[Sliding Window Attention] BlockManagerV2 allocate with SWA

[yangsijia-serena]

Hi there, I'm new to vllm and I may have missed something, but in BlockManagerV2, I only see consideration of the sliding window in the can_allocate function, like the following code snippet:

def can_allocate(self, seq_group: SequenceGroup) -> AllocStatus:
            # other code ...
              if self.max_block_sliding_window is not None:
                 num_required_blocks = min(num_required_blocks, self.max_block_sliding_window)
            # other code ...

But I don't see any consideration of the sliding window when actually performing the allocation. Is this by design or a potential bug? If it's by design, I'm wondering about a scenario where the entire prompt requires 4 blocks, but the number of free blocks is only 3. In this case, if max_block_sliding_window=3, the can_allocate function would return True, but when it comes to the actual allocation, there wouldn't be enough space for the tokens in the 4th block. Is this a known issue or something that is handled differently?