From e4e929ec1d6bb380da5eed54bd324ab33fc513ee Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?J=C3=A9r=C3=A9my=20NANCEL?= <jeremy.nancel@fr.clara.net>
Date: Mon, 30 Oct 2017 12:47:19 +0100
Subject: [PATCH] MON-78 Add datadog monitor for stream analytics

---
 cloud/azure/stream-analytics/inputs.tf        | 44 +++++++++
 .../monitors-stream-analytics.tf              | 92 +++++++++++++++++++
 2 files changed, 136 insertions(+)
 create mode 100644 cloud/azure/stream-analytics/inputs.tf
 create mode 100644 cloud/azure/stream-analytics/monitors-stream-analytics.tf

diff --git a/cloud/azure/stream-analytics/inputs.tf b/cloud/azure/stream-analytics/inputs.tf
new file mode 100644
index 0000000..e9bc507
--- /dev/null
+++ b/cloud/azure/stream-analytics/inputs.tf
@@ -0,0 +1,44 @@
+variable "hno_escalation_group" {}
+variable "ho_escalation_group" {}
+
+variable "environment" {}
+
+variable "notify_no_data" {
+  default = "false"
+}
+
+variable "delay" {
+  default = "600"
+}
+
+variable "su_utilization_warning" {
+  default = 60
+}
+
+variable "su_utilization_critical" {
+  default = 80
+}
+
+variable "failed_function_requests_warning" {
+  default = 0
+}
+
+variable "failed_function_requests_critical" {
+  default = 10
+}
+
+variable "conversion_errors_warning" {
+  default = 0
+}
+
+variable "conversion_errors_critical" {
+  default = 10
+}
+
+variable "runtime_errors_warning" {
+  default = 0
+}
+
+variable "runtime_errors_critical" {
+  default = 0
+}
diff --git a/cloud/azure/stream-analytics/monitors-stream-analytics.tf b/cloud/azure/stream-analytics/monitors-stream-analytics.tf
new file mode 100644
index 0000000..f18d7f1
--- /dev/null
+++ b/cloud/azure/stream-analytics/monitors-stream-analytics.tf
@@ -0,0 +1,92 @@
+resource "datadog_monitor" "SU_utilization" {
+  name    = "[${var.environment} SU utilization at more than ${var.su_utilization_critical}% on {{host.identifier}}]"
+  message = "{{#is_alert}}\n${var.hno_escalation_group} \n{{/is_alert}} \n{{#is_recovery}}\n${var.hno_escalation_group}\n{{/is_recovery}}"
+
+  query = "avg(last_5m):avg:azure.streamanalytics_streamingjobs.resource_utilization{*} by {name,resource_group} > ${var.su_utilization_critical}"
+  type  = "query alert"
+
+  notify_no_data      = "${var.notify_no_data}"
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 60
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+  new_host_delay      = "${var.delay}"
+  no_data_timeframe   = 20
+  thresholds {
+    warning  = "${var.su_utilization_warning}"
+    critical = "${var.su_utilization_critical}"
+  }
+}
+
+resource "datadog_monitor" "failed_function_requests" {
+  name    = "[${var.environment} More than ${var.failed_function_requests_critical} failed function requests on {{host.identifier}}]"
+  message = "{{#is_alert}}\n${var.hno_escalation_group} \n{{/is_alert}} \n{{#is_recovery}}\n${var.hno_escalation_group}\n{{/is_recovery}}"
+
+  query = "avg(last_5m):avg:azure.streamanalytics_streamingjobs.aml_callout_failed_requests{*} by {name,resource_group} > ${var.failed_function_requests_critical}"
+  type  = "query alert"
+
+  notify_no_data      = "${var.notify_no_data}"
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 60
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+  new_host_delay      = "${var.delay}"
+  no_data_timeframe   = 20
+  thresholds {
+    warning  = "${var.failed_function_requests_warning}"
+    critical = "${var.failed_function_requests_critical}"
+  }
+}
+
+resource "datadog_monitor" "conversion_errors" {
+  name    = "[${var.environment} More than ${var.conversion_errors_critical} conversion errors on {{host.identifier}}]"
+  message = "{{#is_alert}}\n${var.hno_escalation_group} \n{{/is_alert}} \n{{#is_recovery}}\n${var.hno_escalation_group}\n{{/is_recovery}}"
+
+  query = "avg(last_5m):avg:azure.streamanalytics_streamingjobs.conversion_errors{*} by {name,resource_group} > ${var.conversion_errors_critical}"
+  type  = "query alert"
+
+  notify_no_data      = "${var.notify_no_data}"
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 60
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+  new_host_delay      = "${var.delay}"
+  no_data_timeframe   = 20
+  thresholds {
+    warning  = "${var.conversion_errors_warning}"
+    critical = "${var.conversion_errors_critical}"
+  }
+}
+
+resource "datadog_monitor" "runtime_errors" {
+  name    = "[${var.environment} More than ${var.runtime_errors_critical} runtime errors on {{host.identifier}}]"
+  message = "{{#is_alert}}\n${var.hno_escalation_group} \n{{/is_alert}} \n{{#is_recovery}}\n${var.hno_escalation_group}\n{{/is_recovery}}"
+
+  query = "avg(last_5m):avg:azure.streamanalytics_streamingjobs.errors{*} by {name,resource_group} > ${var.runtime_errors_critical}"
+  type  = "query alert"
+
+  notify_no_data      = "${var.notify_no_data}"
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 60
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+  new_host_delay      = "${var.delay}"
+  no_data_timeframe   = 20
+  thresholds {
+    warning  = "${var.runtime_errors_warning}"
+    critical = "${var.runtime_errors_critical}"
+  }
+}
+