From 36a0a300703992c763f6403df8ca98aa786a4658 Mon Sep 17 00:00:00 2001
From: Maha Benzekri <maha.benzekri@scality.com>
Date: Tue, 8 Oct 2024 12:26:14 +0200
Subject: [PATCH] fixup post review

---
 monitoring/mongodb/alerts.test.yaml | 43 +++++++++++++++--------------
 monitoring/mongodb/alerts.yaml      |  4 +--
 2 files changed, 25 insertions(+), 22 deletions(-)

diff --git a/monitoring/mongodb/alerts.test.yaml b/monitoring/mongodb/alerts.test.yaml
index c6d3bf337..1a6c3423c 100644
--- a/monitoring/mongodb/alerts.test.yaml
+++ b/monitoring/mongodb/alerts.test.yaml
@@ -329,28 +329,31 @@ tests:
               description: "MongoDB pod `data-db-mongodb-sharded-mongos-0` has been in the 'STARTUP2' state for more than 1 hour. Please ensure that the instance is running properly."
               summary: MongoDB node in STARTUP2 state for too long
     
-
   - name: MongoDbRSNotSynced
     interval: 1m
     input_series:
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-0", member_state="PRIMARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-0.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 1x10
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-0", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-1.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 2x10
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-0", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-2.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 2x10
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-1", member_state="PRIMARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-0.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 1x10
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-1", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-1.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 2x10
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-1", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-2.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 2x10
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-2", member_state="PRIMARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-0.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 1x10
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-2", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-1.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 2x10
-      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-2", member_state="(not reachable/healthy)", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="data-db-mongodb-sharded-shard0-data-2.data-db-mongodb-sharded-headless.zenko.svc.cluster.local:27017"}
-        values: 2 _ _ _ _ _ _ _ _ _
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-0", member_state="PRIMARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-0"}
+        values: 1x10 1x10
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-0", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-1"}
+        values: 2x10 2x10
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-0", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-2"}
+        values: 2x10 stale
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-0", member_state="(not reachable/healthy)", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-2"}
+        values: _x10 8x10
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-1", member_state="PRIMARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-0"}
+        values: 1x10 1x10
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-1", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-1"}
+        values: 2x10 2x10
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-1", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-2"}
+        values: 2x10 stale
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-1", member_state="(not reachable/healthy)", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-2"}
+        values: _x10 8x10
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-2", member_state="PRIMARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-0"}
+        values: 1x10 stale
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-2", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-1"}
+        values: 2x10 stale
+      - series: mongodb_rs_members_state{namespace="zenko", pod="data-db-mongodb-sharded-shard0-data-2", member_state="SECONDARY", rs_nm="data-db-mongodb-sharded-shard-0", member_idx="shard0-data-2"}
+        values: 2x10 stale
 
     alert_rule_test:
       - alertname: MongoDbRSNotSynced
@@ -360,5 +363,5 @@ tests:
               severity: warning
               rs_nm: data-db-mongodb-sharded-shard-0
             exp_annotations:
-              description: "MongoDB replica set `data-db-mongodb-sharded-shard-0` is not in the expected state. It currently has `1` SECONDARY members instead of the expected number. Please ensure that all instances are running properly."
+              description: "MongoDB replica set `data-db-mongodb-sharded-shard-0` is not in the expected state. It does not have the expected number of SECONDARY members. Please ensure that all instances are running properly."
               summary: MongoDB replica set out of sync
diff --git a/monitoring/mongodb/alerts.yaml b/monitoring/mongodb/alerts.yaml
index f5acde22f..cdbe2b3c8 100644
--- a/monitoring/mongodb/alerts.yaml
+++ b/monitoring/mongodb/alerts.yaml
@@ -183,10 +183,10 @@ groups:
 
   - alert: MongoDbRSNotSynced
     expr: |
-      floor(avg by(rs_nm)(count by (rs_nm, pod)(mongodb_rs_members_state{namespace="${namespace}", pod=~"${service}.*", member_state="SECONDARY"}))) != (${replicas} - 1) 
+      group by(rs_nm) ( count by(rs_nm, pod) (mongodb_rs_members_state{namespace="${namespace}", pod=~"${service}.*", member_state="SECONDARY"}) != (${replicas} - 1) )
     for: 10m
     labels:
       severity: warning
     annotations:
-      description: "MongoDB replica set `{{ $labels.rs_nm }}` is not in the expected state. It currently has `{{ $value }}` SECONDARY members instead of the expected number. Please ensure that all instances are running properly."
+      description: "MongoDB replica set `{{ $labels.rs_nm }}` is not in the expected state. It does not have the expected number of SECONDARY members. Please ensure that all instances are running properly."
       summary: MongoDB replica set out of sync