From 1c00f748d6e75585d313c7ea334d5849836295e6 Mon Sep 17 00:00:00 2001
From: Alex Lemaresquier <alexandre.lemaresquier@fr.clara.net>
Date: Fri, 20 Jul 2018 23:09:38 +0200
Subject: [PATCH 01/12] MON-248-monitors-for-nginx-ingress-contr: add Nginx
 ingress controller 5xx monitor

---
 README.md                            |  3 ++
 caas/k8s/ingress/inputs.tf           | 63 ++++++++++++++++++++++++++++
 caas/k8s/ingress/monitors-ingress.tf | 42 +++++++++++++++++++
 3 files changed, 108 insertions(+)
 create mode 100644 caas/k8s/ingress/inputs.tf
 create mode 100644 caas/k8s/ingress/monitors-ingress.tf

diff --git a/README.md b/README.md
index 7090482..3be9395 100644
--- a/README.md
+++ b/README.md
@@ -72,6 +72,9 @@ The `//` is very important, it's a terraform specific syntax used to separate gi
 
 ### Monitors summary ###
 
+- [caas](https://bitbucket.org/morea/terraform.feature.datadog/src/master/caas/)
+	- [k8s](https://bitbucket.org/morea/terraform.feature.datadog/src/master/caas/k8s/)
+		- [ingress](https://bitbucket.org/morea/terraform.feature.datadog/src/master/caas/k8s/ingress/)
 - [cloud](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/)
 	- [aws](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/aws/)
 		- [alb](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/aws/alb/)
diff --git a/caas/k8s/ingress/inputs.tf b/caas/k8s/ingress/inputs.tf
new file mode 100644
index 0000000..87e229a
--- /dev/null
+++ b/caas/k8s/ingress/inputs.tf
@@ -0,0 +1,63 @@
+# Global Terraform
+variable "team" {
+  type    = "string"
+  default = "k8s"
+}
+
+variable "environment" {
+  description = "Architecture Environment"
+  type        = "string"
+}
+
+# Global DataDog
+variable "delay" {
+  description = "Delay in seconds for the metric evaluation"
+  default     = 900
+}
+
+variable "message" {
+  description = "Message sent when an alert is triggered"
+}
+
+variable "filter_tags_use_defaults" {
+  description = "Use default filter tags convention"
+  default     = "true"
+}
+
+variable "filter_tags_custom" {
+  description = "Tags used for custom filtering when filter_tags_use_defaults is false"
+  default     = "*"
+}
+
+#Ingress
+variable "ingress_5xx_silenced" {
+  description = "Groups to mute for Ingress 5xx errors monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "ingress_5xx_message" {
+  description = "Message sent when an alert is triggered"
+  default     = ""
+}
+
+variable "ingress_5xx_timeframe" {
+  description = "Monitor timeframe for Ingress 5xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
+variable "ingress_5xx_threshold_critical" {
+  type    = "string"
+  default = "10"
+}
+
+variable "ingress_5xx_threshold_warning" {
+  type    = "string"
+  default = "5"
+}
+
+variable "artificial_requests_count" {
+  default     = 5
+  description = "Number of false requests used to mitigate false positive in case of low trafic"
+}
diff --git a/caas/k8s/ingress/monitors-ingress.tf b/caas/k8s/ingress/monitors-ingress.tf
new file mode 100644
index 0000000..fb4d64a
--- /dev/null
+++ b/caas/k8s/ingress/monitors-ingress.tf
@@ -0,0 +1,42 @@
+data "template_file" "filter" {
+  template = "$${filter}"
+
+  vars {
+    filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_k8s_ingress:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}"
+  }
+}
+
+resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
+  name    = "[${var.environment}] [${var.team}] Nginx Ingress 5xx errors too high for {{ingress_class.name}} on {{upstream.name}} {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
+  message = "${coalesce(var.ingress_5xx_message, var.message)}"
+
+  query = <<EOF
+    sum(${var.ingress_5xx_timeframe}): (
+      default(
+        avg:nginx_ingress.nginx_upstream_responses_total{status_code:5xx,${data.template_file.filter.rendered},!upstream:upstream-default-backend} by {upstream,ingress_class} /
+        (avg:nginx_ingress.nginx_upstream_requests_total{${data.template_file.filter.rendered},!upstream:upstream-default-backend} by {upstream,ingress_class} + ${var.artificial_requests_count}),
+      0) * 100
+    ) > ${var.ingress_5xx_threshold_critical}
+  EOF
+
+  type = "metric alert"
+
+  thresholds {
+    warning  = "${var.ingress_5xx_threshold_warning}"
+    critical = "${var.ingress_5xx_threshold_critical}"
+  }
+
+  notify_no_data      = false
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = false
+  new_host_delay      = "${var.delay}"
+
+  silenced = "${var.ingress_5xx_silenced}"
+
+  tags = ["env:${var.environment}", "resource:ingress", "team:${var.team}", "provider:k8s"]
+}

From 40f25abcabfa6cc2dcd7d8c9b981971e01f4fa6b Mon Sep 17 00:00:00 2001
From: Alex Lemaresquier <alexandre.lemaresquier@fr.clara.net>
Date: Thu, 30 Aug 2018 16:30:42 +0200
Subject: [PATCH 02/12] MON-248-monitors-for-nginx-ingress-contr: add README.md

---
 caas/k8s/ingress/README.md | 59 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 59 insertions(+)
 create mode 100644 caas/k8s/ingress/README.md

diff --git a/caas/k8s/ingress/README.md b/caas/k8s/ingress/README.md
new file mode 100644
index 0000000..87c2949
--- /dev/null
+++ b/caas/k8s/ingress/README.md
@@ -0,0 +1,59 @@
+Inputs
+------
+
+| Name | Description | Type | Default | Required |
+|------|-------------|:----:|:-----:|:-----:|
+| artificial_requests_count | Number of false requests used to mitigate false positive in case of low trafic | string | `5` | no |
+| delay | Delay in seconds for the metric evaluation | string | `900` | no |
+| environment | Architecture Environment | string | - | yes |
+| filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
+| filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
+| ingress_5xx_message | Message sent when an alert is triggered | string | `` | no |
+| ingress_5xx_silenced | Groups to mute for Ingress 5xx errors monitor | map | `<map>` | no |
+| ingress_5xx_threshold_critical |  | string | `10` | no |
+| ingress_5xx_threshold_warning |  | string | `5` | no |
+| ingress_5xx_timeframe | Monitor timeframe for Ingress 5xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
+| message | Message sent when an alert is triggered | string | - | yes |
+| team | Global Terraform | string | `k8s` | no |
+
+Outputs
+-------
+
+| Name | Description |
+|------|-------------|
+| Nginx_ingress_too_many_5xx_id | id for monitor Nginx_ingress_too_many_5xx |
+
+Related documentation
+---------------------
+
+DataDog blog: https://www.datadoghq.com/blog/monitor-prometheus-metrics
+https://github.com/kubernetes/ingress-nginx/pull/423/commits/1d38e3a38425f08de2f75fcae13896a3fec4d144
+
+Nginx Ingress Controller setup
+------------------------------
+Enable the following flags in the Nginx Ingress Controller chart
+controller.stats.enabled=true,controller.metrics.enabled=true
+and the following Datadog agent configuration for each ingress controller:
+```
+datadog:
+  confd:
+    prometheus.yaml: |-
+      #nginx_upstream_responses_total{ingress_class,namespace,server,status_code:{1xx,2xx,3xx,4xx,5xx},upstream}
+      #nginx_upstream_requests_total{ingress_class,namespace,server,upstream}
+      init_config:
+      instances:
+        # The prometheus endpoint to query from
+        - prometheus_url: http://nginx-ingress-controller-metrics:9913/metrics
+          # This is NOT the ingress namespace, it is the prefix that will be used for the custom metrics
+          namespace: nginx-ingress
+          # Filter on the following metrics only
+          metrics:
+            - "nginx_upstream_requests_total"
+            - "nginx_upstream_responses_total"
+          # Adapt the tags to the current convention and verify that the monitor will match
+          tags:
+              - dd_monitoring:enabled
+              - dd_k8s_ingress:enabled
+              - dd_ingress_class:nginx
+              - env:prod
+```

From e2a583670d7e40af7b6bd17e7171373225c33e66 Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Thu, 30 Aug 2018 17:55:25 +0200
Subject: [PATCH 03/12] MON-248 update readme

---
 caas/k8s/ingress/README.md | 26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)

diff --git a/caas/k8s/ingress/README.md b/caas/k8s/ingress/README.md
index 87c2949..09284b6 100644
--- a/caas/k8s/ingress/README.md
+++ b/caas/k8s/ingress/README.md
@@ -1,5 +1,24 @@
-Inputs
-------
+# CAAS K8S INGRESS DataDog monitors
+
+## How to use this module
+
+```
+module "datadog-monitors-caas-k8s-ingress" {
+  source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//caas/k8s/ingress?ref={revision}"
+
+  environment = "${var.environment}"
+  message     = "${module.datadog-message-alerting.alerting-message}"
+}
+
+```
+
+## Purpose
+
+Creates DataDog monitors with the following checks:
+
+- Nginx Ingress 5xx errors too high for {{ingress_class.name}} on {{upstream.name}}
+
+## Inputs
 
 | Name | Description | Type | Default | Required |
 |------|-------------|:----:|:-----:|:-----:|
@@ -16,8 +35,7 @@ Inputs
 | message | Message sent when an alert is triggered | string | - | yes |
 | team | Global Terraform | string | `k8s` | no |
 
-Outputs
--------
+## Outputs
 
 | Name | Description |
 |------|-------------|

From ab2647ea9cff6b594456458afb731cdd18ef88df Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Thu, 30 Aug 2018 17:55:45 +0200
Subject: [PATCH 04/12] MON-248 add outputs

---
 caas/k8s/ingress/outputs.tf | 4 ++++
 1 file changed, 4 insertions(+)
 create mode 100644 caas/k8s/ingress/outputs.tf

diff --git a/caas/k8s/ingress/outputs.tf b/caas/k8s/ingress/outputs.tf
new file mode 100644
index 0000000..cacbec3
--- /dev/null
+++ b/caas/k8s/ingress/outputs.tf
@@ -0,0 +1,4 @@
+output "Nginx_ingress_too_many_5xx_id" {
+  description = "id for monitor Nginx_ingress_too_many_5xx"
+  value       = "${datadog_monitor.Nginx_ingress_too_many_5xx.*.id}"
+}

From 0766bce61eb4895743c2b8f8011da6f854d5fc52 Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Thu, 30 Aug 2018 18:03:46 +0200
Subject: [PATCH 05/12] MON-248 use filter tags modules

---
 caas/k8s/ingress/modules.tf          | 19 +++++++++++++++++++
 caas/k8s/ingress/monitors-ingress.tf |  4 ++--
 2 files changed, 21 insertions(+), 2 deletions(-)
 create mode 100644 caas/k8s/ingress/modules.tf

diff --git a/caas/k8s/ingress/modules.tf b/caas/k8s/ingress/modules.tf
new file mode 100644
index 0000000..8cf5cd7
--- /dev/null
+++ b/caas/k8s/ingress/modules.tf
@@ -0,0 +1,19 @@
+module "filter-tags" {
+  source = "../../../common/filter-tags"
+
+  environment              = "${var.environment}"
+  resource                 = "ingress"
+  filter_tags_use_defaults = "${var.filter_tags_use_defaults}"
+  filter_tags_custom       = "${var.filter_tags_custom}"
+  extra_tags               = ["!upstream:upstream-default-backend"]
+}
+
+module "filter-tags-5xx" {
+  source = "../../../common/filter-tags"
+
+  environment              = "${var.environment}"
+  resource                 = "ingress"
+  filter_tags_use_defaults = "${var.filter_tags_use_defaults}"
+  filter_tags_custom       = "${var.filter_tags_custom}"
+  extra_tags               = ["!upstream:upstream-default-backend,status_code:5xx"]
+}
diff --git a/caas/k8s/ingress/monitors-ingress.tf b/caas/k8s/ingress/monitors-ingress.tf
index fb4d64a..9b23ed3 100644
--- a/caas/k8s/ingress/monitors-ingress.tf
+++ b/caas/k8s/ingress/monitors-ingress.tf
@@ -13,8 +13,8 @@ resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
   query = <<EOF
     sum(${var.ingress_5xx_timeframe}): (
       default(
-        avg:nginx_ingress.nginx_upstream_responses_total{status_code:5xx,${data.template_file.filter.rendered},!upstream:upstream-default-backend} by {upstream,ingress_class} /
-        (avg:nginx_ingress.nginx_upstream_requests_total{${data.template_file.filter.rendered},!upstream:upstream-default-backend} by {upstream,ingress_class} + ${var.artificial_requests_count}),
+        avg:nginx_ingress.nginx_upstream_responses_total${module.filter-tags-5xx.query_alert} by {upstream,ingress_class} /
+        (avg:nginx_ingress.nginx_upstream_requests_total${module.filter-tags.query_alert} by {upstream,ingress_class} + ${var.artificial_requests_count}),
       0) * 100
     ) > ${var.ingress_5xx_threshold_critical}
   EOF

From 048e3a8338b949a12d22683799a0b1008d1e613e Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Thu, 30 Aug 2018 18:05:31 +0200
Subject: [PATCH 06/12] MON-248 split delay

---
 caas/k8s/ingress/inputs.tf           | 9 +++++++--
 caas/k8s/ingress/monitors-ingress.tf | 4 ++--
 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/caas/k8s/ingress/inputs.tf b/caas/k8s/ingress/inputs.tf
index 87e229a..5d94e11 100644
--- a/caas/k8s/ingress/inputs.tf
+++ b/caas/k8s/ingress/inputs.tf
@@ -10,9 +10,14 @@ variable "environment" {
 }
 
 # Global DataDog
-variable "delay" {
+variable "evaluation_delay" {
   description = "Delay in seconds for the metric evaluation"
-  default     = 900
+  default     = 15
+}
+
+variable "new_host_delay" {
+  description = "Delay in seconds before monitor new resource"
+  default     = 300
 }
 
 variable "message" {
diff --git a/caas/k8s/ingress/monitors-ingress.tf b/caas/k8s/ingress/monitors-ingress.tf
index 9b23ed3..9327473 100644
--- a/caas/k8s/ingress/monitors-ingress.tf
+++ b/caas/k8s/ingress/monitors-ingress.tf
@@ -27,14 +27,14 @@ resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
   }
 
   notify_no_data      = false
-  evaluation_delay    = "${var.delay}"
+  evaluation_delay    = "${var.evaluation_delay}"
+  new_host_delay      = "${var.new_host_delay}"
   renotify_interval   = 0
   notify_audit        = false
   timeout_h           = 0
   include_tags        = true
   locked              = false
   require_full_window = false
-  new_host_delay      = "${var.delay}"
 
   silenced = "${var.ingress_5xx_silenced}"
 

From cf09393abbedc0236b694fd1c22335c9aa23acd1 Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Thu, 30 Aug 2018 18:06:14 +0200
Subject: [PATCH 07/12] MON-248 auto update

---
 caas/k8s/ingress/README.md | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/caas/k8s/ingress/README.md b/caas/k8s/ingress/README.md
index 09284b6..deac474 100644
--- a/caas/k8s/ingress/README.md
+++ b/caas/k8s/ingress/README.md
@@ -23,8 +23,8 @@ Creates DataDog monitors with the following checks:
 | Name | Description | Type | Default | Required |
 |------|-------------|:----:|:-----:|:-----:|
 | artificial_requests_count | Number of false requests used to mitigate false positive in case of low trafic | string | `5` | no |
-| delay | Delay in seconds for the metric evaluation | string | `900` | no |
 | environment | Architecture Environment | string | - | yes |
+| evaluation_delay | Delay in seconds for the metric evaluation | string | `15` | no |
 | filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
 | filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
 | ingress_5xx_message | Message sent when an alert is triggered | string | `` | no |
@@ -33,6 +33,7 @@ Creates DataDog monitors with the following checks:
 | ingress_5xx_threshold_warning |  | string | `5` | no |
 | ingress_5xx_timeframe | Monitor timeframe for Ingress 5xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | message | Message sent when an alert is triggered | string | - | yes |
+| new_host_delay | Delay in seconds before monitor new resource | string | `300` | no |
 | team | Global Terraform | string | `k8s` | no |
 
 ## Outputs

From 0433c7dabcf62f96b533462d77d4c5b77ce14732 Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Thu, 30 Aug 2018 18:10:34 +0200
Subject: [PATCH 08/12] MON-248 new tag convention and extra tag and disable
 feature

---
 caas/k8s/ingress/inputs.tf           | 12 ++++++++++++
 caas/k8s/ingress/monitors-ingress.tf | 11 ++---------
 2 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/caas/k8s/ingress/inputs.tf b/caas/k8s/ingress/inputs.tf
index 5d94e11..b96012e 100644
--- a/caas/k8s/ingress/inputs.tf
+++ b/caas/k8s/ingress/inputs.tf
@@ -41,6 +41,18 @@ variable "ingress_5xx_silenced" {
   default     = {}
 }
 
+variable "ingress_5xx_enabled" {
+  description = "Flag to enable Ingress 5xx errors monitor"
+  type        = "string"
+  default     = "true"
+}
+
+variable "ingress_5xx_extra_tags" {
+  description = "Extra tags for Ingress 5xx errors monitor"
+  type        = "list"
+  default     = []
+}
+
 variable "ingress_5xx_message" {
   description = "Message sent when an alert is triggered"
   default     = ""
diff --git a/caas/k8s/ingress/monitors-ingress.tf b/caas/k8s/ingress/monitors-ingress.tf
index 9327473..25c7977 100644
--- a/caas/k8s/ingress/monitors-ingress.tf
+++ b/caas/k8s/ingress/monitors-ingress.tf
@@ -1,12 +1,5 @@
-data "template_file" "filter" {
-  template = "$${filter}"
-
-  vars {
-    filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_k8s_ingress:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}"
-  }
-}
-
 resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
+  count   = "${var.ingress_5xx_enabled ? 1 : 0}"
   name    = "[${var.environment}] [${var.team}] Nginx Ingress 5xx errors too high for {{ingress_class.name}} on {{upstream.name}} {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
   message = "${coalesce(var.ingress_5xx_message, var.message)}"
 
@@ -38,5 +31,5 @@ resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
 
   silenced = "${var.ingress_5xx_silenced}"
 
-  tags = ["env:${var.environment}", "resource:ingress", "team:${var.team}", "provider:k8s"]
+  tags = ["env:${var.environment}", "type:caas", "provider:prometheus", "resource:nginx-ingress-controller", "team:claranet", "created-by:terraform", "${var.ingress_5xx_extra_tags}"]
 }

From af1cb3f9ffb77adb51914258d83d2733ed8eb56d Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Thu, 30 Aug 2018 18:11:20 +0200
Subject: [PATCH 09/12] MON-248 auto update

---
 caas/k8s/ingress/README.md | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/caas/k8s/ingress/README.md b/caas/k8s/ingress/README.md
index deac474..819248f 100644
--- a/caas/k8s/ingress/README.md
+++ b/caas/k8s/ingress/README.md
@@ -27,6 +27,8 @@ Creates DataDog monitors with the following checks:
 | evaluation_delay | Delay in seconds for the metric evaluation | string | `15` | no |
 | filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
 | filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
+| ingress_5xx_enabled | Flag to enable Ingress 5xx errors monitor | string | `true` | no |
+| ingress_5xx_extra_tags | Extra tags for Ingress 5xx errors monitor | list | `<list>` | no |
 | ingress_5xx_message | Message sent when an alert is triggered | string | `` | no |
 | ingress_5xx_silenced | Groups to mute for Ingress 5xx errors monitor | map | `<map>` | no |
 | ingress_5xx_threshold_critical |  | string | `10` | no |

From c7285c7a0c793f949438a820040f3a6175d2ce84 Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Thu, 30 Aug 2018 18:12:03 +0200
Subject: [PATCH 10/12] MON-248 require full window to true

---
 caas/k8s/ingress/monitors-ingress.tf | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/caas/k8s/ingress/monitors-ingress.tf b/caas/k8s/ingress/monitors-ingress.tf
index 25c7977..ee472bc 100644
--- a/caas/k8s/ingress/monitors-ingress.tf
+++ b/caas/k8s/ingress/monitors-ingress.tf
@@ -27,7 +27,7 @@ resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
   timeout_h           = 0
   include_tags        = true
   locked              = false
-  require_full_window = false
+  require_full_window = true
 
   silenced = "${var.ingress_5xx_silenced}"
 

From 881a0fe78e7d5e515251c4b09a901c31eed8a78c Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Fri, 31 Aug 2018 13:07:17 +0200
Subject: [PATCH 11/12] MON-248 rename k8s to kubernetes

---
 README.md                                            |  4 ++--
 caas/{k8s => kubernetes}/ingress/README.md           | 11 +++++------
 caas/{k8s => kubernetes}/ingress/inputs.tf           |  5 -----
 caas/{k8s => kubernetes}/ingress/modules.tf          |  0
 caas/{k8s => kubernetes}/ingress/monitors-ingress.tf |  2 +-
 caas/{k8s => kubernetes}/ingress/outputs.tf          |  0
 6 files changed, 8 insertions(+), 14 deletions(-)
 rename caas/{k8s => kubernetes}/ingress/README.md (91%)
 rename caas/{k8s => kubernetes}/ingress/inputs.tf (96%)
 rename caas/{k8s => kubernetes}/ingress/modules.tf (100%)
 rename caas/{k8s => kubernetes}/ingress/monitors-ingress.tf (81%)
 rename caas/{k8s => kubernetes}/ingress/outputs.tf (100%)

diff --git a/README.md b/README.md
index 3be9395..4efcd0c 100644
--- a/README.md
+++ b/README.md
@@ -73,8 +73,8 @@ The `//` is very important, it's a terraform specific syntax used to separate gi
 ### Monitors summary ###
 
 - [caas](https://bitbucket.org/morea/terraform.feature.datadog/src/master/caas/)
-	- [k8s](https://bitbucket.org/morea/terraform.feature.datadog/src/master/caas/k8s/)
-		- [ingress](https://bitbucket.org/morea/terraform.feature.datadog/src/master/caas/k8s/ingress/)
+	- [kubernetes](https://bitbucket.org/morea/terraform.feature.datadog/src/master/caas/kubernetes/)
+		- [ingress](https://bitbucket.org/morea/terraform.feature.datadog/src/master/caas/kubernetes/ingress/)
 - [cloud](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/)
 	- [aws](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/aws/)
 		- [alb](https://bitbucket.org/morea/terraform.feature.datadog/src/master/cloud/aws/alb/)
diff --git a/caas/k8s/ingress/README.md b/caas/kubernetes/ingress/README.md
similarity index 91%
rename from caas/k8s/ingress/README.md
rename to caas/kubernetes/ingress/README.md
index 819248f..7ba4994 100644
--- a/caas/k8s/ingress/README.md
+++ b/caas/kubernetes/ingress/README.md
@@ -1,10 +1,10 @@
-# CAAS K8S INGRESS DataDog monitors
+# CAAS KUBERNETES INGRESS DataDog monitors
 
 ## How to use this module
 
 ```
-module "datadog-monitors-caas-k8s-ingress" {
-  source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//caas/k8s/ingress?ref={revision}"
+module "datadog-monitors-caas-kubernetes-ingress" {
+  source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//caas/kubernetes/ingress?ref={revision}"
 
   environment = "${var.environment}"
   message     = "${module.datadog-message-alerting.alerting-message}"
@@ -16,7 +16,7 @@ module "datadog-monitors-caas-k8s-ingress" {
 
 Creates DataDog monitors with the following checks:
 
-- Nginx Ingress 5xx errors too high for {{ingress_class.name}} on {{upstream.name}}
+- Nginx Ingress 5xx errors
 
 ## Inputs
 
@@ -36,7 +36,6 @@ Creates DataDog monitors with the following checks:
 | ingress_5xx_timeframe | Monitor timeframe for Ingress 5xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | message | Message sent when an alert is triggered | string | - | yes |
 | new_host_delay | Delay in seconds before monitor new resource | string | `300` | no |
-| team | Global Terraform | string | `k8s` | no |
 
 ## Outputs
 
@@ -74,7 +73,7 @@ datadog:
           # Adapt the tags to the current convention and verify that the monitor will match
           tags:
               - dd_monitoring:enabled
-              - dd_k8s_ingress:enabled
+              - dd_ingress:enabled
               - dd_ingress_class:nginx
               - env:prod
 ```
diff --git a/caas/k8s/ingress/inputs.tf b/caas/kubernetes/ingress/inputs.tf
similarity index 96%
rename from caas/k8s/ingress/inputs.tf
rename to caas/kubernetes/ingress/inputs.tf
index b96012e..65ed192 100644
--- a/caas/k8s/ingress/inputs.tf
+++ b/caas/kubernetes/ingress/inputs.tf
@@ -1,9 +1,4 @@
 # Global Terraform
-variable "team" {
-  type    = "string"
-  default = "k8s"
-}
-
 variable "environment" {
   description = "Architecture Environment"
   type        = "string"
diff --git a/caas/k8s/ingress/modules.tf b/caas/kubernetes/ingress/modules.tf
similarity index 100%
rename from caas/k8s/ingress/modules.tf
rename to caas/kubernetes/ingress/modules.tf
diff --git a/caas/k8s/ingress/monitors-ingress.tf b/caas/kubernetes/ingress/monitors-ingress.tf
similarity index 81%
rename from caas/k8s/ingress/monitors-ingress.tf
rename to caas/kubernetes/ingress/monitors-ingress.tf
index ee472bc..2cce3c3 100644
--- a/caas/k8s/ingress/monitors-ingress.tf
+++ b/caas/kubernetes/ingress/monitors-ingress.tf
@@ -1,6 +1,6 @@
 resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
   count   = "${var.ingress_5xx_enabled ? 1 : 0}"
-  name    = "[${var.environment}] [${var.team}] Nginx Ingress 5xx errors too high for {{ingress_class.name}} on {{upstream.name}} {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
+  name    = "[${var.environment}] Nginx Ingress 5xx errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
   message = "${coalesce(var.ingress_5xx_message, var.message)}"
 
   query = <<EOF
diff --git a/caas/k8s/ingress/outputs.tf b/caas/kubernetes/ingress/outputs.tf
similarity index 100%
rename from caas/k8s/ingress/outputs.tf
rename to caas/kubernetes/ingress/outputs.tf

From efaa611202c8835813d997b34bdfade54ed26ccb Mon Sep 17 00:00:00 2001
From: Quentin Manfroi <quentin.manfroi@fr.clara.net>
Date: Fri, 31 Aug 2018 13:25:34 +0200
Subject: [PATCH 12/12] MON-248 add 4xx monitor

---
 caas/kubernetes/ingress/README.md           | 15 +++++--
 caas/kubernetes/ingress/inputs.tf           | 43 ++++++++++++++++++++-
 caas/kubernetes/ingress/modules.tf          | 10 +++++
 caas/kubernetes/ingress/monitors-ingress.tf | 38 +++++++++++++++++-
 caas/kubernetes/ingress/outputs.tf          | 11 ++++--
 5 files changed, 108 insertions(+), 9 deletions(-)

diff --git a/caas/kubernetes/ingress/README.md b/caas/kubernetes/ingress/README.md
index 7ba4994..0a7ba2a 100644
--- a/caas/kubernetes/ingress/README.md
+++ b/caas/kubernetes/ingress/README.md
@@ -16,6 +16,7 @@ module "datadog-monitors-caas-kubernetes-ingress" {
 
 Creates DataDog monitors with the following checks:
 
+- Nginx Ingress 4xx errors
 - Nginx Ingress 5xx errors
 
 ## Inputs
@@ -27,12 +28,19 @@ Creates DataDog monitors with the following checks:
 | evaluation_delay | Delay in seconds for the metric evaluation | string | `15` | no |
 | filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
 | filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
+| ingress_4xx_enabled | Flag to enable Ingress 4xx errors monitor | string | `true` | no |
+| ingress_4xx_extra_tags | Extra tags for Ingress 4xx errors monitor | list | `<list>` | no |
+| ingress_4xx_message | Message sent when an alert is triggered | string | `` | no |
+| ingress_4xx_silenced | Groups to mute for Ingress 4xx errors monitor | map | `<map>` | no |
+| ingress_4xx_threshold_critical |  | string | `40` | no |
+| ingress_4xx_threshold_warning |  | string | `20` | no |
+| ingress_4xx_timeframe | Monitor timeframe for Ingress 4xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | ingress_5xx_enabled | Flag to enable Ingress 5xx errors monitor | string | `true` | no |
 | ingress_5xx_extra_tags | Extra tags for Ingress 5xx errors monitor | list | `<list>` | no |
 | ingress_5xx_message | Message sent when an alert is triggered | string | `` | no |
 | ingress_5xx_silenced | Groups to mute for Ingress 5xx errors monitor | map | `<map>` | no |
-| ingress_5xx_threshold_critical |  | string | `10` | no |
-| ingress_5xx_threshold_warning |  | string | `5` | no |
+| ingress_5xx_threshold_critical |  | string | `20` | no |
+| ingress_5xx_threshold_warning |  | string | `10` | no |
 | ingress_5xx_timeframe | Monitor timeframe for Ingress 5xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | message | Message sent when an alert is triggered | string | - | yes |
 | new_host_delay | Delay in seconds before monitor new resource | string | `300` | no |
@@ -41,7 +49,8 @@ Creates DataDog monitors with the following checks:
 
 | Name | Description |
 |------|-------------|
-| Nginx_ingress_too_many_5xx_id | id for monitor Nginx_ingress_too_many_5xx |
+| nginx_ingress_too_many_4xx_id | id for monitor nginx_ingress_too_many_4xx |
+| nginx_ingress_too_many_5xx_id | id for monitor nginx_ingress_too_many_5xx |
 
 Related documentation
 ---------------------
diff --git a/caas/kubernetes/ingress/inputs.tf b/caas/kubernetes/ingress/inputs.tf
index 65ed192..7be1590 100644
--- a/caas/kubernetes/ingress/inputs.tf
+++ b/caas/kubernetes/ingress/inputs.tf
@@ -61,12 +61,51 @@ variable "ingress_5xx_timeframe" {
 
 variable "ingress_5xx_threshold_critical" {
   type    = "string"
-  default = "10"
+  default = "20"
 }
 
 variable "ingress_5xx_threshold_warning" {
   type    = "string"
-  default = "5"
+  default = "10"
+}
+
+variable "ingress_4xx_silenced" {
+  description = "Groups to mute for Ingress 4xx errors monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "ingress_4xx_enabled" {
+  description = "Flag to enable Ingress 4xx errors monitor"
+  type        = "string"
+  default     = "true"
+}
+
+variable "ingress_4xx_extra_tags" {
+  description = "Extra tags for Ingress 4xx errors monitor"
+  type        = "list"
+  default     = []
+}
+
+variable "ingress_4xx_message" {
+  description = "Message sent when an alert is triggered"
+  default     = ""
+}
+
+variable "ingress_4xx_timeframe" {
+  description = "Monitor timeframe for Ingress 4xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
+variable "ingress_4xx_threshold_critical" {
+  type    = "string"
+  default = "40"
+}
+
+variable "ingress_4xx_threshold_warning" {
+  type    = "string"
+  default = "20"
 }
 
 variable "artificial_requests_count" {
diff --git a/caas/kubernetes/ingress/modules.tf b/caas/kubernetes/ingress/modules.tf
index 8cf5cd7..ee1e148 100644
--- a/caas/kubernetes/ingress/modules.tf
+++ b/caas/kubernetes/ingress/modules.tf
@@ -17,3 +17,13 @@ module "filter-tags-5xx" {
   filter_tags_custom       = "${var.filter_tags_custom}"
   extra_tags               = ["!upstream:upstream-default-backend,status_code:5xx"]
 }
+
+module "filter-tags-4xx" {
+  source = "../../../common/filter-tags"
+
+  environment              = "${var.environment}"
+  resource                 = "ingress"
+  filter_tags_use_defaults = "${var.filter_tags_use_defaults}"
+  filter_tags_custom       = "${var.filter_tags_custom}"
+  extra_tags               = ["!upstream:upstream-default-backend,status_code:4xx"]
+}
diff --git a/caas/kubernetes/ingress/monitors-ingress.tf b/caas/kubernetes/ingress/monitors-ingress.tf
index 2cce3c3..ab44196 100644
--- a/caas/kubernetes/ingress/monitors-ingress.tf
+++ b/caas/kubernetes/ingress/monitors-ingress.tf
@@ -1,4 +1,4 @@
-resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
+resource "datadog_monitor" "nginx_ingress_too_many_5xx" {
   count   = "${var.ingress_5xx_enabled ? 1 : 0}"
   name    = "[${var.environment}] Nginx Ingress 5xx errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
   message = "${coalesce(var.ingress_5xx_message, var.message)}"
@@ -33,3 +33,39 @@ resource "datadog_monitor" "Nginx_ingress_too_many_5xx" {
 
   tags = ["env:${var.environment}", "type:caas", "provider:prometheus", "resource:nginx-ingress-controller", "team:claranet", "created-by:terraform", "${var.ingress_5xx_extra_tags}"]
 }
+
+resource "datadog_monitor" "nginx_ingress_too_many_4xx" {
+  count   = "${var.ingress_4xx_enabled ? 1 : 0}"
+  name    = "[${var.environment}] Nginx Ingress 4xx errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
+  message = "${coalesce(var.ingress_4xx_message, var.message)}"
+
+  query = <<EOF
+    sum(${var.ingress_4xx_timeframe}): (
+      default(
+        avg:nginx_ingress.nginx_upstream_responses_total${module.filter-tags-4xx.query_alert} by {upstream,ingress_class} /
+        (avg:nginx_ingress.nginx_upstream_requests_total${module.filter-tags.query_alert} by {upstream,ingress_class} + ${var.artificial_requests_count}),
+      0) * 100
+    ) > ${var.ingress_4xx_threshold_critical}
+  EOF
+
+  type = "metric alert"
+
+  thresholds {
+    warning  = "${var.ingress_4xx_threshold_warning}"
+    critical = "${var.ingress_4xx_threshold_critical}"
+  }
+
+  notify_no_data      = false
+  evaluation_delay    = "${var.evaluation_delay}"
+  new_host_delay      = "${var.new_host_delay}"
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+
+  silenced = "${var.ingress_4xx_silenced}"
+
+  tags = ["env:${var.environment}", "type:caas", "provider:prometheus", "resource:nginx-ingress-controller", "team:claranet", "created-by:terraform", "${var.ingress_4xx_extra_tags}"]
+}
diff --git a/caas/kubernetes/ingress/outputs.tf b/caas/kubernetes/ingress/outputs.tf
index cacbec3..28582f8 100644
--- a/caas/kubernetes/ingress/outputs.tf
+++ b/caas/kubernetes/ingress/outputs.tf
@@ -1,4 +1,9 @@
-output "Nginx_ingress_too_many_5xx_id" {
-  description = "id for monitor Nginx_ingress_too_many_5xx"
-  value       = "${datadog_monitor.Nginx_ingress_too_many_5xx.*.id}"
+output "nginx_ingress_too_many_5xx_id" {
+  description = "id for monitor nginx_ingress_too_many_5xx"
+  value       = "${datadog_monitor.nginx_ingress_too_many_5xx.*.id}"
+}
+
+output "nginx_ingress_too_many_4xx_id" {
+  description = "id for monitor nginx_ingress_too_many_4xx"
+  value       = "${datadog_monitor.nginx_ingress_too_many_4xx.*.id}"
 }