mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-10-07 01:22:59 +08:00
fix bug for trigger preempted (#4228)
This commit is contained in:
@@ -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:
|
||||||
|
@@ -387,14 +387,20 @@ 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']}"
|
||||||
)
|
)
|
||||||
cache_info = {
|
if tasks[i].get("error_msg", None) is not None:
|
||||||
"request_id": tasks[i].request_id,
|
cache_info = {
|
||||||
"device_ids": self.cfg.device_ids.split(","),
|
"request_id": tasks[i].request_id,
|
||||||
"ip": self.cfg.host_ip,
|
"error_msg": tasks[i].get("error_msg"),
|
||||||
"rdma_ports": self.cfg.disaggregate_info["cache_info"]["rdma"]["rdma_port"],
|
}
|
||||||
"transfer_protocol": "rdma",
|
else:
|
||||||
"dest_block_ids": tasks[i].disaggregate_info["block_tables"],
|
cache_info = {
|
||||||
}
|
"request_id": tasks[i].request_id,
|
||||||
|
"device_ids": self.cfg.device_ids.split(","),
|
||||||
|
"ip": self.cfg.host_ip,
|
||||||
|
"rdma_ports": self.cfg.disaggregate_info["cache_info"]["rdma"]["rdma_port"],
|
||||||
|
"transfer_protocol": "rdma",
|
||||||
|
"dest_block_ids": tasks[i].disaggregate_info["block_tables"],
|
||||||
|
}
|
||||||
if addr not in temp_cache_info:
|
if addr not in temp_cache_info:
|
||||||
temp_cache_info[addr] = []
|
temp_cache_info[addr] = []
|
||||||
|
|
||||||
|
Reference in New Issue
Block a user