Concurrency 2 (Tue Apr 5, lect 21) | previous | next | slides |

Continuing the discussion on this critical aspect of scaling

Logistics

Magic Word is
Status of projects
Quick review of Implement and test the validation API
Discuss architecture of OTP

Tuning Threads and Processes

Why

In tuning an application for performance you need to configure the web server, app server, database server
How many threads, how many processes, etc
If done wrong the peformance will actually get worse
Or you will have very weird bugs

Review the major players

The web server (thin, puma, other)
Your sinatra app itself
Any gems you use (i.e. ActiveRecord itself)
Database servers

What makes it Thread safe?

Simply put: when threads don’t share mutable state
So, a class method (i.e. def self.do_this_thing) is fine
But a class instance variable (.e. def self do this thing; @a = 100) is asking for trouble

Mini case study



glass CurrentState

  def self.favorite_user_id
     @favorite_user_id
  end

  def self.favorite_user_id= curr
     @favorite_user_id = curr
  end

  def self.favorite_user
     @favorite_user ||= User.find(@favorite_user_id)
  end
end

# And then
def handle_form
  CurrentState.favorite_user_id = some_user_id
#...
  CurrentState.favorite_user.name 
#...
end

A convenient way to globally share the current user
Without regard to thread or to Sinatra dependency
Terrible idea! You are sharing mutable state between threads

Discussion: Would this be a problem if your web server was running just one thread? How about if it was running two processes, one thread each? So how would you safely share the current user when you are multi threaded?

Ruby

None of the core data structures of MRI Ruby are thread safe (except for Queue)
Any ruby expression that generates more than one “ruby operation” (e.g. +=) is not safe!
Only if they are used with thread shared data
You can avoid using all of them and use thread safe equivalents
For example Ruby Concurrency Library
Has thread-safe versions of array, hash, set, map, tuple and others
So: Your ruby code is probably not thread-safe!

What about the GIL?

GIL (Global Interpreter Lock) ensures that your Ruby code doesn’t run in parallel - at the same time
But operating system threads get interrupted (by the clock) at arbitrary times
GIL is what makes MRI itself threadsafe but it doesnt make your code threadsafe

What about Rails (and its key dependency, ActiveRecord)

It is threadsafe!
You need to “just” worry about your own code
And any gems or libraries you use!
How to make your code threadsafe>

Avoid global state

Ruby global variables: $logger or $redis
Ruby class variables and class instance variables (see example above)
Memoization can accidentally get you in trouble (it happened to me)

Session

As a SESSION belongs to one request
Thread safe from the perspective of a multi-threaded web server

Puma

1 or more workers - they are processes
- No thread safety problems
- Each process uses a complete memory space
1 or more threads per worker
- Pay attention to thread safety
- If a “global” value changes out from under you – suspect a race condition