The queue system allows placing items in a queue and processing them later. The system tries to ensure that only one consumer can process an item.
Before a queue can be used it needs to be created by DrupalQueueInterface::createQueue().
Items can be added to the queue by passing an arbitrary data object to DrupalQueueInterface::createItem().
To process an item, call DrupalQueueInterface::claimItem() and specify how long you want to have a lease for working on that item. When finished processing, the item needs to be deleted by calling DrupalQueueInterface::deleteItem(). If the consumer dies, the item will be made available again by the DrapalQueueInterface implementation once the lease expires. Another consumer will then be able to receive it when calling DrupalQueueInterface::claimItem().
The $item object used by the DrupalQueueInterface can contain arbitrary metadata depending on the implementation. Systems using the interface should only rely on the data property which will contain the information passed to DrupalQueueInterface::createItem(). The full queue item returned by DrupalQueueInterface::createItem() needs to be passed to DrupalQueueInterface::deleteItem() once processing is completed.
While the queue system makes a best effort to preserve order in messages, due to the pluggable nature of the queue, there is no guarantee that items will be delivered on claim in the order they were sent. For example, some implementations like beanstalkd or others with distributed back-ends like Amazon SQS will be managing jobs for a large set of producers and consumers where a strict FIFO ordering will likely not be preserved.
The system also makes no guarantees about a task only being executed once: callers that have non-idempotent tasks either need to live with the possiblity of the task being invoked multiple times in cases where a claim lease expires, or need to implement their own transactions to make their tasks idempotent.
et al. Can you say 'registry'? Drupal now maintains an internal registry of
all functions or classes in the system, allowing it to lazy-load code files as
needed (reducing the amount of code that must be parsed on each request). The
list of included files is cached per menu callback for subsequent loading by
the menu router. This way, a given page request will have all the code it needs
but little else, minimizing time spent parsing unneeded code.
Angie Byron
Dries Buytaert