages out of the queue you need a consumer:
...
consumer, err := testQueue.AddConsumer("testconsumer")
if err != nil {
panic(err)
}
package, err := consumer.Get()
if err != nil {
panic(err)
}
fmt.Println(package.Payload)
...
}
Payload will hold the original string, while package will have some additional header information.
To remove a package from the queue you have to Ack() it:
...
package, err := consumer.Get()
if err != nil {
panic(err)
}
err = package.Ack()
if err != nil {
panic(err)
}
...
}
Buffered Queues
When input speed is of the essence BufferedQueues will scratch that itch.
They pipeline multiple puts into one fast operation. The only issue is that upon crashing or restart the packages in
the buffer that haven't been written yet will be lost. So it's advised to wait one second before terminating your program to flush the buffer.
The usage is as easy as it gets:
...
bufferSize := 100
testQueue := redismq.CreateBufferedQueue("localhost", "6379", "", 9, "clicks", bufferSize)
testQueue.Start()
...
}
Put() and Get() stay exactly the same.
I have found anything over 200 as bufferSize not to increase performance any further.
To ensure that no packages are left in the buffer when you shut down your program you need to call
FlushBuffer() which will tell the queue to flush the buffer and wait till it's empty.
testQueue.FlushBuffer()
Multi Get
Like BufferedQueues for Get() MultiGet() speeds up the fetching of messages. The good news it comes without the buffer loss issues.
Usage is pretty straight forward with the only difference being the MultiAck():
...
packages, err := consumer.MultiGet(100)
if err != nil {
panic(err)
}
for i := range packages {
fmt.Println(p[i].Payload)
}
packages[len(p)-1].MultiAck()
...
}
MultiAck() can be called on any package in the array with all the prior packages being "acked". This way you can Fail() single packages.
Reject and Failed Queues
Similar to AMQP redismq supports Failed Queues meaning that packages that are rejected by a consumer will be stored in separate queue for further inspection. Alternatively a consumer can also Requeue() a package and put it back into the queue:
...
package, err := consumer.Get()
if err != nil {
panic(err)
}
err = package.Requeue()
if err != nil {
panic(err)
}
...
}
To push the message into the Failed Queue of this consumer simply use Fail():
...
package, err := consumer.Get()
if err != nil {
panic(err)
}
err = package.Fail()
if err != nil {
panic(err)
}
package, err = suite.consumer.GetUnacked()
...
}
As you can see there is also a command to get messages from the Failed Queue.
How fast is it
Even though the original implementation wasn't aiming for high speeds the addition of BufferedQueues and MultiGet
make it go something like this.
All of the following benchmarks were conducted on a MacBook Retina with a 2.4 GHz i7. The InputRate is the number of messages per second that get inserted, WorkRate the messages per second consumed.
Single Publisher, Two Consumers only atomic Get and Put
InputRate: 12183
WorkRate: 12397
Single Publisher, Two Consumers using BufferedQueues and MultiGet
InputRate: 46994
WorkRate: 25000
And yes that is a persistent message queue that can move over 70k messages per second.
If you want to find out for yourself checkout the example folder. The load.go or buffered_queue.go
will start a web server that will display performance stats under http://localhost:9999/stats.
How persistent is it
As redis is the underlying storage engine you can set your desired persistence somewhere between YOLO and fsync(). With somewhat sane settings you should see no significant performance decrease.
Copyright
redismq is Copyright © 2014 adjust GmbH.
It is free software, and may be redistributed under the terms specified in the LICENSE file.
Documentation
¶
Index ¶
- func SetSSLMode(enabled bool)
- type BufferedQueue
- type Consumer
- func (consumer *Consumer) Get() (*Package, error)
- func (consumer *Consumer) GetFailed() (*Package, error)
- func (consumer *Consumer) GetUnacked() (*Package, error)
- func (consumer *Consumer) GetUnackedLength() int64
- func (consumer *Consumer) HasUnacked() bool
- func (consumer *Consumer) MultiGet(length int) ([]*Package, error)
- func (consumer *Consumer) NoWaitGet() (*Package, error)
- func (consumer *Consumer) Quit()
- func (consumer *Consumer) RequeueWorking() error
- func (consumer *Consumer) ResetWorking() error
- type ConsumerStat
- type Observer
- type Package
- type Queue
- func (queue *Queue) AddConsumer(name string) (c *Consumer, err error)
- func (queue *Queue) Delete() error
- func (queue *Queue) GetFailedLength() int64
- func (queue *Queue) GetInputLength() int64
- func (queue *Queue) Put(payload string) error
- func (queue *Queue) RequeueFailed() error
- func (queue *Queue) ResetFailed() error
- func (queue *Queue) ResetInput() error
- type QueueStat
- type Server
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func SetSSLMode ¶
func SetSSLMode(enabled bool)
Types ¶
type BufferedQueue ¶
type BufferedQueue struct {
*Queue
BufferSize int
Buffer chan *Package
// contains filtered or unexported fields
}
BufferedQueue provides an queue with buffered writes for increased performance. Only one buffered queue (per name) can be started at a time. Before terminating the queue should be flushed using FlushBuffer() to avoid package loss
func CreateBufferedQueue ¶
func CreateBufferedQueue(redisHost, redisPort, redisPassword string, redisDB int64, name string, bufferSize int) *BufferedQueue
CreateBufferedQueue returns BufferedQueue. To start writing the buffer to redis use Start(). Optimal BufferSize seems to be around 200. Works like SelectBufferedQueue for existing queues
func SelectBufferedQueue ¶
func SelectBufferedQueue(redisHost, redisPort, redisPassword string, redisDB int64, name string, bufferSize int) (queue *BufferedQueue, err error)
SelectBufferedQueue returns a BufferedQueue if a queue with the name exists
func (*BufferedQueue) FlushBuffer ¶
func (queue *BufferedQueue) FlushBuffer()
FlushBuffer tells the background writer to flush the buffer to redis
func (*BufferedQueue) Put ¶
func (queue *BufferedQueue) Put(payload string) error
Put writes the payload to the buffer
func (*BufferedQueue) Start ¶
func (queue *BufferedQueue) Start() error
Start dispatches the background writer that flushes the buffer. If there is already a BufferedQueue running it will return an error.
type Consumer ¶
Consumer are used for reading from queues
func (*Consumer) GetFailed ¶
GetFailed returns a single packages from the failed queue of this consumer
func (*Consumer) GetUnacked ¶
GetUnacked returns a single packages from the working queue of this consumer
func (*Consumer) GetUnackedLength ¶
GetUnackedLength returns the number of packages in the unacked queue
func (*Consumer) HasUnacked ¶
HasUnacked returns true if the consumers has unacked packages
func (*Consumer) NoWaitGet ¶
NoWaitGet returns a single package from the queue (returns nil, nil if no package in queue)
func (*Consumer) RequeueWorking ¶
RequeueWorking requeues all packages from working to input
func (*Consumer) ResetWorking ¶
ResetWorking deletes! all messages in the working queue of this consumer
type ConsumerStat ¶
ConsumerStat collects data about a queues consumer
type Observer ¶
Observer is a very simple implementation of an statistics observer far more complex things could be implemented with the way stats are written for now it allows basic access to throughput rates and queue size averaged over seconds, minutes and hours
func NewObserver ¶
NewObserver returns an Oberserver to monitor different statistics from redis
func (*Observer) GetAllQueues ¶
GetAllQueues returns a list of all registed queues
func (*Observer) UpdateAllStats ¶
func (observer *Observer) UpdateAllStats()
UpdateAllStats fetches stats for all queues and all their consumers
func (*Observer) UpdateQueueStats ¶
UpdateQueueStats fetches stats for one specific queue and its consumers
type Package ¶
type Package struct {
Payload string
CreatedAt time.Time
Queue interface{} `json:"-"`
Consumer *Consumer `json:"-"`
Collection *[]*Package `json:"-"`
Acked bool `json:"-"`
}
Package provides headers and handling functions around payloads
type Queue ¶
type Queue struct {
Name string
// contains filtered or unexported fields
}
Queue is the central element of this library. Packages can be put into or get from the queue. To read from a queue you need a consumer.
func CreateQueue ¶
CreateQueue return a queue that you can Put() or AddConsumer() to Works like SelectQueue for existing queues
func SelectQueue ¶
func SelectQueue(redisHost, redisPort, redisPassword string, redisDB int64, name string) (queue *Queue, err error)
SelectQueue returns a Queue if a queue with the name exists
func (*Queue) AddConsumer ¶
AddConsumer returns a conumser that can write from the queue
func (*Queue) Delete ¶
Delete clears all input and failed queues as well as all consumers will not proceed as long as consumers are running
func (*Queue) GetFailedLength ¶
GetFailedLength returns the number of packages in the failed queue
func (*Queue) GetInputLength ¶
GetInputLength returns the number of packages in the input queue
func (*Queue) RequeueFailed ¶
RequeueFailed moves all failed packages back to the input queue
func (*Queue) ResetFailed ¶
ResetFailed deletes all packages from the failed queue
func (*Queue) ResetInput ¶
ResetInput deletes all packages from the input queue
type QueueStat ¶
type QueueStat struct {
InputSizeSecond int64
InputSizeMinute int64
InputSizeHour int64
FailSizeSecond int64
FailSizeMinute int64
FailSizeHour int64
InputRateSecond int64
InputRateMinute int64
InputRateHour int64
WorkRateSecond int64
WorkRateMinute int64
WorkRateHour int64
ConsumerStats map[string]*ConsumerStat
}
QueueStat collects data about a queue
Source Files
Directories