aboutsummaryrefslogtreecommitdiffstats
path: root/terraform-ci-infra/1n_nmd/prometheus
diff options
context:
space:
mode:
authorpmikus <pmikus@cisco.com>2021-03-05 11:36:29 +0000
committerPeter Mikus <pmikus@cisco.com>2021-03-05 11:51:51 +0000
commit5551ffb9e6fb4f6de54331d9105bc1e4bbd0641e (patch)
tree806ea258a70207bdfc61d8a25a6078171231e983 /terraform-ci-infra/1n_nmd/prometheus
parent85785b4fac91e2bc443d4b62b80df418d8a99699 (diff)
FIX: JenkinsJob alert condition
Signed-off-by: pmikus <pmikus@cisco.com> Change-Id: I1b996d0ed6252ed2b790656879e9068ff9c52039
Diffstat (limited to 'terraform-ci-infra/1n_nmd/prometheus')
-rw-r--r--terraform-ci-infra/1n_nmd/prometheus/conf/nomad/prometheus.hcl4
1 files changed, 2 insertions, 2 deletions
diff --git a/terraform-ci-infra/1n_nmd/prometheus/conf/nomad/prometheus.hcl b/terraform-ci-infra/1n_nmd/prometheus/conf/nomad/prometheus.hcl
index d851628fcd..235a04228c 100644
--- a/terraform-ci-infra/1n_nmd/prometheus/conf/nomad/prometheus.hcl
+++ b/terraform-ci-infra/1n_nmd/prometheus/conf/nomad/prometheus.hcl
@@ -191,7 +191,7 @@ groups:
- name: "Jenkins Job Health Exporter"
rules:
- alert: JenkinsJobHealthExporterFailures
- expr: jenkins_job_failure{id=~".*"} >= 10
+ expr: jenkins_job_failure{id=~".*"} > jenkins_job_success{id=~".*"}
for: 0m
labels:
severity: critical
@@ -199,7 +199,7 @@ groups:
summary: "Jenkins Job Health detected high failure rate on jenkins jobs."
description: "Job: {{ $labels.id }}"
- alert: JenkinsJobHealthExporterUnstable
- expr: jenkins_job_unstable{id=~".*"} >= 10
+ expr: jenkins_job_unstable{id=~".*"} > jenkins_job_success{id=~".*"}
for: 0m
labels:
severity: warning