Merge "Tracked Steps which errored out in the runbook metrics"

-- Branch commit log -- commit 60c1640c0e4d6297b0626546f6d61a031780efe5 Author: Kwesi Graham <[email protected]> Date: 2024-09-21T04:06:39Z Tracked Steps which errored out in the runbook metrics Change-Id: Ie3a72ba15f5a2b76c1cd03f759fb5b47dc5eabd3 GitOrigin-RevId: 2ed613000c6d136edf3de1427f494f5dc6be0a81
GoogleCloudPlatform · Sep 23, 2024 · 7d97161 · 7d97161
1 parent 0aff6f5
commit 7d97161
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 6 deletions.
diff --git a/gcpdiag/runbook/command.py b/gcpdiag/runbook/command.py
@@ -21,7 +21,7 @@
 import re
 import sys
 import warnings
-from typing import List
+from typing import List, Tuple
 
 import yaml
 
@@ -312,7 +312,7 @@ def _initialize_output():
   return output
 
 
-def run_and_get_report(argv=None, credentials: str = None) -> dict:
+def run_and_get_report(argv=None, credentials: str = None) -> Tuple[int, dict]:
   # Initialize argument parser
   parser = _init_runbook_args_parser()
   args = parser.parse_args(argv[1:])
@@ -389,8 +389,10 @@ def run_and_get_report(argv=None, credentials: str = None) -> dict:
   output.display_footer(dt_engine.interface.rm)
   # Clean up the kubeconfig file generated for gcpdiag
   kubectl.clean_up()
-  return report
+  # return success if we get to this point and the report..
+  return (0, report)
 
 
 def run(argv) -> None:
-  run_and_get_report(argv)
+  code, _ = run_and_get_report(argv)
+  sys.exit(code)
diff --git a/gcpdiag/runbook/gce/generalized_steps.py b/gcpdiag/runbook/gce/generalized_steps.py
@@ -124,7 +124,7 @@ class HighVmDiskUtilization(runbook.Step):
   project_id: str
   zone: str
   instance_name: str
-  serial_console_file: str
+  serial_console_file: str = ''
 
   def execute(self):
     """Verifying VM's Boot disk space utilization is within optimal levels."""

diff --git a/gcpdiag/runbook/report.py b/gcpdiag/runbook/report.py
@@ -236,20 +236,21 @@ def get_totals_by_status(self) -> Dict[str, int]:
   def generate_report_metrics(self, report: Report) -> Dict[str, dict]:
     reports_metrics: Dict[str, Any] = {}
     all_step_metrics = []
+    reports_metrics['execution_mode'] = report.execution_mode
     for result in report.results.values():
       step_metrics: Dict[str, dict] = collections.defaultdict(dict)
       start = util.parse_time_input(result.start_time)
       end = util.parse_time_input(result.end_time)
       duration = (end - start).total_seconds() * 1000
       step_metrics[result.step.id]['execution_duration'] = duration
       step_metrics[result.step.id]['totals_by_status'] = result.totals_by_status
+      step_metrics[result.step.id]['error'] = bool(result.step_error)
       all_step_metrics.append(step_metrics)
     if report.runbook_name:
       start = util.parse_time_input(report.run_start_time)
       end = util.parse_time_input(report.run_end_time)
       duration = (end - start).total_seconds() * 1000
       reports_metrics['runbook_name'] = report.runbook_name
-      reports_metrics['execution_mode'] = report.execution_mode
       reports_metrics['run_duration'] = duration
       reports_metrics['totals_by_status'] = report.get_totals_by_status()
       reports_metrics['steps'] = all_step_metrics