Merge "Fix building report with customized stepd names + add disable_…

…paging parameters to the log queries"

-- Branch commit log --
commit b488c91ad7786b03c8e4ca07340bd58b64b1d061
Author:  Natalia Kaczor <[email protected]>
Date:    2025-02-05T17:40:29Z

    Fix building report with customized stepd names + add disable_paging parameters to the log queries

Change-Id: I8dfb09564f1535130342c9ebf01077f222e2cca0
GitOrigin-RevId: 32ba4f6cfbada92d2a70cd5ddc8b96c90c006257

gcpdiag/queries/logs.py

@@ -44,11 +44,12 @@
 from typing import (Any, Deque, Dict, List, Mapping, Optional, Sequence, Set,
                     Tuple, Union)
 
+import apiclient.errors
 import dateutil.parser
 import ratelimit
 from boltons.iterutils import get_path
 
-from gcpdiag import caching, config, models
+from gcpdiag import caching, config, models, utils
 from gcpdiag.queries import apis
 
 
@@ -162,7 +163,12 @@ def query(project_id: str, resource_type: str, log_name: str,
                   period=config.get('logging_ratelimit_period_seconds'))
 def _ratelimited_execute(req):
   """Wrapper to req.execute() with rate limiting to avoid hitting quotas."""
-  return req.execute(num_retries=config.API_RETRIES)
+  try:
+    return req.execute(num_retries=config.API_RETRIES)
+  except apiclient.errors.HttpError as err:
+    logging.error('failed to execute logging request for request %s. Error: %s',
+                  req, err)
+    raise utils.GcpApiError(err) from err
 
 
 def _execute_query_job(job: _LogsQueryJob):
@@ -240,7 +246,11 @@ def _execute_query_job(job: _LogsQueryJob):
 
 
 @caching.cached_api_call
-def realtime_query(project_id, filter_str, start_time, end_time):
+def realtime_query(project_id,
+                   filter_str,
+                   start_time,
+                   end_time,
+                   disable_paging=False):
   """Intended for use in only runbooks. use logs.query() for lint rules."""
   logging_api = apis.get_api('logging', 'v2', project_id)
 
@@ -283,6 +293,8 @@ def realtime_query(project_id, filter_str, start_time, end_time):
           'maximum query runtime for log query reached (project: %s, query: %s).',
           project_id, filter_str.replace('\n', ' AND '))
       return deque
+    if disable_paging:
+      break
     req = logging_api.entries().list_next(req, res)
     if req is not None:
       logging.info('still fetching logs (project: %s, max wait: %ds)',

gcpdiag/runbook/lb/snapshots/unhealthy_backends.txt

@@ -49,13 +49,13 @@ lb/unhealthy-backends: Load Balancer Unhealthy Backends Analyzer.
      Group https://www.googleapis.com/compute/v1/projects/gcpdiag-lb2-aaaa/zones/us-east1-b/instanceGroups/lb-backend-example has 2/2 unhealthy backends
 
 
-[GATEWAY]: Check if health check logging is enabled.
+[GATEWAY]: Verify health check logging enabled for backend service "web-backend-service" in region "global".
 
    - gcpdiag-lb2-aaaa/http-basic-check                                    [OK]
      [REASON]
      Health check logging is enabled for health check projects/gcpdiag-lb1-aaaa/global/healthChecks/http-basic-check.
 
-[GATEWAY]: Look for the latest health check logs and based on that decide what to do next.
+[GATEWAY]: Analyze latest health check log for backend service "web-backend-service" in region "global".
 [AUTOMATED STEP]: Analyze TIMEOUT health check logs for backend service "web-backend-service" in region "global".
 
    - gcpdiag-lb2-aaaa/web-backend-service                                 [UNCERTAIN]
@@ -276,13 +276,13 @@ lb/unhealthy-backends: Load Balancer Unhealthy Backends Analyzer.
      Group https://www.googleapis.com/compute/v1/projects/gcpdiag-lb2-aaaa/zones/europe-west4-b/networkEndpointGroups/neg1 has 1/1 unhealthy backends
 
 
-[GATEWAY]: Check if health check logging is enabled.
+[GATEWAY]: Verify health check logging enabled for backend service "backend-service-2" in region "europe-west4".
 
    - gcpdiag-lb2-aaaa/tcp-basic-check-2                                   [OK]
      [REASON]
      Health check logging is enabled for health check projects/gcpdiag-lb2-aaaa/regions/europe-west4/healthChecks/tcp-basic-check-2.
 
-[GATEWAY]: Look for the latest health check logs and based on that decide what to do next.
+[GATEWAY]: Analyze latest health check log for backend service "backend-service-2" in region "europe-west4".
 [AUTOMATED STEP]: Analyze TIMEOUT health check logs for backend service "backend-service-2" in region "europe-west4".
 
    - gcpdiag-lb2-aaaa/backend-service-2                                   [UNCERTAIN]