background v0.1.0-beta.2

Fast background job processing

Onyx::Background

Built with Crystal Travis CI build API docs Latest release

A fast background job processing for Crystal.

About

This is a Redis-based background jobs processing for Crystal, alternative to sidekiq and mosquito. It's a component of Onyx Framework, but can be used separately.

Goals

Just like all other @onyxframework components, Onyx::Background aims to be as much novice-friendly as possible, still being able to scale with a developer's knowledge of Crystal, thus having these goals:

  • Speed — taking the best from Crystal performance
  • Simplicity — the shard API is simpler than alternatives'
  • Modularity — most of the components are replaceable
  • Expandability — every component may be re-opened and added with new functionality
  • Failure-safety — the whole system is stateless, allowing any of its parts to fail and re-run safely without locking overhead
  • No Crystal lock-in — a simple manager can be written in any language with Redis driver

Features

This publicly available open-source version has the following functionality:

  • Enqueuing jobs for immediate processing
  • Enqueuing jobs for delayed processing (i.e. in: 5.minutes or at: Time.now + 1.hour)
  • Different job queues ("default" by default)
  • Concurrent jobs processing (with a separate Redis client for each fiber)
  • Verbose Redis logging of all the activity (i.e. every attempt made)
  • Rescuing errors and moving failed jobs to the failed list
  • Moving stale jobs (i.e. with dead workers) to the failed list
  • CLI

If you want more features and professional support, consider purchasing a commercial license at onyxframework.com.

Performance

Thorough benchmaring is to be done yet, however, currently it is able to process more than 9000 jobs per second on my 0.9GHz machine with Redis instance running on itself.

It's over 9000!

Installation

⚠️ Note: Onyx::Background works with Redis version ~> 5.0

Add this to your application's shard.yml:

dependencies:
  onyx-background:
    github: onyxframework/background
    version: ~> 0.1.0

This shard follows Semantic Versioning v2.0.0, so check releases and change the version accordingly.

Usage

API docs

Please refer to API documentation available online: https://api.onyxframework.org/background

Note: it is automatically updated on every master branch commit

Example

require "onyx-background"

struct Jobs::Nap
  include Onyx::Background::Job

  def initialize(@sleep : Int32 = 1)
  end

  def perform
    sleep(@sleep)
  end
end

manager = Onyx::Background::Manager.new
manager.enqueue(Jobs::Nap.new)

puts "Enqueued"

logger = Logger.new(STDOUT, Logger::DEBUG)
worker = Onyx::Background::Worker.new(logger: logger)
worker.run
$ crystal app.cr
Enqueued
I -- worker: Working...
D -- worker: Waiting for a new job...
D -- worker: [fa5b6d65-46fe-4c88-829f-d69023c4c6de] Attempting
D -- worker: [fa5b6d65-46fe-4c88-829f-d69023c4c6de] Performing Jobs::Nap {"sleep":1}...
D -- worker: Waiting for a new job...
D -- worker: [fa5b6d65-46fe-4c88-829f-d69023c4c6de] Completed

It's also highly recommended to run a Watcher process to watch for stale jobs. If you're queuing delayed jobs, the Watcher is required to move the jobs to the ready queue on time:

# watcher.cr
require "onyx-background"

watcher = Onyx::Background::Watcher.new
watcher.run

CLI

A simple Onyx::Background::CLI is used to interact with Onyx::Background.

# src/background-cli.cr
require "onyx-background/cli"
$ crystal build -o cli src/background-cli.cr
$ ./cli -h
usage:
    onyx-background-cli [command] [options]
commands:
    status                           Display system status
options:
    -h, --help                       Show this help

Status

The status command displays the current system status:

$ ./cli status -h
usage:
    onyx-background-cli status [options]
options:
    -q, --queue QUEUE                Comma-separated queue(s) ("default")
    -r, --redis REDIS_URL            Redis URL
    -n, --namespace NAMESPACE        Redis namespace ("onyx-background")
    -v, --verbose                    Enable verbose mode
    -h, --help                       Show this help
$ ./cli status -q high,low
high
workers fibers  jps ready scheduled processing  completed failed
      4      4    0     0         0          0     500000      0

low
workers fibers  jps ready scheduled processing  completed failed
      0      0    0     0         0          0          0      0

Tip: use watch -n 1 ./cli status to continuously monitor the status

Architecture

In this section the system architecture is explained.

Storing a job

Tip: only this functionality is likely to be implemented in another language driver

A job hash is saved into Redis with an unique UUID at "jobs:<uuid>" key after Onyx::Background::Manager#enqueue call:

manager = Onyx::Background::Manager.new
manager.enqueue(Jobs::Nap.new, in: 5.minutes)
Key Example Description
que default The job's queue
cls Jobs::Nap The job's class, should be available in the Worker program
arg {"sleep":1} The job's arguments, usually a JSON string
qat 1545906554934 The time the job's been queued at
pat 1545906870368 The time the job's been scheduled to be processed at (none if immediate)

The job UUID is also pushed to the "ready:<queue>" list or "scheduled:<queue>" sorted set (with score equal to scheduled time in milliseconds) for further processing.

Working

When an Onyx::Background::Worker runs, its Redis client name is set to "onyx-background-worker:<queues>" (with comma-separated queues this worker is watching). It then BLPOPs the "ready:<queue>" list and spawns a separate fiber with a new instance of Redis client (it's pooled internally for better performance) for every popped job UUID.

This fiber's Redis client name is set to "onyx-background-worker-fiber:<worker_client_id>". Afterwards it saves an unique attempt hash for a particular job at "attempts:<attempt_uuid>" key, which looks like this:

Key Example Description
sta 1545907080109 The time the attempt was started at
job d62f4c9a-... The job's UUID
wrk 342 The fiber's Redis client ID
que default The actual processing queue

And adds the attempt UUID to the "processing:<queue>" set.

Thereafter, in case of successful job processing, the attempt is updated with finished at and processing time values:

Key Example Description
fin 1545907080110 The time the attempt was finished at
time 1.0 The job processing time, in milliseconds

If the job raised an error, the attempt is updated with these values:

Key Example Description
fin 1545907080110 The time the attempt was finished at
time 1.0 The job processing time, in milliseconds
err IndexOutOfBounds The unhandled exception class

The attempt UUID is then removed from the "processing:<queue>" set and added into "completed:<queue>" or "failed:<queue>" sorted set (with score equal to current time in milliseconds) depending on the result.

Watching

Onyx::Background::Watcher is used to watch for stale attempts (which workers are offline) and move jobs from the "scheduled:<queue>" sorted set to the "ready:<queue>" list:

require "onyx-background"

watcher = Onyx::Background::Watcher.new
watcher.run

The watcher loops every interval (defaults to 1 second) and does the following:

  1. Gets all the clients list from Redis and compares it with attempts stored in the "processing:<queue>" set. If an attempt's fiber client ID (the "wrk" key) is not found in the list of active clients, it's considered stale then and fails with a "Worker Timeout" error
  2. Gets all jobs from "scheduled:<queue>" sorted set which have a score less than Time.now.to_unix_ms and moves them to the "ready:<queue>" list.

⚠️ Note: in case of multiple watchers watching the same queue, a duplication error may appear!

For a deeper understaing of the architecture, feel free to dive into the source code, it's quite well documented!

Development

Redis is flushed during the spec, so you must specify a safe-to-flush Redis database in the REDIS_URL. To run the specs, use the following command:

$ env REDIS_URL=redis://localhost:6379/1 crystal spec

Contributing

  1. Fork it (https://github.com/onyxframework/background/fork)
  2. Create your feature branch (git checkout -b my-new-feature)
  3. Commit your changes (git commit -am 'Add some feature')
  4. Push to the branch (git push origin my-new-feature)
  5. Create a new Pull Request

Contributors

Licensing

This software is licensed under BSD 3-Clause License.

Open Source Initiative

Repository

background

Owner
Statistic
  • 11
  • 0
  • 3
  • 0
  • 2
  • almost 6 years ago
  • November 7, 2018
License

BSD 3-Clause "New" or "Revised" License

Links
Synced at

Wed, 06 Nov 2024 10:41:58 GMT

Languages