[fix] initialize available_gpu_block_num with max_gpu_block_num (#4193)

Author: liyonghua0910

fastdeploy/cache_manager/prefix_cache_manager.py

@@ -113,6 +113,10 @@ def __init__(
             + f"{self.num_cpu_blocks}, bytes_per_layer_per_block {self.cache_config.bytes_per_layer_per_block}"
         )
 
+        main_process_metrics.max_gpu_block_num.set(self.num_gpu_blocks)
+        main_process_metrics.available_gpu_block_num.set(self.num_gpu_blocks)
+        main_process_metrics.available_gpu_resource.set(1.0)
+
     @property
     def available_gpu_resource(self):
         return len(self.gpu_free_block_list) / self.num_gpu_blocks if self.num_gpu_blocks > 0 else 0.0
@@ -269,6 +273,7 @@ def update_cache_config(self, cache_config):
         self.node_id_pool = list(range(self.num_gpu_blocks + self.num_cpu_blocks))
 
         main_process_metrics.max_gpu_block_num.set(self.num_gpu_blocks)
+        main_process_metrics.available_gpu_block_num.set(self.num_gpu_blocks)
         main_process_metrics.available_gpu_resource.set(1.0)
 
     def can_allocate_gpu_blocks(self, num_blocks: int):