dev: debug

clore_hosting/main.py

@@ -554,10 +554,13 @@ class CloreClient:
             try:
                 await monitoring.put("specs_service")
                 current_specs = await specs.get()
+                print(current_specs)
                 if self.last_hw_specs_submit < (utils.unix_timestamp()-1800):
                     self.last_hw_specs_submit=utils.unix_timestamp()
                     await self.submit_specs(current_specs)
+                    print("submit specs")
                 await self.update_realtime_data(current_specs)
+                print("update realtime")
                 try:
                     if self.xfs_state == "active" and len(current_specs["gpus"]["nvidia"]) > 0 and not self.runned_pull_selftest:
                         await clore_partner.check_to_pull_selftest(current_specs)

lib/get_specs.py

@@ -291,8 +291,6 @@ def get_gpu_info():
                         "mem_used": parts[12]
                     }
 
-                    print(xl_gpu_info)
-
                     try:
                         pci_query = parts[2][parts[2].find(':')+1:]
                         for index, valid_pci_dev in enumerate(valid_pci_dev_list):
@@ -302,8 +300,6 @@ def get_gpu_info():
                                     xl_gpu_info["pcie_width"]=bus_spec.width
                                     xl_gpu_info["pcie_revision"]=bus_spec.revision
                     except Exception as e:
-                        print("err")
-                        print(e)
                         pass
                     gpus["nvidia"].append(xl_gpu_info)
             lines = nvidia_smi_stdout.split('\n')
@@ -313,21 +309,13 @@ def get_gpu_info():
 
                     gpu_name = parts[1].strip()
                     gpu_id = parts[5].strip();
-                    print("parts")
-                    print(parts)
-                    print("p1:" + gpu_name)
-                    print("p1:" + gpu_id)
                     if gpu_name == "NVIDIA Graphics Device" and gpu_id in GPU_ID_TO_NAME:
                         gpu_name = GPU_ID_TO_NAME[gpu_id]
-                        print("p2:" + gpu_name)
 
                     gpu_str = f"{len(lines)-1}x {gpu_name}"
-                    print(gpu_str)
                     gpu_mem = round(int(filter_non_numeric(parts[4]).strip())/1024, 2)
         except Exception as e:
             nvml_err=True
-            print("err2")
-            print(e)
             pass
     else:
         nvml_err=True