[PD Disaggregation] Add timestamp for analyzing splitwise deployment (#5317)

* Add timestamp for analyzing splitwise deployment

* up

* up

* up

* up

* up

* up

* fix format

* fix
This commit is contained in:
Juncai
2025-12-08 10:08:44 +08:00
committed by GitHub
parent 0c66163dfd
commit 80efe98f8d
21 changed files with 287 additions and 188 deletions

View File

@@ -302,7 +302,7 @@ class ResultReader:
add a req to reader, reader will async fetch infer result from redis
"""
with self.lock:
self.reqs[req.request_id] = {"arrival_time": req.arrival_time}
self.reqs[req.request_id] = {"arrival_time": req.metrics.arrival_time}
self.out_buffer[req.request_id] = []
def read(self):
@@ -834,7 +834,7 @@ class InferScheduler:
break
req = self.reqs_queue.popleft()
if cur_time - req.arrival_time > self.ttl:
if cur_time - req.metrics.arrival_time > self.ttl:
logger.error(f"req({req.request_id}) is expired({self.ttl}) when InferScheduler Get Requests")
self.node.finish_req(req.request_id)
continue