From 299beab1a4d5cd3a8c72d1f8cd468dc10b2f7379 Mon Sep 17 00:00:00 2001
From: Alexandre Gaillet <alexandre.gaillet@fr.clara.net>
Date: Fri, 25 May 2018 17:13:50 +0200
Subject: [PATCH] MON-32 - Fix redis_cpu_high monitor

---
 cloud/aws/elasticache/redis/locals.tf         | 92 +++++++++----------
 cloud/aws/elasticache/redis/monitors-redis.tf | 13 +--
 2 files changed, 51 insertions(+), 54 deletions(-)

diff --git a/cloud/aws/elasticache/redis/locals.tf b/cloud/aws/elasticache/redis/locals.tf
index 89858cc..591525d 100644
--- a/cloud/aws/elasticache/redis/locals.tf
+++ b/cloud/aws/elasticache/redis/locals.tf
@@ -1,53 +1,53 @@
 locals {
   core = {
-    cache.t2.micro    = "1"
-    cache.t2.small    = "1"
-    cache.t2.medium   = "2"
-    cache.m3.medium   = "1"
-    cache.m3.large    = "2"
-    cache.m3.xlarge   = "4"
-    cache.m3.2xlarge  = "8"
-    cache.m4.large    = "2"
-    cache.m4.xlarge   = "4"
-    cache.m4.2xlarge  = "8"
-    cache.m4.4xlarge  = "16"
-    cache.m4.10xlarge = "40"
-    cache.r3.large    = "2"
-    cache.r3.xlarge   = "4"
-    cache.r3.2xlarge  = "8"
-    cache.r3.4xlarge  = "16"
-    cache.r3.8xlarge  = "32"
-    cache.r4.large    = "2"
-    cache.r4.xlarge   = "4"
-    cache.r4.2xlarge  = "8"
-    cache.r4.4xlarge  = "16"
-    cache.r4.8xlarge  = "32"
-    cache.r4.16xlarge = "64"
+    cache.t2.micro    = 1
+    cache.t2.small    = 1
+    cache.t2.medium   = 2
+    cache.m3.medium   = 1
+    cache.m3.large    = 2
+    cache.m3.xlarge   = 4
+    cache.m3.2xlarge  = 8
+    cache.m4.large    = 2
+    cache.m4.xlarge   = 4
+    cache.m4.2xlarge  = 8
+    cache.m4.4xlarge  = 16
+    cache.m4.10xlarge = 40
+    cache.r3.large    = 2
+    cache.r3.xlarge   = 4
+    cache.r3.2xlarge  = 8
+    cache.r3.4xlarge  = 16
+    cache.r3.8xlarge  = 32
+    cache.r4.large    = 2
+    cache.r4.xlarge   = 4
+    cache.r4.2xlarge  = 8
+    cache.r4.4xlarge  = 16
+    cache.r4.8xlarge  = 32
+    cache.r4.16xlarge = 64
   }
 
   memory = {
-    cache.t2.micro    = "595926712"
-    cache.t2.small    = "1664299827"
-    cache.t2.medium   = "3457448673"
-    cache.m3.medium   = "2985002270"
-    cache.m3.large    = "6496138035"
-    cache.m3.xlarge   = "14280766259"
-    cache.m3.2xlarge  = "29957396889"
-    cache.m4.large    = "6893422510"
-    cache.m4.xlarge   = "15333033246"
-    cache.m4.2xlarge  = "31890132172"
-    cache.m4.4xlarge  = "65262028062"
-    cache.m4.10xlarge = "166043435663"
-    cache.r3.large    = "14495514624"
-    cache.r3.xlarge   = "30494267801"
-    cache.r3.2xlarge  = "62491774156"
-    cache.r3.4xlarge  = "126701535232"
-    cache.r3.8xlarge  = "254476812288"
-    cache.r4.large    = "13207024435"
-    cache.r4.xlarge   = "26897232691"
-    cache.r4.2xlarge  = "54191749857"
-    cache.r4.4xlarge  = "108855946117"
-    cache.r4.8xlarge  = "218248763146"
-    cache.r4.16xlarge = "437012922368"
+    cache.t2.micro    = 595926712
+    cache.t2.small    = 1664299827
+    cache.t2.medium   = 3457448673
+    cache.m3.medium   = 2985002270
+    cache.m3.large    = 6496138035
+    cache.m3.xlarge   = 14280766259
+    cache.m3.2xlarge  = 29957396889
+    cache.m4.large    = 6893422510
+    cache.m4.xlarge   = 15333033246
+    cache.m4.2xlarge  = 31890132172
+    cache.m4.4xlarge  = 65262028062
+    cache.m4.10xlarge = 166043435663
+    cache.r3.large    = 14495514624
+    cache.r3.xlarge   = 30494267801
+    cache.r3.2xlarge  = 62491774156
+    cache.r3.4xlarge  = 126701535232
+    cache.r3.8xlarge  = 254476812288
+    cache.r4.large    = 13207024435
+    cache.r4.xlarge   = 26897232691
+    cache.r4.2xlarge  = 54191749857
+    cache.r4.4xlarge  = 108855946117
+    cache.r4.8xlarge  = 218248763146
+    cache.r4.16xlarge = 437012922368
   }
 }
diff --git a/cloud/aws/elasticache/redis/monitors-redis.tf b/cloud/aws/elasticache/redis/monitors-redis.tf
index 8aa63dd..dddb799 100644
--- a/cloud/aws/elasticache/redis/monitors-redis.tf
+++ b/cloud/aws/elasticache/redis/monitors-redis.tf
@@ -53,20 +53,17 @@ resource "datadog_monitor" "redis_cpu_high" {
   name    = "[${var.environment}] Elasticache redis CPU {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
   message = "${coalesce(var.cpu_high_message, var.message)}"
 
+  count = "${length(keys(local.core))}"
+
   type = "metric alert"
 
   query = <<EOF
     ${var.cpu_high_time_aggregator}(${var.cpu_high_timeframe}): (
-      avg:aws.elasticache.cpuutilization{${data.template_file.filter.rendered}} by {region,cacheclusterid,cachenodeid}
-    ) > ( ${var.cpu_high_threshold_critical} / ${local.core[var.elasticache_size]} )
+      avg:aws.elasticache.cpuutilization{dd_monitoring:enabled,dd_aws_red:enabled,env:${var.environment},cache_node_type:${element(keys(local.core), count.index)}} by {region,cacheclusterid,cachenodeid}
+    ) > ${var.cpu_high_threshold_critical / element(values(local.core), count.index)}
   EOF
 
-  thresholds {
-    warning  = "${var.cpu_high_threshold_warning}"
-    critical = "${var.cpu_high_threshold_critical}"
-  }
-
-  notify_no_data      = true
+  notify_no_data      = false
   evaluation_delay    = "${var.delay}"
   renotify_interval   = 0
   notify_audit        = false