I will elaborate further through Java API and technical blog introductions in the future.

Asynchronous

What is Asynchronous?

Processing tasks not in sequential order.

When is it used?

Used for performance improvement.

  • When there are I/O operations
  • When there are CPU bounded jobs (batch jobs)

Considerations when writing asynchronous programs

  • How to use the results of asynchronous tasks?
    • callback
      • Pass a function (routine) to perform additional work upon completion.
    • blocking
      • Wait until all are completed
    • using buffer
      • Able to use data even if I/O is not finished
  • What state are the asynchronous tasks in?
    • Responding according to state
      • If all are completed -> process
      • If in progress -> wait or proceed to the next process
      • If an error occurred -> fail the entire process, etc.
  • How to handle asynchronous I/O tasks?
    • non-blocking (event loop method)
    • blocking (request I/O to Thread in the usual way)
  • When splitting multiple jobs, how to handle completion or failure
    • If all are completed, success
      • How to determine if all succeeded?
    • If even one fails, all fail (fail fast)
      • How to determine if one failed?
    • If even one is completed, treat as success
      • How to determine?
    • Process in the order of completion
  • Concurrency
    • Since the purpose of asynchronous tasks is performance improvement, multiple threads may be used. In this case, concurrency issues must be resolved when accessing the same resource.
    • volatile
      • Ensures that the value referenced is from main memory, not cached in the CPU register (to resolve shared resource inconsistency issues)
    • CompareAndSwap
      • Ensures atomicity of operations at the CPU level for Check Then Act. Only one thread can perform the operation.
      • Changes the value of a variable only if it contains the expected value.
      • Used instead of the synchronization keyword (lock). If a thread is blocked, it cannot immediately wake up from the blocked state.
    • ForkJoinPool
      • Work-stealing technique. Threads can steal jobs assigned to other threads.
      • Tasks cannot be perfectly evenly distributed among threads. Therefore, idle threads can take tasks from busy threads.

Asynchronous processing at the system level?

  • Use a message queue
  • The producer sends a message about the work to the queue, and the consumer reads the message and processes the work
  • Message queues are also used to decouple dependencies in event-driven design
  • Transaction unit
    • Group up to message publishing as a single transaction.
    • Whether the asynchronous task is completed is not of concern.
  • You can think of the message as a task. If the consumer fails to process due to an error, re-publishing is necessary.