From ec0fa220009dc9dc4a30310f6e1514d8bfc87b83 Mon Sep 17 00:00:00 2001
From: Maha Benzekri <maha.benzekri@scality.com>
Date: Tue, 20 Aug 2024 17:28:57 +0200
Subject: [PATCH] fix

---
 monitoring/pra/alerts.yaml | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/monitoring/pra/alerts.yaml b/monitoring/pra/alerts.yaml
index f3e3c7f2e4..3c05c65098 100644
--- a/monitoring/pra/alerts.yaml
+++ b/monitoring/pra/alerts.yaml
@@ -27,9 +27,9 @@ groups:
 
   - alert: DrResourcePausedWhileOtherRunning
     expr: |
-      (sum(up{job="${kafka_connect_src_job}", namespace="${namespace}"}) or vector(0)) 
+      (sum(up{drSinkInstancejob="",job="${kafka_connect_src_job}", namespace="${namespace}"}) or vector(0)) 
       !=
-      (sum(up{job="${kafka_connect_sink_job}", namespace="${namespace}"}) or vector(0))
+      (sum(up{drSinkInstancejob="${dr_sink_instance}", job="${kafka_connect_sink_job}", namespace="${namespace}"}) or vector(0))
     for: 1m
     labels:
       severity: warning
@@ -39,7 +39,7 @@ groups:
 
   - alert: DrResourcePausedForTooLong
     expr: |
-      (sum(up{job="${kafka_connect_src_job}", namespace="${namespace}"}) or vector(0)) == 0
+      (sum(up{drSinkInstancejob="", job="${kafka_connect_src_job}", namespace="${namespace}"}) or vector(0)) == 0
     for: 5m
     labels:
       severity: warning
@@ -49,9 +49,9 @@ groups:
 
   - alert: KafkaConnectOutageSource
     expr: |
-      sum(rate(kafka_connect_task_error_total_record_errors{job="${kafka_connect_src_job}"}[$__rate_interval])) > 0
+      sum(rate(kafka_connect_task_error_total_record_errors{drSinkInstancejob="", job="${kafka_connect_src_job}"}[$__rate_interval])) > 0
       or
-      sum(rate(kafka_connect_task_error_total_record_failures{job="${kafka_connect_src_job}"}[$__rate_interval])) > 0
+      sum(rate(kafka_connect_task_error_total_record_failures{drSinkInstancejob="", job="${kafka_connect_src_job}"}[$__rate_interval])) > 0
     for: 1m
     labels:
       severity: critical
@@ -62,9 +62,9 @@ groups:
 
   - alert: KafkaConnectOutageSink
     expr: |
-      sum(rate(kafka_connect_task_error_total_record_errors{job="${kafka_connect_sink_job}"}[$__rate_interval])) > 0
+      sum(rate(kafka_connect_task_error_total_record_errors{drSinkInstancejob="${dr_sink_instance}", job="${kafka_connect_sink_job}"}[$__rate_interval])) > 0
       or
-      sum(rate(kafka_connect_task_error_total_record_failures{job="${kafka_connect_sink_job}"}[$__rate_interval])) > 0
+      sum(rate(kafka_connect_task_error_total_record_failures{drSinkInstancejob="${dr_sink_instance}", job="${kafka_connect_sink_job}"}[$__rate_interval])) > 0
     for: 1m
     labels:
       severity: critical