vllm.v1.core.sched.request_queue ¶
FCFSRequestQueue ¶
Bases: deque[Request], RequestQueue
A first-come-first-served queue that supports deque operations.
Source code in vllm/v1/core/sched/request_queue.py
__iter__ ¶
__reversed__ ¶
prepend_requests ¶
prepend_requests(requests: RequestQueue) -> None
Prepend all requests from another queue to the front of this queue.
remove_requests ¶
Remove multiple specific requests from the queue.
Source code in vllm/v1/core/sched/request_queue.py
PriorityRequestQueue ¶
Bases: RequestQueue
A priority queue that supports heap operations.
Respects the ordering defined in the Request class, where requests with a smaller value of priority are processed first. If multiple requests have the same priority, the one with the earlier arrival_time is processed first.
Source code in vllm/v1/core/sched/request_queue.py
__init__ ¶
__iter__ ¶
__reversed__ ¶
prepend_request ¶
prepend_request(request: Request) -> None
Add a request to the queue according to priority policy.
Note: In a priority queue, there is no concept of prepending to the front. Requests are ordered by (priority, arrival_time).
Source code in vllm/v1/core/sched/request_queue.py
prepend_requests ¶
prepend_requests(requests: RequestQueue) -> None
Add all requests from another queue according to priority policy.
Note: In a priority queue, there is no concept of prepending to the front. Requests are ordered by (priority, arrival_time).
Source code in vllm/v1/core/sched/request_queue.py
remove_requests ¶
Remove multiple specific requests from the queue.
Source code in vllm/v1/core/sched/request_queue.py
RequestQueue ¶
Bases: ABC
Abstract base class for request queues.
Source code in vllm/v1/core/sched/request_queue.py
__iter__ abstractmethod ¶
__reversed__ abstractmethod ¶
prepend_requests abstractmethod ¶
prepend_requests(requests: RequestQueue) -> None
remove_requests abstractmethod ¶
SchedulingPolicy ¶
create_request_queue ¶
create_request_queue(
policy: SchedulingPolicy,
) -> RequestQueue
Create request queue based on scheduling policy.