[0.9.1][PD] Added support for delay-free blocks in prefill nodes #1691
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: underfituu <hzhucong@163.com>
Signed-off-by: underfituu <hzhucong@163.com>
Signed-off-by: underfituu <hzhucong@163.com>
underfituu
force-pushed
the
add_kv_transfer_finished_status
branch
from
77c41bd to
a8a3954
Compare
underfituu
force-pushed
the
add_kv_transfer_finished_status
branch
from
259ff3c to
2d2fc36
Compare
underfituu
changed the title
Signed-off-by: underfituu <hzhucong@163.com>
Signed-off-by: underfituu <hzhucong@163.com>
| else: | ||
| raise RuntimeError( | ||
| f"LLMDataDistCMgrConnectorWorker: Receiving unexpected request event {event_msg} from remote !" | ||
| ) | ||
|
|
||
| def _increment_task_count(self, request_id: str, tp_rank: int, | ||
| decode_tp_size: int): | ||
| if tp_rank in self.done_receiving_counts[request_id]: |
There was a problem hiding this comment.
Better check if the request_id is already inside the self.done_receiving_counts for safety
There was a problem hiding this comment.
where did you add your request_id ? seems I can't find it in this diff
There was a problem hiding this comment.
Thanks! Added explicit request_id existence check and early initialization for safety.
Signed-off-by: underfituu <hzhucong@163.com>
ganyi1996ppo
approved these changes
Jul 14, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What this PR does / why we need it?
PD Logic Analysis:
In the current implementation, the P-node immediately releases memory blocks after completing inference. Under high concurrency scenarios, if the P-node's inference speed significantly outpaces the D-node's block pulling operations, this leads to memory block contention/corruption issues.
Current Solution:
The D-node sends acknowledgment messages to the worker connector in the P-node's driver worker after completing data reception. The P-node maintains a counter to track these acknowledgments - memory blocks are only released after receiving confirmations from all D-node worker connectors involved in the KV cache transfer.
Does this PR introduce any user-facing change?
How was this patch tested?