File tree Expand file tree Collapse file tree
Expand file tree Collapse file tree Original file line number Diff line number Diff line change @@ -467,14 +467,16 @@ def validate_restored_correct_checkpoint(
467467 namespace = "default" ,
468468 pod_pattern = pod_pattern ,
469469 text_filter = (
470- "(textPayload:\" 'event_type'\" OR jsonPayload.message:\" 'event_type'\" )"
470+ "(textPayload:\" 'event_type'\" OR jsonPayload.message:\" 'event_type'\" OR "
471+ "textPayload:\" restoring from this run's directory step\" OR "
472+ "jsonPayload.message:\" restoring from this run's directory step\" )"
471473 ),
472474 start_time = start_time ,
473475 end_time = end_time ,
474476 )
475477
476478 if not entries :
477- raise AirflowFailException ("No event_type found in the log." )
479+ raise AirflowFailException ("No event_type or restore log found in the log." )
478480
479481 local_saved_steps_before_restore = []
480482 for entry in entries :
@@ -497,15 +499,17 @@ def validate_restored_correct_checkpoint(
497499
498500 local_saved_steps_before_restore .append (int (saved_step_match .group (1 )))
499501
500- elif re .search (r"'event_type': '(emergency_)?restore'" , message ):
502+ elif re .search (
503+ r"'event_type': '(emergency_)?restore'" , message
504+ ) or re .search (r"restoring from this run's directory step \d+" , message ):
501505 logging .info ("Found restore event: %s" , message )
502506 logging .info (
503507 "Saved steps before restore: %s" , local_saved_steps_before_restore
504508 )
505509
506510 restored_step_match = re .search (
507511 r"'step':\s*(?:np\.int32\()?(\d+)" , message
508- )
512+ ) or re . search ( r"restoring from this run's directory step (\d+)" , message )
509513 restored_step = (
510514 int (restored_step_match .group (1 )) if restored_step_match else None
511515 )
You can’t perform that action at this time.
0 commit comments