Skip to content

Latest commit

 

History

History
362 lines (261 loc) · 7.62 KB

README.md

File metadata and controls

362 lines (261 loc) · 7.62 KB

Fiber Scheduler

Ruby 3 has Fiber Scheduler hooks that enable asynchronous programming. In order to make this work you need a Fiber Scheduler implementation, but Ruby does not provide a default one.

This gem aims to fill that void by providing a Fiber Scheduler class that makes a great default. It's easy to use, performant, and can be used with just built-in Ruby methods.

fiber_scheduler's killer feature 💣 is full compatibility with any other Fiber Scheduler implementation, including the async gem. Write code using fiber_scheduler and it works seamlessly with async, bsync or whatever other _sync gem comes in the future.

Learn more about the Ruby's Fiber Scheduler feature.

Installation

gem install fiber_scheduler

Requires Ruby 3.1.

Highlights

  • Enables asynchronous (colorless) programming in Ruby.
  • Killer feature: full compatibility with any other Fiber Scheduler implementation, including the async gem.
  • Not a framework: no DSL or new APIs. Can be used with just built-in Ruby methods: Fiber.set_scheduler and Fiber.schedule.
  • ~500 LOC of pure Ruby, no C extensions.
  • No dependencies.

Setup

  1. With a block (recommended)
  2. Set Fiber.scheduler directly

With a block (recommended)

FiberScheduler do
  # Your code here, e.g. Fiber.schedule { ... }
end

Recommended because:

Set Fiber.scheduler directly

Fiber.set_scheduler(FiberScheduler.new)

# Your code here, e.g. Fiber.schedule { ... }

Fiber.scheduler is set until the end of the current thread, unless manually unset with Fiber.set_scheduler(nil).

Pros:

  • Uses only built-in Ruby methods Fiber.set_scheduler and Fiber.schedule.

Cons:

  • No compatibility with other fiber schedulers.

Examples

Basic example

This example runs two HTTP requests in parallel:

require "fiber_scheduler"
require "open-uri"

FiberScheduler do
  Fiber.schedule do
    URI.open("https://httpbin.org/delay/2")
  end

  Fiber.schedule do
    URI.open("https://httpbin.org/delay/2")
  end
end

Advanced example

This example runs various operations in parallel. The example total running time is slightly more than 2 seconds, which indicates all the operations ran in parallel.

Note that all the operations used in Fiber.schedule blocks below are either common gems or built-in Ruby methods. They all work asynchronously with this library, no monkey patching!

require "fiber_scheduler"
require "httparty"
require "open-uri"
require "redis"
require "sequel"

DB = Sequel.postgres
Sequel.extension(:fiber_concurrency)

FiberScheduler do
  Fiber.schedule do
    # This HTTP request takes 2 seconds (slightly more because of the latency)
    URI.open("https://httpbin.org/delay/2")
  end

  Fiber.schedule do
    # Use any HTTP library
    HTTParty.get("https://httpbin.org/delay/2")
  end

  Fiber.schedule do
    # Works with any TCP protocol library
    Redis.new.blpop("abc123", 2)
  end

  Fiber.schedule do
    # Make database queries
    DB.run("SELECT pg_sleep(2)")
  end

  Fiber.schedule do
    sleep 2
  end

  Fiber.schedule do
    # Run system commands
    `sleep 2`
  end
end

Scaling example

Easily run thousands and thousands of blocking operations in parallel. This program finishes in about 2.5 seconds.

require "fiber_scheduler"

FiberScheduler do
  10_000.times do
    Fiber.schedule do
      sleep 2
    end
  end
end

Gotcha: be careful about the overheads when scaling things. The below snippet runs sleep which is an "inexpensive" operation. But, if we were to run thousands of network requests there would be more overhead (establishing TCP connections, SSL handshakes etc) which would prolong program running time.

Nested Fiber.schedule example

It's possible to nest Fiber.schedule blocks arbitrarily deep.

All the sleep operations in this snippet run in parallel and the program finishes in 2 seconds total.

require "fiber_scheduler"

FiberScheduler do
  Fiber.schedule do
    Fiber.schedule do
      sleep 2
    end

    Fiber.schedule do
      sleep 2
    end

    sleep 2
  end

  Fiber.schedule do
    sleep 2
  end
end

Waiting Fiber.schedule example

Sometimes it's conventient for the parent to wait on the child fiber to complete. Use Fiber.schedule(:waiting) to achieve that.

In the below example fiber labeled parent will wait for the child fiber to complete. Note that only the parent fiber waits, other fibers run as usual.

This example takes 4 seconds to finish.

require "fiber_scheduler"

FiberScheduler do
  Fiber.schedule do # parent
    Fiber.schedule(:waiting) do # child
      sleep 2
    end
    # The fiber stops here until the waiting child fiber completes.

    sleep 2
  end

  Fiber.schedule do
    sleep 2
  end
end

Blocking Fiber.schedule example

Blocking fibers "block" all the other fibers from running until they're finished.

This example takes 4 seconds to finish.

require "fiber_scheduler"

FiberScheduler do
  Fiber.schedule do
    Fiber.schedule(:blocking) do
      sleep 2
    end
  end

  Fiber.schedule do
    sleep 2
  end
end

Volatile Fiber.schedule example

Volatile fibers end when all the "durable" fibers finish. Volatile fibers (by design) may not complete all their work.

This is useful if you have a neverending task that performs some cleanup work that should finish when the rest of the program completes.

This example takes 2 seconds to finish.

require "fiber_scheduler"

FiberScheduler do
  Fiber.schedule(:volatile) do
    # This fiber will live for only 2 seconds.

    loop do
      cleanup_work # this method will run only once

      sleep 10
    end
  end

  Fiber.schedule do
    sleep 2
  end
end

Compatibility with other fiber schedulers

async is an awesome asynchronous programming library, if not a framework. If async is like Rails, then fiber_scheduler is plain Ruby.

fiber_scheduler is fully compatible with async:

Async do |task|
  task.async do
    # code ...
  end

  FiberScheduler do
    Fiber.schedule do
      # code ...
    end
  end

  # ...
end

Note that currently the opposite doesn't work:

FiberScheduler do
  Async do
    # ...
  end

  Fiber.schedule do # No scheduler is available! (RuntimeError)
    # ...
  end
end

Other Fiber Scheduler implementations

fiber_scheduler gem works with any other Fiber Scheduler class (current and future ones). Example:

Fiber.set_scheduler(AnotherScheduler.new)

# stuff

FiberScheduler do
  # works just fine
end

# more stuff

fiber_scheduler is like choosing pure Ruby: it's a safe choice because you know it works and will continue working with everything else in Ruby's asynchronous eco-system.

Performance

This basic perf benchmark looks promising.

HINT: make sure to install io-event gem alongside fiber_scheduler for a performance improvement.

Credits

Samuel Williams for:

  • Implementing Ruby's Fiber Scheduler hooks.
  • The default selector used in this gem.

License

MIT