Choosing a data structure for a variant of producer consumer problem

后端 未结 2 1552
温柔的废话
温柔的废话 2021-01-13 12:03

Right now, I have a queue, with multiple producers and single consumer.

Consumer thread operation is slow. Also, consumer takes element from queue through a peek ope

相关标签:
2条回答
  • 2021-01-13 12:06

    It sounds like you should really have two queues:

    • Unprocessed
    • In progress

    A consumer would atomically (via a lock) pull from the unprocessed queue and add to the in-progress queue. That way multiple consumers can work concurrently... but the producer can still take a snapshot of both queues when it needs to. When the consumer is finished with a task, it removes it from the in-progress queue. (That doesn't really need to be a queue, as nothing's "pulling" from it as such. Just some collection you can easily add to and remove from.)

    Given that you'll need locking to make the transfer atomic, you probably don't need the underlying queues to be the concurrent ones - you'll be guarding all shared access already.

    0 讨论(0)
  • 2021-01-13 12:12

    I'd agree with Jon Skeet (+1) in that you need two stores for recording waiting and in-progress items. I would use a LinkedBlockingQueue and have each of your consumers call take() on it. When an element arrives on the queue it will be taken by one of the consumers.

    Recording what is in progress and what is completed would be separate operation. I would maintain a HashSet of all the items that have not yet completed and my producer would first (atomically) add the item to the HashSet of non-completed items and then pop the item on the queue. Once a consumer have finished it's work, it removes the item from the HashSet.

    Your producer can scan the HashSet to determine what is outstanding.

    0 讨论(0)
提交回复
热议问题