o-concurrent-ruby 1.1.11

data/lib/concurrent-ruby/concurrent/atomic/reentrant_read_write_lock.rb

@@ -0,0 +1,377 @@
+require 'thread'
+require 'concurrent/atomic/atomic_reference'
+require 'concurrent/errors'
+require 'concurrent/synchronization'
+require 'concurrent/atomic/thread_local_var'
+
+module Concurrent
+
+  # Re-entrant read-write lock implementation
+  #
+  # Allows any number of concurrent readers, but only one concurrent writer
+  # (And while the "write" lock is taken, no read locks can be obtained either.
+  # Hence, the write lock can also be called an "exclusive" lock.)
+  #
+  # If another thread has taken a read lock, any thread which wants a write lock
+  # will block until all the readers release their locks. However, once a thread
+  # starts waiting to obtain a write lock, any additional readers that come along
+  # will also wait (so writers are not starved).
+  #
+  # A thread can acquire both a read and write lock at the same time. A thread can
+  # also acquire a read lock OR a write lock more than once. Only when the read (or
+  # write) lock is released as many times as it was acquired, will the thread
+  # actually let it go, allowing other threads which might have been waiting
+  # to proceed. Therefore the lock can be upgraded by first acquiring
+  # read lock and then write lock and that the lock can be downgraded by first
+  # having both read and write lock a releasing just the write lock.
+  #
+  # If both read and write locks are acquired by the same thread, it is not strictly
+  # necessary to release them in the same order they were acquired. In other words,
+  # the following code is legal:
+  #
+  # @example
+  #   lock = Concurrent::ReentrantReadWriteLock.new
+  #   lock.acquire_write_lock
+  #   lock.acquire_read_lock
+  #   lock.release_write_lock
+  #   # At this point, the current thread is holding only a read lock, not a write
+  #   # lock. So other threads can take read locks, but not a write lock.
+  #   lock.release_read_lock
+  #   # Now the current thread is not holding either a read or write lock, so
+  #   # another thread could potentially acquire a write lock.
+  #
+  # This implementation was inspired by `java.util.concurrent.ReentrantReadWriteLock`.
+  #
+  # @example
+  #   lock = Concurrent::ReentrantReadWriteLock.new
+  #   lock.with_read_lock  { data.retrieve }
+  #   lock.with_write_lock { data.modify! }
+  #
+  # @see http://docs.oracle.com/javase/7/docs/api/java/util/concurrent/locks/ReentrantReadWriteLock.html java.util.concurrent.ReentrantReadWriteLock
+  class ReentrantReadWriteLock < Synchronization::Object
+
+    # Implementation notes:
+    #
+    # A goal is to make the uncontended path for both readers/writers mutex-free
+    # Only if there is reader-writer or writer-writer contention, should mutexes be used
+    # Otherwise, a single CAS operation is all we need to acquire/release a lock
+    #
+    # Internal state is represented by a single integer ("counter"), and updated
+    #   using atomic compare-and-swap operations
+    # When the counter is 0, the lock is free
+    # Each thread which has one OR MORE read locks increments the counter by 1
+    #   (and decrements by 1 when releasing the read lock)
+    # The counter is increased by (1 << 15) for each writer waiting to acquire the
+    #   write lock, and by (1 << 29) if the write lock is taken
+    #
+    # Additionally, each thread uses a thread-local variable to count how many times
+    #   it has acquired a read lock, AND how many times it has acquired a write lock.
+    # It uses a similar trick; an increment of 1 means a read lock was taken, and
+    #   an increment of (1 << 15) means a write lock was taken
+    # This is what makes re-entrancy possible
+    #
+    # 2 rules are followed to ensure good liveness properties:
+    # 1) Once a writer has queued up and is waiting for a write lock, no other thread
+    #    can take a lock without waiting
+    # 2) When a write lock is released, readers are given the "first chance" to wake
+    #    up and acquire a read lock
+    # Following these rules means readers and writers tend to "take turns", so neither
+    #   can starve the other, even under heavy contention
+
+    # @!visibility private
+    READER_BITS    = 15
+    # @!visibility private
+    WRITER_BITS    = 14
+
+    # Used with @Counter:
+    # @!visibility private
+    WAITING_WRITER = 1 << READER_BITS
+    # @!visibility private
+    RUNNING_WRITER = 1 << (READER_BITS + WRITER_BITS)
+    # @!visibility private
+    MAX_READERS    = WAITING_WRITER - 1
+    # @!visibility private
+    MAX_WRITERS    = RUNNING_WRITER - MAX_READERS - 1
+
+    # Used with @HeldCount:
+    # @!visibility private
+    WRITE_LOCK_HELD = 1 << READER_BITS
+    # @!visibility private
+    READ_LOCK_MASK  = WRITE_LOCK_HELD - 1
+    # @!visibility private
+    WRITE_LOCK_MASK = MAX_WRITERS
+
+    safe_initialization!
+
+    # Create a new `ReentrantReadWriteLock` in the unlocked state.
+    def initialize
+      super()
+      @Counter    = AtomicFixnum.new(0)       # single integer which represents lock state
+      @ReadQueue  = Synchronization::Lock.new # used to queue waiting readers
+      @WriteQueue = Synchronization::Lock.new # used to queue waiting writers
+      @HeldCount  = ThreadLocalVar.new(0)     # indicates # of R & W locks held by this thread
+    end
+
+    # Execute a block operation within a read lock.
+    #
+    # @yield the task to be performed within the lock.
+    #
+    # @return [Object] the result of the block operation.
+    #
+    # @raise [ArgumentError] when no block is given.
+    # @raise [Concurrent::ResourceLimitError] if the maximum number of readers
+    #   is exceeded.
+    def with_read_lock
+      raise ArgumentError.new('no block given') unless block_given?
+      acquire_read_lock
+      begin
+        yield
+      ensure
+        release_read_lock
+      end
+    end
+
+    # Execute a block operation within a write lock.
+    #
+    # @yield the task to be performed within the lock.
+    #
+    # @return [Object] the result of the block operation.
+    #
+    # @raise [ArgumentError] when no block is given.
+    # @raise [Concurrent::ResourceLimitError] if the maximum number of readers
+    #   is exceeded.
+    def with_write_lock
+      raise ArgumentError.new('no block given') unless block_given?
+      acquire_write_lock
+      begin
+        yield
+      ensure
+        release_write_lock
+      end
+    end
+
+    # Acquire a read lock. If a write lock is held by another thread, will block
+    # until it is released.
+    #
+    # @return [Boolean] true if the lock is successfully acquired
+    #
+    # @raise [Concurrent::ResourceLimitError] if the maximum number of readers
+    #   is exceeded.
+    def acquire_read_lock
+      if (held = @HeldCount.value) > 0
+        # If we already have a lock, there's no need to wait
+        if held & READ_LOCK_MASK == 0
+          # But we do need to update the counter, if we were holding a write
+          #   lock but not a read lock
+          @Counter.update { |c| c + 1 }
+        end
+        @HeldCount.value = held + 1
+        return true
+      end
+
+      while true
+        c = @Counter.value
+        raise ResourceLimitError.new('Too many reader threads') if max_readers?(c)
+
+        # If a writer is waiting OR running when we first queue up, we need to wait
+        if waiting_or_running_writer?(c)
+          # Before going to sleep, check again with the ReadQueue mutex held
+          @ReadQueue.synchronize do
+            @ReadQueue.ns_wait if waiting_or_running_writer?
+          end
+          # Note: the above 'synchronize' block could have used #wait_until,
+          #   but that waits repeatedly in a loop, checking the wait condition
+          #   each time it wakes up (to protect against spurious wakeups)
+          # But we are already in a loop, which is only broken when we successfully
+          #   acquire the lock! So we don't care about spurious wakeups, and would
+          #   rather not pay the extra overhead of using #wait_until
+
+          # After a reader has waited once, they are allowed to "barge" ahead of waiting writers
+          # But if a writer is *running*, the reader still needs to wait (naturally)
+          while true
+            c = @Counter.value
+            if running_writer?(c)
+              @ReadQueue.synchronize do
+                @ReadQueue.ns_wait if running_writer?
+              end
+            elsif @Counter.compare_and_set(c, c+1)
+              @HeldCount.value = held + 1
+              return true
+            end
+          end
+        elsif @Counter.compare_and_set(c, c+1)
+          @HeldCount.value = held + 1
+          return true
+        end
+      end
+    end
+
+    # Try to acquire a read lock and return true if we succeed. If it cannot be
+    # acquired immediately, return false.
+    #
+    # @return [Boolean] true if the lock is successfully acquired
+    def try_read_lock
+      if (held = @HeldCount.value) > 0
+        if held & READ_LOCK_MASK == 0
+          # If we hold a write lock, but not a read lock...
+          @Counter.update { |c| c + 1 }
+        end
+        @HeldCount.value = held + 1
+        return true
+      else
+        c = @Counter.value
+        if !waiting_or_running_writer?(c) && @Counter.compare_and_set(c, c+1)
+          @HeldCount.value = held + 1
+          return true
+        end
+      end
+      false
+    end
+
+    # Release a previously acquired read lock.
+    #
+    # @return [Boolean] true if the lock is successfully released
+    def release_read_lock
+      held = @HeldCount.value = @HeldCount.value - 1
+      rlocks_held = held & READ_LOCK_MASK
+      if rlocks_held == 0
+        c = @Counter.update { |counter| counter - 1 }
+        # If one or more writers were waiting, and we were the last reader, wake a writer up
+        if waiting_or_running_writer?(c) && running_readers(c) == 0
+          @WriteQueue.signal
+        end
+      elsif rlocks_held == READ_LOCK_MASK
+        raise IllegalOperationError, "Cannot release a read lock which is not held"
+      end
+      true
+    end
+
+    # Acquire a write lock. Will block and wait for all active readers and writers.
+    #
+    # @return [Boolean] true if the lock is successfully acquired
+    #
+    # @raise [Concurrent::ResourceLimitError] if the maximum number of writers
+    #   is exceeded.
+    def acquire_write_lock
+      if (held = @HeldCount.value) >= WRITE_LOCK_HELD
+        # if we already have a write (exclusive) lock, there's no need to wait
+        @HeldCount.value = held + WRITE_LOCK_HELD
+        return true
+      end
+
+      while true
+        c = @Counter.value
+        raise ResourceLimitError.new('Too many writer threads') if max_writers?(c)
+
+        # To go ahead and take the lock without waiting, there must be no writer
+        #   running right now, AND no writers who came before us still waiting to
+        #   acquire the lock
+        # Additionally, if any read locks have been taken, we must hold all of them
+        if held > 0 && @Counter.compare_and_set(1, c+RUNNING_WRITER)
+          # If we are the only one reader and successfully swap the RUNNING_WRITER bit on, then we can go ahead
+          @HeldCount.value = held + WRITE_LOCK_HELD
+          return true
+        elsif @Counter.compare_and_set(c, c+WAITING_WRITER)
+          while true
+            # Now we have successfully incremented, so no more readers will be able to increment
+            #   (they will wait instead)
+            # However, readers OR writers could decrement right here
+            @WriteQueue.synchronize do
+              # So we have to do another check inside the synchronized section
+              # If a writer OR another reader is running, then go to sleep
+              c = @Counter.value
+              @WriteQueue.ns_wait if running_writer?(c) || running_readers(c) != held
+            end
+            # Note: if you are thinking of replacing the above 'synchronize' block
+            # with #wait_until, read the comment in #acquire_read_lock first!
+
+            # We just came out of a wait
+            # If we successfully turn the RUNNING_WRITER bit on with an atomic swap,
+            #   then we are OK to stop waiting and go ahead
+            # Otherwise go back and wait again
+            c = @Counter.value
+            if !running_writer?(c) &&
+               running_readers(c) == held &&
+               @Counter.compare_and_set(c, c+RUNNING_WRITER-WAITING_WRITER)
+              @HeldCount.value = held + WRITE_LOCK_HELD
+              return true
+            end
+          end
+        end
+      end
+    end
+
+    # Try to acquire a write lock and return true if we succeed. If it cannot be
+    # acquired immediately, return false.
+    #
+    # @return [Boolean] true if the lock is successfully acquired
+    def try_write_lock
+      if (held = @HeldCount.value) >= WRITE_LOCK_HELD
+        @HeldCount.value = held + WRITE_LOCK_HELD
+        return true
+      else
+        c = @Counter.value
+        if !waiting_or_running_writer?(c) &&
+           running_readers(c) == held &&
+           @Counter.compare_and_set(c, c+RUNNING_WRITER)
+           @HeldCount.value = held + WRITE_LOCK_HELD
+          return true
+        end
+      end
+      false
+    end
+
+    # Release a previously acquired write lock.
+    #
+    # @return [Boolean] true if the lock is successfully released
+    def release_write_lock
+      held = @HeldCount.value = @HeldCount.value - WRITE_LOCK_HELD
+      wlocks_held = held & WRITE_LOCK_MASK
+      if wlocks_held == 0
+        c = @Counter.update { |counter| counter - RUNNING_WRITER }
+        @ReadQueue.broadcast
+        @WriteQueue.signal if waiting_writers(c) > 0
+      elsif wlocks_held == WRITE_LOCK_MASK
+        raise IllegalOperationError, "Cannot release a write lock which is not held"
+      end
+      true
+    end
+
+    private
+
+    # @!visibility private
+    def running_readers(c = @Counter.value)
+      c & MAX_READERS
+    end
+
+    # @!visibility private
+    def running_readers?(c = @Counter.value)
+      (c & MAX_READERS) > 0
+    end
+
+    # @!visibility private
+    def running_writer?(c = @Counter.value)
+      c >= RUNNING_WRITER
+    end
+
+    # @!visibility private
+    def waiting_writers(c = @Counter.value)
+      (c & MAX_WRITERS) >> READER_BITS
+    end
+
+    # @!visibility private
+    def waiting_or_running_writer?(c = @Counter.value)
+      c >= WAITING_WRITER
+    end
+
+    # @!visibility private
+    def max_readers?(c = @Counter.value)
+      (c & MAX_READERS) == MAX_READERS
+    end
+
+    # @!visibility private
+    def max_writers?(c = @Counter.value)
+      (c & MAX_WRITERS) == MAX_WRITERS
+    end
+  end
+end

data/lib/concurrent-ruby/concurrent/atomic/ruby_thread_local_var.rb

@@ -0,0 +1,181 @@
+require 'thread'
+require 'concurrent/atomic/abstract_thread_local_var'
+
+module Concurrent
+
+  # @!visibility private
+  # @!macro internal_implementation_note
+  class RubyThreadLocalVar < AbstractThreadLocalVar
+
+    # Each thread has a (lazily initialized) array of thread-local variable values
+    # Each time a new thread-local var is created, we allocate an "index" for it
+    # For example, if the allocated index is 1, that means slot #1 in EVERY
+    #   thread's thread-local array will be used for the value of that TLV
+    #
+    # The good thing about using a per-THREAD structure to hold values, rather
+    #   than a per-TLV structure, is that no synchronization is needed when
+    #    reading and writing those values (since the structure is only ever
+    #    accessed by a single thread)
+    #
+    # Of course, when a TLV is GC'd, 1) we need to recover its index for use
+    #   by other new TLVs (otherwise the thread-local arrays could get bigger
+    #   and bigger with time), and 2) we need to null out all the references
+    #   held in the now-unused slots (both to avoid blocking GC of those objects,
+    #   and also to prevent "stale" values from being passed on to a new TLV
+    #   when the index is reused)
+    # Because we need to null out freed slots, we need to keep references to
+    #   ALL the thread-local arrays -- ARRAYS is for that
+    # But when a Thread is GC'd, we need to drop the reference to its thread-local
+    #   array, so we don't leak memory
+
+    FREE                = []
+    LOCK                = Mutex.new
+    THREAD_LOCAL_ARRAYS = {} # used as a hash set
+
+    # synchronize when not on MRI
+    # on MRI using lock in finalizer leads to "can't be called from trap context" error
+    # so the code is carefully written to be tread-safe on MRI relying on GIL
+
+    if Concurrent.on_cruby?
+      # @!visibility private
+      def self.semi_sync(&block)
+        block.call
+      end
+    else
+      # @!visibility private
+      def self.semi_sync(&block)
+        LOCK.synchronize(&block)
+      end
+    end
+
+    private_constant :FREE, :LOCK, :THREAD_LOCAL_ARRAYS
+
+    # @!macro thread_local_var_method_get
+    def value
+      if (array = get_threadlocal_array)
+        value = array[@index]
+        if value.nil?
+          default
+        elsif value.equal?(NULL)
+          nil
+        else
+          value
+        end
+      else
+        default
+      end
+    end
+
+    # @!macro thread_local_var_method_set
+    def value=(value)
+      me = Thread.current
+      # We could keep the thread-local arrays in a hash, keyed by Thread
+      # But why? That would require locking
+      # Using Ruby's built-in thread-local storage is faster
+      unless (array = get_threadlocal_array(me))
+        array = set_threadlocal_array([], me)
+        self.class.semi_sync { THREAD_LOCAL_ARRAYS[array.object_id] = array }
+        ObjectSpace.define_finalizer(me, self.class.thread_finalizer(array.object_id))
+      end
+      array[@index] = (value.nil? ? NULL : value)
+      value
+    end
+
+    protected
+
+    # @!visibility private
+    def allocate_storage
+      @index = FREE.pop || next_index
+
+      ObjectSpace.define_finalizer(self, self.class.thread_local_finalizer(@index))
+    end
+
+    # @!visibility private
+    def self.thread_local_finalizer(index)
+      proc do
+        semi_sync do
+          # The cost of GC'ing a TLV is linear in the number of threads using TLVs
+          # But that is natural! More threads means more storage is used per TLV
+          # So naturally more CPU time is required to free more storage
+          #
+          # DO NOT use each_value which might conflict with new pair assignment
+          # into the hash in #value= method
+          THREAD_LOCAL_ARRAYS.values.each { |array| array[index] = nil }
+          # free index has to be published after the arrays are cleared
+          FREE.push(index)
+        end
+      end
+    end
+
+    # @!visibility private
+    def self.thread_finalizer(id)
+      proc do
+        semi_sync do
+          # The thread which used this thread-local array is now gone
+          # So don't hold onto a reference to the array (thus blocking GC)
+          THREAD_LOCAL_ARRAYS.delete(id)
+        end
+      end
+    end
+
+    private
+
+    # noinspection RubyClassVariableUsageInspection
+    @@next = 0
+    # noinspection RubyClassVariableUsageInspection
+    def next_index
+      LOCK.synchronize do
+        result = @@next
+        @@next += 1
+        result
+      end
+    end
+
+    if Thread.instance_methods.include?(:thread_variable_get)
+
+      def get_threadlocal_array(thread = Thread.current)
+        thread.thread_variable_get(:__threadlocal_array__)
+      end
+
+      def set_threadlocal_array(array, thread = Thread.current)
+        thread.thread_variable_set(:__threadlocal_array__, array)
+      end
+
+    else
+
+      def get_threadlocal_array(thread = Thread.current)
+        thread[:__threadlocal_array__]
+      end
+
+      def set_threadlocal_array(array, thread = Thread.current)
+        thread[:__threadlocal_array__] = array
+      end
+    end
+
+    # This exists only for use in testing
+    # @!visibility private
+    def value_for(thread)
+      if (array = get_threadlocal_array(thread))
+        value = array[@index]
+        if value.nil?
+          get_default
+        elsif value.equal?(NULL)
+          nil
+        else
+          value
+        end
+      else
+        get_default
+      end
+    end
+
+    # @!visibility private
+    def get_default
+      if @default_block
+        raise "Cannot use default_for with default block"
+      else
+        @default
+      end
+    end
+  end
+end