From d2e1aa5efddea62258790c9b0afff8dea0d51cf4 Mon Sep 17 00:00:00 2001
From: Laurent Piroelle <laurent.piroelle@fr.clara.net>
Date: Mon, 30 Oct 2017 11:40:16 +0100
Subject: [PATCH] MON-77 Azure Event Hub monitors

---
 cloud/azure/eventhub/inputs.tf            | 31 ++++++++
 cloud/azure/eventhub/monitors-eventhub.tf | 86 +++++++++++++++++++++++
 cloud/azure/eventhub/outputs.tf           | 11 +++
 3 files changed, 128 insertions(+)
 create mode 100644 cloud/azure/eventhub/inputs.tf
 create mode 100644 cloud/azure/eventhub/monitors-eventhub.tf
 create mode 100644 cloud/azure/eventhub/outputs.tf

diff --git a/cloud/azure/eventhub/inputs.tf b/cloud/azure/eventhub/inputs.tf
new file mode 100644
index 0000000..a1c7ec4
--- /dev/null
+++ b/cloud/azure/eventhub/inputs.tf
@@ -0,0 +1,31 @@
+variable "environment" {}
+
+variable "down_message" {}
+
+variable "failed_requests_message" {}
+
+variable "errors_message" {}
+
+variable "delay" {
+  default = 600
+}
+
+variable "failed_requests_rate_thresold_critical" {
+  default = 5
+}
+
+variable "failed_requests_rate_thresold_warning" {
+  default = 3
+}
+
+variable "errors_rate_thresold_critical" {
+  default = 5
+}
+
+variable "errors_rate_thresold_warning" {
+  default = 3
+}
+
+variable "use_filter_tags" {
+  default = "true"
+}
diff --git a/cloud/azure/eventhub/monitors-eventhub.tf b/cloud/azure/eventhub/monitors-eventhub.tf
new file mode 100644
index 0000000..7c22418
--- /dev/null
+++ b/cloud/azure/eventhub/monitors-eventhub.tf
@@ -0,0 +1,86 @@
+resource "datadog_monitor" "eventhub_status" {
+  name    = "[${var.environment}] Event Hub status"
+  message = "${var.down_message}"
+
+  query = <<EOF
+      avg(last_5m): avg:azure.eventhub_namespaces.status{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} != 1
+      EOF
+  type  = "query alert"
+
+  notify_no_data      = true
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 60
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+  new_host_delay      = "${var.delay}"
+  no_data_timeframe   = 20
+}
+
+resource "datadog_monitor" "eventhub_failed_requests" {
+  name    = "[${var.environment}] Event Hub failed requests"
+  message = "${var.failed_requests_message}"
+
+  query = <<EOF
+        avg(last_5m): (
+          avg:azure.eventhub_namespaces.failed_requests{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region}
+        ) * 100 / (
+          avg:azure.eventhub_namespaces.successful_requests{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
+          avg:azure.eventhub_namespaces.failed_requests{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region}
+        ) > ${var.failed_requests_rate_thresold_critical}
+        EOF
+  type  = "query alert"
+
+  thresholds {
+    critical = "${var.failed_requests_rate_thresold_critical}"
+    warning  = "${var.failed_requests_rate_thresold_warning}"
+  }
+
+  notify_no_data      = false
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 60
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+  new_host_delay      = "${var.delay}"
+  no_data_timeframe   = 20
+}
+
+resource "datadog_monitor" "eventhub_errors" {
+  name    = "[${var.environment}] Event Hub errors"
+  message = "${var.errors_message}"
+
+  query = <<EOF
+        avg(last_5m): (
+          avg:azure.eventhub_namespaces.internal_server_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
+          avg:azure.eventhub_namespaces.server_busy_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
+          avg:azure.eventhub_namespaces.other_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region}
+        ) * 100 / (
+          avg:azure.eventhub_namespaces.successful_requests{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
+          avg:azure.eventhub_namespaces.internal_server_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
+          avg:azure.eventhub_namespaces.server_busy_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region} +
+          avg:azure.eventhub_namespaces.other_errors{${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_eventhub:enabled,env:%s", var.environment) : "*"}} by {name,resource_group,region}
+        ) > ${var.errors_rate_thresold_critical}
+        EOF
+  type  = "query alert"
+
+  thresholds {
+    critical = "${var.errors_rate_thresold_critical}"
+    warning  = "${var.errors_rate_thresold_warning}"
+  }
+
+  notify_no_data      = false
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 60
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+  new_host_delay      = "${var.delay}"
+  no_data_timeframe   = 20
+}
diff --git a/cloud/azure/eventhub/outputs.tf b/cloud/azure/eventhub/outputs.tf
new file mode 100644
index 0000000..b9d1822
--- /dev/null
+++ b/cloud/azure/eventhub/outputs.tf
@@ -0,0 +1,11 @@
+output "status_monitor_id" {
+  value = "${datadog_monitor.eventhub_failed_requests.id}"
+}
+
+output "failed_requests_monitor_id" {
+  value = "${datadog_monitor.eventhub_status.id}"
+}
+
+output "errors_monitor_id" {
+  value = "${datadog_monitor.eventhub_errors.id}"
+}