Thread Pool Updates

[jdantonio]

Addresses #248, #227, and #149.

Changelog

• Added at_exit handler to Ruby thread pools (already in Java thread pools)
  • Ruby handler stores the object id and retrieves from ObjectSpace
  • JRuby disables ObjectSpace by default so that handler stores the object reference
• Added a :stop_on_exit option to thread pools to enable/disable at_exit handler
• Updated thread pool docs to better explain shutting down thread pools
• Simpler :executor option syntax for all abstractions which support this option
• Added Executor#auto_terminate? predicate method (for thread pools)
• Added at_exit handler to TimerSet
• Simplified auto-termination of the global executors
  • Can now disable auto-termination of global executors
  • Added shutdown/kill/wait_for_termination variants for global executors
• Can now disable auto-termination for all executors (the nuclear option)
• Simplified auto-termination of the global executors
• Deprecated terms "task pool" and "operation pool"
  • New terms are "io executor" and "fast executor"
  • New functions added with new names
  • Deprecation warnings added to functions referencing old names
• Moved all thread pool related functions from Concurrent::Configuration to Concurrent
  • Old functions still exist with deprecation warnings
  • New functions have updated names as appropriate
• All high-level abstractions default to the "io executor"
• Fixed bug in Actor causing it to prematurely warm global thread pools on gem load
  • This also fixed a RejectedExecutionError bug when running with minitest/autorun via JRuby
• Added LazyReference, a simpler and faster varition of Delay
  • Updated most internal uses of Delay with LazyReference
• Moved global logger up to the Concurrent namespace and refactored the code

To-Do

From #248

• Add a new option to the constructor of the Java-backed thread pools that controls the creation of the at_exit handler
  • The default will be to enable the at_exit handler
  • This will support the "simple and easy" base case
  • And will ensure backward compatibility with the current release
• Create an auto_shutdown? (or similarly named) predicate method to our Java-backed thread pools that indicates whether the at_exit handler has been set
• Update the pure-Ruby thread pools with the exact same at_exit behavior as the Java-backed thread pools
  • It was an oversight that our pure-Ruby thread pools are not 100% compatible with our Java-backed thread pools
  • However, we need to make sure that this behavior is backward compatible with the current release
• Update the global config to ensure only one at_exit handler ever exists for each global thread pool
  • Our default global thread pools should be created without their own at_exit handlers
  • [ ] Let the at_exit handlers created within the global config handle the global thread pools
• Update the behavior of the global configuration auto_terminate feature
  • Remove unnecessary respond_to? checks
  • [ ] Ensure proper handing of at_exit handler creation for all thread pools automatically created within the global config
  • [ ] Ensure proper handling of at_exit when the user explicitly sets a global thread pool
• Create a Configuration.shutdown_global_thread_pools (or similarly named) method
  • Give advanced users who disable the at_exit handlers the ability to explicitly shutdown the global thread pools
  • Potentially create shutdown/kill/wait methods that mirror the methods on the thread pools themselves
• Create a global "autotest friendly" configuration option
  • [ ] It should include a require 'concurrent/autotest' feature for simple setup
  • It should prevent any at_exit handlers from being created for all thread pools, regardless of how they are created
  • Yet it should also ensure that the JVM should not hang because the thread pools are not shutdown
  • And it should be as unobtrusive as possible (I hope to avoid coupling thread pools to some global registry if possible)
• Document, document, document

From #149

• keep them lazy initialized
• rename: task -> short-non-blocking, operation -> long-blocking
• [ ] add more global executors like an immediate executor
• remove configuration object, move global pools to Concurrent space
• possibly improve parsing of :executor option
  • [ ] to also handle agent two pools
  • to understand :immediate too
  • etc.
• what to do with logging?

[jdantonio]

ALL: With the exception of documentation, I believe this PR to be complete. I plan to merge it Tuesday night after work. All feedback is welcome.

[pitr-ch]

@jdantonio I was on PTO for a week so I am still catching up on the conversations. Could you give me little more time to go over it?

[jdantonio]

@pitr-ch Absolutely!

[pitr-ch]

I think the --format documentation was there to see the last test where it was hanging. Is there another way?

[jdantonio]

I'm not sure if there's another way, but --format documentation has been in .travis.yml since June of last year. I only removed it a couple of weeks ago--then immediately added it back when we had a hanging test. I think we should keep it for now. Once we stabilize our tests we can remove it, but the full output is pretty valuable right now.

[pitr-ch]

Thanks for all the work on executors! And sorry that it took me so long to get to it :/