Refa: asyncio.to_thread to ThreadPoolExecutor to break thread limitat… (#12716)

### Type of change

- [x] Refactoring

rag/flow/parser/parser.py

@@ -40,6 +40,10 @@ from rag.llm.cv_model import Base as VLM
 from rag.utils.base64_image import image2id
 
 
+
+
+from common.misc_utils import thread_pool_exec
+
 class ParserParam(ProcessParamBase):
     def __init__(self):
         super().__init__()
@@ -845,7 +849,7 @@ class Parser(ProcessBase):
         for p_type, conf in self._param.setups.items():
             if from_upstream.name.split(".")[-1].lower() not in conf.get("suffix", []):
                 continue
-            await asyncio.to_thread(function_map[p_type], name, blob)
+            await thread_pool_exec(function_map[p_type], name, blob)
             done = True
             break
 

rag/flow/tokenizer/tokenizer.py

@@ -12,7 +12,6 @@
 #  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
-import asyncio
 import logging
 import random
 import re
@@ -31,6 +30,7 @@ from common import settings
 from rag.svr.task_executor import embed_limiter
 from common.token_utils import truncate
 
+from common.misc_utils import thread_pool_exec
 
 class TokenizerParam(ProcessParamBase):
     def __init__(self):
@@ -84,7 +84,7 @@ class Tokenizer(ProcessBase):
         cnts_ = np.array([])
         for i in range(0, len(texts), settings.EMBEDDING_BATCH_SIZE):
             async with embed_limiter:
-                vts, c = await asyncio.to_thread(batch_encode,texts[i : i + settings.EMBEDDING_BATCH_SIZE],)
+                vts, c = await thread_pool_exec(batch_encode,texts[i : i + settings.EMBEDDING_BATCH_SIZE],)
             if len(cnts_) == 0:
                 cnts_ = vts
             else: