Added alert signalling no new logs sent via Elasticsearch exporter.
Related-PROD: PROD-34821
Change-Id: Ied7279d3a107fad302d14b10f218ce1b53ca3f9f
diff --git a/prometheus/meta/prometheus.yml b/prometheus/meta/prometheus.yml
index 752a84f..46b6e9a 100644
--- a/prometheus/meta/prometheus.yml
+++ b/prometheus/meta/prometheus.yml
@@ -1,5 +1,5 @@
{%- if pillar.prometheus is defined %}
-{%- from "prometheus/map.jinja" import server, alertmanager, remote_storage_adapter, monitoring, relay, sf_notifier with context %}
+{%- from "prometheus/map.jinja" import server, alertmanager, remote_storage_adapter, monitoring, relay, sf_notifier, elasticsearch_exporter with context %}
{%- if server.get('enabled', False) or remote_storage_adapter.get('enabled', False) or alertmanager.get('enabled', False) or relay.get('enabled', False) or sf_notifier.get('enabled', False) %}
server:
alert:
@@ -183,6 +183,19 @@
description: "The sf-notifier service fails to authenticate to Salesforce for 2 minutes."
{%- endraw %}
{%- endif %}
+{%- if elasticsearch_exporter.get('enabled', False) %}
+{%- raw %}
+ ElasticsearchExporterNoDailyLogs:
+ if: >-
+ (sum by (host) (changes(logs_program_host_doc_count[3h])) or sum by (host) (up{host!=""})*0) == 0
+ labels:
+ severity: warning
+ service: elasticsearch-exporter
+ annotations:
+ summary: "No new logs sent from a node within the last 3 hours"
+ description: "There were no new logs shipped from {{ $labels.host }} node within the last 3 hours."
+{%- endraw %}
+{%- endif %}
{%- if server.get("enabled", False) and not server.get("is_container", True) and relay.get('enabled', False) or sf_notifier.get('enabled', False) %}
target:
{%- if sf_notifier.get('enabled', False) %}