Play project with code examples to some toy problems to refresh my knowledge of Java Concurrency API.

NOTE: Some examples contain idioms not good for production code but are here just to remind me of oddities / gotchas / things to remember.

How to Generate Random Numbers in Concurrent Applications

In a concurrent application if different threads generate random number using Random class e.g. by doing Random.nextInt(10) then it can cause heavy contention on the seed and poor performance.

Simulate Producer Consumer Problem

Variations:

• Producer Consumer using synchronized - wait
• Using BlockedQueue
• Using Lock

Simulate ReaderWriter Problem

3 Variations:

1. Unbiased - Ensure that the reads and writes are in FIFO order and fair

2. Writer Biased - Once a writer requests, no reads are entertained.

3. Reader Biased - If a reader already has lock, another reader can continue to read even if a Writer is waiting.

Semaphore

• Semaphores and concept of lock ownership
• Semaphores like a CountdownLatch
• Controlling access to a pool of resources using Semaphores

CountDownLatch

• Demonstrate that countdown is just a count and not necessarily number of threads which counted down.

Executors

• CachedThreadPool (ThreadPoolExecutor)
• FixedSizeThreadPool (ThreadPoolExecutor)
• SingleThreadExecutor (Executor)
• ScheduledThreadpoolExecutors (ThreadPoolExecutor)
  • Running Tasks periodically at a fixed rate
  • Running Task Only once after an initial delay
  • Running tasks periodically with a fixed delay between 2 instances
• Running tasks which don't return results
• Running tasks which return result
  • Running multiple tasks and processing first result
  • Running multiple tasks and processing all results
• FutureTask - How to execute another function post the task is complete.

AtomicLong / AtomicInteger

• How to use CAS for better performance than locks

AtomicArrays

• Using CAS operations on an array of atomic variables for better performance

LongAdder

• LongAdder as maintaining total
• LongAdder as maintaining counter in stats

In case of maintaining total/counter, LongAdder results in better performance than AtomicLong under heavy contention.

Note: getting result is racy

LongAccumulators

In case of maintaining stats, this results in better performance than AtomicLong under heavy contention.

Exchanger

• Play ping pong between two threads using Exchanger

CyclicBarrier

• Divide and Conquer using CyclicBarrier.
• Demonstrate that the barrier is automatically reset.
• Demonstrate that CyclicBarrier takes into account the number of threads and not the number of invocations

CompletionService

Separating concerns of submitting tasks and processing results

Phaser

• Sample program to illustrate register(), arriveAndAwaitAdvance(), and arriveAndDeregister()
• Simulate a test where all students finish exercises together. Print messages whenever phase changes. Simulate a late student joining the exam.

ForkJoinPool

1. - Understanding ForkJoinTask, RecursiveAction, RecursiveTask
- pool.invoke() vs. pool.execute() vs pool.submit()
2. Divide and Conquer using ForkJoinPool
- Using ForkJoinTask.invokeAll() [ Synchronous but better reuse of threads] - Grouping results of sub-tasks
3. Asynchronous methods - ForkJoinTask.fork() and ForkJoinTask.join()

TODOs:

How to handle uncaught exceptions in the run() method of threads? ThreadGroups ThreadFactory

CloseableFuture

StampedLocks

TransferQueue

1. Show an example of deadlock.

2. Most textbook says to prevent deadlock, should take locks in the same order. Is it really true? Can you implement a situation where this leads to a "deadlock"

Hint: effect is same but it is called something else (not livelock :)