Since 0.3.5

this is young feature - functionality could be dramatically enriched in the future.

By default, queues stored in a directory-based style. Each element of queue pipes directly from incoming requests, as well as to lambda without caching. It means - RAM usage is almost constant regardless of request sizes and a number of elements in a queue.

Security restrictions (policies) checked twice: on append time and before lambda execution in the same way as it defined in security.

Queues that bound to the lambda could be found in Overview -> Endpoint page.

A queue can be re-assigned to another lambda without destroying it.

In case of failure, the task will be re-tried after a defined interval with a limited number of attempts. 0 retry means no additional attempts - at least once the task will be processed. After failure, a queue worker will wait the required time, and it will not process other tasks.

After lambda removal, linked queues are also will be automatically removed.

Designed for

  • to provide async processing for long-running tasks;

NOT designed for

  • load balancing (but possible using multiple queues);

Endpoint: /q/:queue-name

Allowed queue name: latin letters, numbers, dash, and 3 to 64 symbols length.

One queue always linked to one lambda, but one lambda can be linked to multiple queues.