Merge pull request #48525 from rhcs-dashboard/fix-PgImbalance-alert

mgr/dashboard: fix CephPGImbalance alert

Reviewed-by: Avan Thakkar <athakkar@redhat.com>
Reviewed-by: Ernesto Puerta <epuertat@redhat.com>
Reviewed-by: Nizamudeen A <nia@redhat.com>
Reviewed-by: Tatjana Dehler <tdehler@suse.com>

src/pybind/mgr/balancer/module.py

@@ -324,6 +324,7 @@ class Module(MgrModule):
     last_optimize_started = ''
     last_optimize_duration = ''
     optimize_result = ''
+    no_optimization_needed = False
     success_string = 'Optimization plan created successfully'
     in_progress_string = 'in progress'
 
@@ -342,6 +343,7 @@ def show_status(self) -> Tuple[int, str, str]:
             'last_optimize_started': self.last_optimize_started,
             'last_optimize_duration': self.last_optimize_duration,
             'optimize_result': self.optimize_result,
+            'no_optimization_needed': self.no_optimization_needed,
             'mode': self.get_module_option('mode'),
         }
         return (0, json.dumps(s, indent=4, sort_keys=True), '')
@@ -1037,6 +1039,7 @@ def do_upmap(self, plan: Plan) -> Tuple[int, str]:
                 break
         self.log.info('prepared %d/%d changes' % (total_did, max_optimizations))
         if total_did == 0:
+            self.no_optimization_needed = True
             return -errno.EALREADY, 'Unable to find further optimization, ' \
                                     'or pool(s) pg_num is decreasing, ' \
                                     'or distribution is already perfect'

src/pybind/mgr/dashboard/controllers/prometheus.py

@@ -7,6 +7,7 @@
 
 from ..exceptions import DashboardException
 from ..security import Scope
+from ..services import ceph_service
 from ..settings import Settings
 from . import APIDoc, APIRouter, BaseController, Endpoint, RESTController, Router
 
@@ -41,6 +42,9 @@ def alert_proxy(self, method, path, params=None, payload=None):
     def _get_api_url(self, host):
         return host.rstrip('/') + '/api/v1'
 
+    def balancer_status(self):
+        return ceph_service.CephService.send_command('mon', 'balancer status')
+
     def _proxy(self, base_url, method, path, api_name, params=None, payload=None, verify=True):
         # type (str, str, str, str, dict, dict, bool)
         try:
@@ -57,8 +61,16 @@ def _proxy(self, base_url, method, path, api_name, params=None, payload=None, ve
             raise DashboardException(
                 "Error parsing Prometheus Alertmanager response: {}".format(e.msg),
                 component='prometheus')
-        if content['status'] == 'success':
+        balancer_status = self.balancer_status()
+        if content['status'] == 'success':  # pylint: disable=R1702
             if 'data' in content:
+                if balancer_status['active'] and balancer_status['no_optimization_needed'] and path == '/alerts':  # noqa E501  #pylint: disable=line-too-long
+                    for alert in content['data']:
+                        for k, v in alert.items():
+                            if k == 'labels':
+                                for key, value in v.items():
+                                    if key == 'alertname' and value == 'CephPGImbalance':
+                                        content['data'].remove(alert)
                 return content['data']
             return content
         raise DashboardException(content, http_status_code=400, component='prometheus')