fix bug for trigger preempted (#4228)

This commit is contained in:
chenjian
2025-09-23 20:34:51 +08:00
committed by GitHub
parent acecd5bebe
commit 12043fc476
2 changed files with 16 additions and 9 deletions

View File

@@ -77,6 +77,7 @@ class EngineService:
self.llm_logger = llm_logger self.llm_logger = llm_logger
self.scheduler = cfg.scheduler_config.scheduler() self.scheduler = cfg.scheduler_config.scheduler()
self.enable_decode_cache_task = envs.FD_ENABLE_CACHE_TASK == "1"
if envs.ENABLE_V1_KVCACHE_SCHEDULER: if envs.ENABLE_V1_KVCACHE_SCHEDULER:
self.resource_manager = ResourceManagerV1( self.resource_manager = ResourceManagerV1(
@@ -623,7 +624,7 @@ class EngineService:
for tmp_task in need_delete_tasks: for tmp_task in need_delete_tasks:
tasks.remove(tmp_task) tasks.remove(tmp_task)
# release resource in P # release resource in P
self.resource_manager.prerelease_resource(task) self.resource_manager.prerelease_resource(tmp_task)
if self.cfg.scheduler_config.splitwise_role == "prefill": if self.cfg.scheduler_config.splitwise_role == "prefill":
# to send cache info to cache messager # to send cache info to cache messager
if tasks: if tasks:

View File

@@ -387,6 +387,12 @@ class SplitwiseConnector:
f"{tasks[i].disaggregate_info['cache_info']['rdma']['ip']}:" f"{tasks[i].disaggregate_info['cache_info']['rdma']['ip']}:"
+ f"{tasks[i].disaggregate_info['cache_info']['rdma']['port']}" + f"{tasks[i].disaggregate_info['cache_info']['rdma']['port']}"
) )
if tasks[i].get("error_msg", None) is not None:
cache_info = {
"request_id": tasks[i].request_id,
"error_msg": tasks[i].get("error_msg"),
}
else:
cache_info = { cache_info = {
"request_id": tasks[i].request_id, "request_id": tasks[i].request_id,
"device_ids": self.cfg.device_ids.split(","), "device_ids": self.cfg.device_ids.split(","),