6161 StaticXgmiPlpd ,
6262 ValueUnit ,
6363)
64- from nodescraper .utils import get_exception_details , get_exception_traceback
64+ from nodescraper .utils import get_exception_traceback
6565
6666
6767class AmdSmiCollector (InBandDataCollector [AmdSmiDataModel , None ]):
@@ -294,12 +294,12 @@ def _get_amdsmi_data(self) -> Optional[AmdSmiDataModel]:
294294 firmware = firmware ,
295295 static = statics ,
296296 )
297- except ValidationError as e :
298- self .logger .warning ("Validation err: %s" , e )
297+ except ValidationError as err :
298+ self .logger .warning ("Validation err: %s" , err )
299299 self ._log_event (
300300 category = EventCategory .APPLICATION ,
301301 description = "Failed to build AmdSmiDataModel" ,
302- data = get_exception_details ( e ) ,
302+ data = { "errors" : err . errors ( include_url = False )} ,
303303 priority = EventPriority .ERROR ,
304304 )
305305 return None
@@ -325,11 +325,11 @@ def _get_amdsmi_version(self) -> Optional[AmdSmiVersion]:
325325 amdsmi_library_version = version_data .get ("amdsmi_library_version" , "" ),
326326 rocm_version = version_data .get ("rocm_version" , "" ),
327327 )
328- except ValidationError as e :
328+ except ValidationError as err :
329329 self ._log_event (
330330 category = EventCategory .APPLICATION ,
331331 description = "Failed to build AmdSmiVersion" ,
332- data = get_exception_details ( e ) ,
332+ data = { "errors" : err . errors ( include_url = False )} ,
333333 priority = EventPriority .WARNING ,
334334 )
335335 return None
@@ -368,11 +368,11 @@ def _to_int(x: Any, default: int = 0) -> int:
368368 partition_id = _to_int (item .get ("partition_id" , 0 )),
369369 )
370370 )
371- except ValidationError as e :
371+ except ValidationError as err :
372372 self ._log_event (
373373 category = EventCategory .APPLICATION ,
374374 description = "Failed to build AmdSmiListItem" ,
375- data = {"exception " : get_exception_traceback ( e ), "item" : item },
375+ data = {"errors " : err . errors ( include_url = False ), "item" : item },
376376 priority = EventPriority .WARNING ,
377377 )
378378
@@ -441,12 +441,12 @@ def get_process(self) -> Optional[list[Processes]]:
441441 )
442442 )
443443 )
444- except ValidationError as e :
444+ except ValidationError as err :
445445 self ._log_event (
446446 category = EventCategory .APPLICATION ,
447447 description = "Failed to build ProcessListItem; skipping entry" ,
448448 data = {
449- "exception " : get_exception_traceback ( e ),
449+ "errors " : err . errors ( include_url = False ),
450450 "gpu_index" : gpu_idx ,
451451 "entry" : repr (entry ),
452452 },
@@ -456,11 +456,11 @@ def get_process(self) -> Optional[list[Processes]]:
456456
457457 try :
458458 out .append (Processes (gpu = gpu_idx , process_list = plist ))
459- except ValidationError as e :
459+ except ValidationError as err :
460460 self ._log_event (
461461 category = EventCategory .APPLICATION ,
462462 description = "Failed to build Processes" ,
463- data = {"exception " : get_exception_traceback ( e ), "gpu_index" : gpu_idx },
463+ data = {"errors " : err . errors ( include_url = False ), "gpu_index" : gpu_idx },
464464 priority = EventPriority .WARNING ,
465465 )
466466
@@ -492,12 +492,12 @@ def get_partition(self) -> Optional[Partition]:
492492 memparts .append (
493493 PartitionMemory (gpu_id = gpu_idx , partition_type = str (mem_pt ) if mem_pt else None )
494494 )
495- except ValidationError as e :
495+ except ValidationError as err :
496496 self ._log_event (
497497 category = EventCategory .APPLICATION ,
498498 description = "Failed to build PartitionMemory" ,
499499 data = {
500- "exception " : get_exception_traceback ( e ),
500+ "errors " : err . errors ( include_url = False ),
501501 "gpu_index" : gpu_idx ,
502502 "data" : mem_pt ,
503503 },
@@ -510,12 +510,12 @@ def get_partition(self) -> Optional[Partition]:
510510 gpu_id = gpu_idx , partition_type = str (comp_pt ) if comp_pt else None
511511 )
512512 )
513- except ValidationError as e :
513+ except ValidationError as err :
514514 self ._log_event (
515515 category = EventCategory .APPLICATION ,
516516 description = "Failed to build PartitionCompute" ,
517517 data = {
518- "exception " : get_exception_traceback ( e ),
518+ "errors " : err . errors ( include_url = False ),
519519 "gpu_index" : gpu_idx ,
520520 "data" : comp_pt ,
521521 },
@@ -524,11 +524,11 @@ def get_partition(self) -> Optional[Partition]:
524524
525525 try :
526526 return Partition (memory_partition = memparts , compute_partition = computeparts )
527- except ValidationError as e :
527+ except ValidationError as err :
528528 self ._log_event (
529529 category = EventCategory .APPLICATION ,
530530 description = "Failed to build Partition" ,
531- data = {"exception " : get_exception_traceback ( e )},
531+ data = {"errors " : err . errors ( include_url = False )},
532532 priority = EventPriority .WARNING ,
533533 )
534534 return None
@@ -577,11 +577,11 @@ def get_firmware(self) -> Optional[list[Fw]]:
577577
578578 try :
579579 out .append (Fw (gpu = gpu_idx , fw_list = normalized ))
580- except ValidationError as e :
580+ except ValidationError as err :
581581 self ._log_event (
582582 category = EventCategory .APPLICATION ,
583583 description = "Failed to build Fw" ,
584- data = {"exception " : get_exception_traceback ( e ), "gpu_index" : gpu_idx },
584+ data = {"errors " : err . errors ( include_url = False ), "gpu_index" : gpu_idx },
585585 priority = EventPriority .WARNING ,
586586 )
587587
@@ -757,12 +757,12 @@ def get_static(self) -> Optional[list[AmdSmiStatic]]:
757757 clock = clock_dict_model ,
758758 )
759759 )
760- except ValidationError as e :
761- self .logger .error (e )
760+ except ValidationError as err :
761+ self .logger .error (err )
762762 self ._log_event (
763763 category = EventCategory .APPLICATION ,
764764 description = "Failed to build AmdSmiStatic" ,
765- data = {"exception " : get_exception_traceback ( e ), "gpu_index" : gpu_idx },
765+ data = {"errors " : err . errors ( include_url = False ), "gpu_index" : gpu_idx },
766766 priority = EventPriority .WARNING ,
767767 )
768768
@@ -920,11 +920,11 @@ def _as_list_str(v: Any) -> list[str]:
920920 num_cache_instance = num_cache_instance ,
921921 )
922922 )
923- except ValidationError as ve :
923+ except ValidationError as err :
924924 self ._log_event (
925925 category = EventCategory .APPLICATION ,
926926 description = "Bad cache info entry from amd-smi; skipping" ,
927- data = {"entry" : repr (e ), "exception " : get_exception_traceback ( ve )},
927+ data = {"entry" : repr (e ), "errors " : err . errors ( include_url = False )},
928928 priority = EventPriority .WARNING ,
929929 )
930930 continue
0 commit comments