From bc4f4b486278fdd5284ae6169022e4b1dcd84953 Mon Sep 17 00:00:00 2001
From: Alex Vandiver <alexmv@zulip.com>
Date: Mon, 13 Jun 2022 14:58:37 -0700
Subject: [PATCH] nagios: Make the pageable/not/flaky tri-state clearer.

---
 puppet/zulip_ops/files/nagios4/conf.d/hostgroups.cfg      | 5 +++++
 puppet/zulip_ops/files/nagios4/conf.d/services.cfg        | 6 +++---
 puppet/zulip_ops/templates/nagios4/hosts.cfg.template.erb | 4 ++--
 3 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/puppet/zulip_ops/files/nagios4/conf.d/hostgroups.cfg b/puppet/zulip_ops/files/nagios4/conf.d/hostgroups.cfg
index 1d1073f0e9..6ddbbf0af5 100644
--- a/puppet/zulip_ops/files/nagios4/conf.d/hostgroups.cfg
+++ b/puppet/zulip_ops/files/nagios4/conf.d/hostgroups.cfg
@@ -25,6 +25,11 @@ define hostgroup {
         alias           Servers for whom we do not want to page
         }
 
+define hostgroup {
+        hostgroup_name  flaky_servers
+        alias           Servers for whom we do not page, and have flakier net
+        }
+
 define hostgroup {
         hostgroup_name  web
         alias           Web servers
diff --git a/puppet/zulip_ops/files/nagios4/conf.d/services.cfg b/puppet/zulip_ops/files/nagios4/conf.d/services.cfg
index 9235846967..c3ff49fe83 100644
--- a/puppet/zulip_ops/files/nagios4/conf.d/services.cfg
+++ b/puppet/zulip_ops/files/nagios4/conf.d/services.cfg
@@ -16,7 +16,7 @@ define service {
 
 define service {
         use                             generic-service
-        hostgroup_name                  zmirror
+        hostgroup_name                  flaky_servers
         service_description             SSH for flaky machines
         check_command                   check_ssh
         normal_check_interval           2
@@ -43,7 +43,7 @@ define service {
 
 define service {
         use                             generic-service
-        hostgroup_name                  zmirror
+        hostgroup_name                  flaky_servers
         service_description             remote disk for flaky machines
         check_command                   check_remote_disk!20%!10%
         normal_check_interval           2
@@ -79,7 +79,7 @@ define service {
 
 define service {
         use                             generic-service
-        hostgroup_name                  zmirror
+        hostgroup_name                  flaky_servers
         service_description             remote load for flaky machines
         check_command                   check_remote_load!22!5.0,4.0,3.0!10.0,6.0,4.0
         normal_check_interval           2
diff --git a/puppet/zulip_ops/templates/nagios4/hosts.cfg.template.erb b/puppet/zulip_ops/templates/nagios4/hosts.cfg.template.erb
index 4210b590d4..4daf9ef341 100644
--- a/puppet/zulip_ops/templates/nagios4/hosts.cfg.template.erb
+++ b/puppet/zulip_ops/templates/nagios4/hosts.cfg.template.erb
@@ -34,7 +34,7 @@ define host{
        host_name        <%= host %>
        alias            <%= host %>
        address          <%= host %>.<%= @default_host_domain %>
-       hostgroups       all,non_aws_host,zmirror,zmirror_main
+       hostgroups       all,non_aws_host,zmirror,zmirror_main,flaky_servers
        }
 <% end -%>
 
@@ -44,7 +44,7 @@ define host{
        host_name        <%= host %>
        alias            <%= host %>
        address          <%= host %>.<%= @default_host_domain %>
-       hostgroups       all,non_aws_host,zmirror,zmirrorp
+       hostgroups       all,non_aws_host,zmirror,zmirrorp,flaky_servers
        }
 <% end -%>