fix pp problem and remove err logs when duplicate create (#191)

Author: qyh111

ucm/csrc/ucmnfsstore/cc/api/ucmnfsstore/ucmnfsstore.cc

@@ -67,7 +67,7 @@ int32_t Setup(const SetupParam& param)
 int32_t Alloc(const std::string& blockId)
 {
     auto s = Singleton<SpaceManager>::Instance()->NewBlock(blockId);
-    if (s.Failure()) {
+    if (s.Failure() && s != Status::DuplicateKey()) {
         UC_ERROR("Failed to allocate kv cache block space, block id: {}, error code: {}.", blockId, s.Underlying());
     }
     return s.Underlying();

ucm/csrc/ucmnfsstore/cc/domain/space/space_manager.cc

@@ -60,7 +60,9 @@ Status SpaceManager::NewBlock(const std::string& blockId) const
     }
     status = file->Open(IFile::OpenFlag::CREATE | IFile::OpenFlag::EXCL | IFile::OpenFlag::READ_WRITE);
     if (status.Failure()) {
-        UC_ERROR("Failed({}) to new block({}).", status, blockId);
+        if (status != Status::DuplicateKey()) {
+            UC_ERROR("Failed({}) to new block({}).", status, blockId);
+        }
         return status;
     }
     status = file->Truncate(this->_blockSize);

ucm/csrc/ucmnfsstore/cc/infra/file/posix_file.cc

@@ -98,7 +98,9 @@ Status PosixFile::Open(const uint32_t flags)
     auto status = this->_handle >= 0 ? Status::OK() : Status::OsApiError();
     if (status.Failure()) {
         if (eno == EEXIST) { status = Status::DuplicateKey(); }
-        UC_ERROR("Failed({},{}) to open file({}) with flags({}).", eno, status, this->Path(), flags);
+        if (status != Status::DuplicateKey()) {
+            UC_ERROR("Failed({},{}) to open file({}) with flags({}).", eno, status, this->Path(), flags);
+        }
     }
     return status;
 }

ucm/integration/vllm/uc_connector.py

@@ -174,7 +174,8 @@ def _initialize_dataoffset(self, vllm_config: "VllmConfig"):
         # layer_id -> rank -> k_offset
         self.k_data_offsets: dict[int, dict[int, int]] = {}
 
-        for layer_id in range(self.num_layers):
+        pp_size = vllm_config.parallel_config.pipeline_parallel_size
+        for layer_id in range(self.num_layers * pp_size):
             self.k_data_offsets[layer_id] = {}
             for rank in range(self.total_tp_size):
                 if self.is_mla: