[RL] add version to the key of cache storage && refine raising error (#6160)

* Waiting for cache transfer manager inited

* up

* up

* up

* up

* up

* fix according comments

* fix unittest

* fix

* fix unittest

* fix error

* pass storage_backend to worker
This commit is contained in:
jc
2026-01-27 10:47:46 +08:00
committed by GitHub
parent 7ffa88bb01
commit b1698a79cb
8 changed files with 121 additions and 107 deletions
+13
View File
@@ -328,6 +328,17 @@ class EngineService:
create=True,
)
cache_transfer_inited_signal_data = np.zeros(
shape=[self.cfg.parallel_config.tensor_parallel_size], dtype=np.int32
)
self.cache_transfer_inited_signal = IPCSignal(
name="cache_transfer_inited_signal",
array=cache_transfer_inited_signal_data,
dtype=np.int32,
suffix=current_suffix,
create=True,
)
model_weights_status = np.zeros([1], dtype=np.int32)
self.model_weights_status_signal = IPCSignal(
name="model_weights_status",
@@ -1750,6 +1761,7 @@ class EngineService:
self.worker_healthy_live_signal.clear()
self.cache_ready_signal.clear()
self.swap_space_ready_signal.clear()
self.cache_transfer_inited_signal.clear()
self.exist_prefill_task_signal.clear()
self.model_weights_status_signal.clear()
self.prefix_tree_status_signal.clear()
@@ -1983,6 +1995,7 @@ class EngineService:
worker_default_none_flag = {
"num_gpu_blocks_override": self.cfg.cache_config.num_gpu_blocks_override,
"kvcache_storage_backend": self.cfg.cache_config.kvcache_storage_backend,
}
for worker_flag, value in worker_default_none_flag.items():
if value: