Add warnings to json result (for UI to use)

5 files changed, 27 insertions, 5 deletions

diff --git a/robusta_krr/core/integrations/prometheus/loader.py b/robusta_krr/core/integrations/prometheus/loader.py
index 9129bb6..c099536 100644
--- a/robusta_krr/core/integrations/prometheus/loader.py
+++ b/robusta_krr/core/integrations/prometheus/loader.py
@@ -70,7 +70,11 @@ class PrometheusMetricsLoader:
         return None
 
     async def load_pods(self, object: K8sObjectData, period: datetime.timedelta) -> list[PodData]:
-        return await self.loader.load_pods(object, period)
+        try:
+            return await self.loader.load_pods(object, period)
+        except Exception as e:
+            logger.exception(f"Failed to load pods for {object}: {e}")
+            return []
 
     async def gather_data(
         self,
diff --git a/robusta_krr/core/integrations/prometheus/metrics_service/prometheus_metrics_service.py b/robusta_krr/core/integrations/prometheus/metrics_service/prometheus_metrics_service.py
index 7894e53..091041e 100644
--- a/robusta_krr/core/integrations/prometheus/metrics_service/prometheus_metrics_service.py
+++ b/robusta_krr/core/integrations/prometheus/metrics_service/prometheus_metrics_service.py
@@ -142,6 +142,11 @@ class PrometheusMetricsService(MetricsService):
         data = await metric_loader.load_data(object, period, step)
 
         if len(data) == 0:
+            if "CPU" in LoaderClass.__name__:
+                object.add_warning("NoPrometheusCPUMetrics")
+            elif "Memory" in LoaderClass.__name__:
+                object.add_warning("NoPrometheusMemoryMetrics")
+
             logger.warning(
                 f"{metric_loader.service_name} returned no {metric_loader.__class__.__name__} metrics for {object}"
             )
diff --git a/robusta_krr/core/models/objects.py b/robusta_krr/core/models/objects.py
index c5f6a98..62149c3 100644
--- a/robusta_krr/core/models/objects.py
+++ b/robusta_krr/core/models/objects.py
@@ -27,6 +27,13 @@ class HPAData(pd.BaseModel):
     target_memory_utilization_percentage: Optional[float]
 
 
+PodWarning = Literal[
+    "NoPrometheusPods",
+    "NoPrometheusCPUMetrics",
+    "NoPrometheusMemoryMetrics",
+]
+
+
 class K8sObjectData(pd.BaseModel):
     # NOTE: Here None means that we are running inside the cluster
     cluster: Optional[str]
@@ -37,6 +44,7 @@ class K8sObjectData(pd.BaseModel):
     namespace: str
     kind: KindLiteral
     allocations: ResourceAllocations
+    warnings: set[PodWarning] = set()
 
     _api_resource = pd.PrivateAttr(None)
 
@@ -46,6 +54,9 @@ class K8sObjectData(pd.BaseModel):
     def __hash__(self) -> int:
         return hash(str(self))
 
+    def add_warning(self, warning: PodWarning) -> None:
+        self.warnings.add(warning)
+
     @property
     def current_pods_count(self) -> int:
         return len([pod for pod in self.pods if not pod.deleted])
diff --git a/robusta_krr/core/runner.py b/robusta_krr/core/runner.py
index 7bf5b32..f50f55f 100644
--- a/robusta_krr/core/runner.py
+++ b/robusta_krr/core/runner.py
@@ -141,6 +141,7 @@ class Runner:
 
             # NOTE: Kubernetes API returned pods, but Prometheus did not
             if object.pods != []:
+                object.add_warning("NoPrometheusPods")
                 logger.warning(
                     f"Was not able to load any pods for {object} from Prometheus.\n\t"
                     "This could mean that Prometheus is missing some required metrics.\n\t"
@@ -154,12 +155,12 @@ class Runner:
             step=self._strategy.settings.timeframe_timedelta,
         )
 
-        logger.debug(f"Calculating recommendations for {object} with {len(metrics)} metrics")
-
         # NOTE: We run this in a threadpool as the strategy calculation might be CPU intensive
         # But keep in mind that numpy calcluations will not block the GIL
         loop = asyncio.get_running_loop()
         result = await loop.run_in_executor(self._executor, self._strategy.run, metrics, object)
+
+        logger.info(f"Calculated recommendations for {object} (using {len(metrics)} metrics)")
         return self._format_result(result)
 
     async def _gather_object_allocations(self, k8s_object: K8sObjectData) -> ResourceScan:
diff --git a/robusta_krr/utils/progress_bar.py b/robusta_krr/utils/progress_bar.py
index 57dc113..b796b6a 100644
--- a/robusta_krr/utils/progress_bar.py
+++ b/robusta_krr/utils/progress_bar.py
@@ -1,6 +1,6 @@
 from alive_progress import alive_bar
 
-from robusta_krr.core.models.config import settings
+# from robusta_krr.core.models.config import settings
 
 
 class ProgressBar:
@@ -12,7 +12,8 @@ class ProgressBar:
     """
 
     def __init__(self, **kwargs) -> None:
-        self.show_bar = not settings.quiet and not settings.log_to_stderr
+        # self.show_bar = not settings.quiet and not settings.log_to_stderr
+        self.show_bar = False  # FIXME: Progress bar is not working good with other logs
         if self.show_bar:
             self.alive_bar = alive_bar(**kwargs, enrich_print=False)