MON-246 - Azure Databases for MySQL servers monitors added

This commit is contained in:
Alexandre Gaillet 2018-07-12 16:08:11 +02:00 committed by Quentin Manfroi
parent 9c3afce524
commit e78c665813
3 changed files with 420 additions and 0 deletions

View File

@ -0,0 +1,72 @@
Azure Database for MySQL servers DataDog monitors
=================================================
How to use this module
----------------------
```
module "datadog-monitors-azure-mysql" {
source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/azure/mysql?ref={revision}"
message = "${module.datadog-message-alerting.alerting-message}"
environment = "${var.environment}"
}
```
Purpose
-------
Creates a DataDog monitors with the following checks :
* CPU Usage
* Memory Usage
* Free disk space
* No active connection
* IO consumption
Inputs
------
| Name | Description | Type | Default | Required |
|------|-------------|:----:|:-----:|:-----:|
| cpu_usage_message | Custom message for Mysql CPU monitor | string | `` | no |
| cpu_usage_silenced | Groups to mute for Mysql CPU monitor | map | `<map>` | no |
| cpu_usage_threshold_critical | Mysql CPU usage in percent (critical threshold) | string | `90` | no |
| cpu_usage_threshold_warning | Mysql CPU usage in percent (warning threshold) | string | `80` | no |
| cpu_usage_time_aggregator | Monitor aggregator for Mysql CPU [available values: min, max or avg] | string | `min` | no |
| cpu_usage_timeframe | Monitor timeframe for Mysql CPU [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_15m` | 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 |
| free_storage_message | Custom message for Mysql Free Storage monitor | string | `` | no |
| free_storage_silenced | Groups to mute for Mysql Free Storage monitor | map | `<map>` | no |
| free_storage_threshold_critical | Mysql Free Storage remaining in percent (critical threshold) | string | `10` | no |
| free_storage_threshold_warning | Mysql Free Storage remaining in percent (warning threshold) | string | `20` | no |
| free_storage_time_aggregator | Monitor aggregator for Mysql Free Storage [available values: min, max or avg] | string | `min` | no |
| free_storage_timeframe | Monitor timeframe for Mysql Free Storage [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_15m` | no |
| io_consumption_message | Custom message for Mysql IO consumption monitor | string | `` | no |
| io_consumption_silenced | Groups to mute for Mysql IO consumption monitor | map | `<map>` | no |
| io_consumption_threshold_critical | Mysql IO consumption in percent (critical threshold) | string | `90` | no |
| io_consumption_threshold_warning | Mysql IO consumption in percent (warning threshold) | string | `80` | no |
| io_consumption_time_aggregator | Monitor aggregator for Mysql IO consumption [available values: min, max or avg] | string | `min` | no |
| io_consumption_timeframe | Monitor timeframe for Mysql IO consumption [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_15m` | no |
| memory_usage_message | Custom message for Mysql memory monitor | string | `` | no |
| memory_usage_silenced | Groups to mute for Mysql memory monitor | map | `<map>` | no |
| memory_usage_threshold_critical | Mysql memory usage in percent (critical threshold) | string | `90` | no |
| memory_usage_threshold_warning | Mysql memory usage in percent (warning threshold) | string | `80` | no |
| memory_usage_time_aggregator | Monitor aggregator for Mysql memory [available values: min, max or avg] | string | `min` | no |
| memory_usage_timeframe | Monitor timeframe for Mysql memory [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_15m` | no |
| message | Message sent when a Redis monitor is triggered | string | - | yes |
| no_connection_message | Custom message for Mysql no connection monitor | string | `` | no |
| no_connection_silenced | Groups to mute for Mysql no connection monitor | map | `<map>` | no |
| no_connection_time_aggregator | Monitor aggregator for Mysql no connection [available values: min, max or avg] | string | `min` | no |
| no_connection_timeframe | Monitor timeframe for Mysql no connection [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
Related documentation
---------------------
DataDog documentation: [https://docs.datadoghq.com/integrations/azure/](https://docs.datadoghq.com/integrations/azure/)
You have to search `mysql`
Azure Database for MySQL servers metrics documentation: [https://docs.microsoft.com/en-us/azure/mysql/concepts-monitoring](https://docs.microsoft.com/en-us/azure/mysql/concepts-monitoring)

186
cloud/azure/mysql/inputs.tf Normal file
View File

@ -0,0 +1,186 @@
# Global Terraform
variable "environment" {
description = "Architecture environment"
type = "string"
}
# Global DataDog
variable "message" {
description = "Message sent when a Redis monitor is triggered"
}
variable "delay" {
description = "Delay in seconds for the metric evaluation"
default = 900
}
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 = "*"
}
# Azure Mysql specific variables
variable "cpu_usage_silenced" {
description = "Groups to mute for Mysql CPU monitor"
type = "map"
default = {}
}
variable "cpu_usage_message" {
description = "Custom message for Mysql CPU monitor"
type = "string"
default = ""
}
variable "cpu_usage_time_aggregator" {
description = "Monitor aggregator for Mysql CPU [available values: min, max or avg]"
type = "string"
default = "min"
}
variable "cpu_usage_timeframe" {
description = "Monitor timeframe for Mysql CPU [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
type = "string"
default = "last_15m"
}
variable "cpu_usage_threshold_warning" {
description = "Mysql CPU usage in percent (warning threshold)"
default = "80"
}
variable "cpu_usage_threshold_critical" {
description = "Mysql CPU usage in percent (critical threshold)"
default = "90"
}
variable "no_connection_silenced" {
description = "Groups to mute for Mysql no connection monitor"
type = "map"
default = {}
}
variable "no_connection_message" {
description = "Custom message for Mysql no connection monitor"
type = "string"
default = ""
}
variable "no_connection_time_aggregator" {
description = "Monitor aggregator for Mysql no connection [available values: min, max or avg]"
type = "string"
default = "min"
}
variable "no_connection_timeframe" {
description = "Monitor timeframe for Mysql no connection [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 "free_storage_silenced" {
description = "Groups to mute for Mysql Free Storage monitor"
type = "map"
default = {}
}
variable "free_storage_message" {
description = "Custom message for Mysql Free Storage monitor"
type = "string"
default = ""
}
variable "free_storage_time_aggregator" {
description = "Monitor aggregator for Mysql Free Storage [available values: min, max or avg]"
type = "string"
default = "min"
}
variable "free_storage_timeframe" {
description = "Monitor timeframe for Mysql Free Storage [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
type = "string"
default = "last_15m"
}
variable "free_storage_threshold_warning" {
description = "Mysql Free Storage remaining in percent (warning threshold)"
default = "20"
}
variable "free_storage_threshold_critical" {
description = "Mysql Free Storage remaining in percent (critical threshold)"
default = "10"
}
variable "io_consumption_silenced" {
description = "Groups to mute for Mysql IO consumption monitor"
type = "map"
default = {}
}
variable "io_consumption_message" {
description = "Custom message for Mysql IO consumption monitor"
type = "string"
default = ""
}
variable "io_consumption_time_aggregator" {
description = "Monitor aggregator for Mysql IO consumption [available values: min, max or avg]"
type = "string"
default = "min"
}
variable "io_consumption_timeframe" {
description = "Monitor timeframe for Mysql IO consumption [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
type = "string"
default = "last_15m"
}
variable "io_consumption_threshold_warning" {
description = "Mysql IO consumption in percent (warning threshold)"
default = "80"
}
variable "io_consumption_threshold_critical" {
description = "Mysql IO consumption in percent (critical threshold)"
default = "90"
}
variable "memory_usage_silenced" {
description = "Groups to mute for Mysql memory monitor"
type = "map"
default = {}
}
variable "memory_usage_message" {
description = "Custom message for Mysql memory monitor"
type = "string"
default = ""
}
variable "memory_usage_time_aggregator" {
description = "Monitor aggregator for Mysql memory [available values: min, max or avg]"
type = "string"
default = "min"
}
variable "memory_usage_timeframe" {
description = "Monitor timeframe for Mysql memory [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
type = "string"
default = "last_15m"
}
variable "memory_usage_threshold_warning" {
description = "Mysql memory usage in percent (warning threshold)"
default = "80"
}
variable "memory_usage_threshold_critical" {
description = "Mysql memory usage in percent (critical threshold)"
default = "90"
}

View File

@ -0,0 +1,162 @@
data "template_file" "filter" {
template = "$${filter}"
vars {
filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_azure_mysql:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}"
}
}
resource "datadog_monitor" "mysql_cpu_usage" {
name = "[${var.environment}] Mysql Server CPU usage is high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = "${coalesce(var.cpu_usage_message, var.message)}"
query = <<EOF
${var.cpu_usage_time_aggregator}(${var.cpu_usage_timeframe}): (
avg:azure.dbformysql_servers.cpu_percent{${data.template_file.filter.rendered}} by {resource_group,region,name}
) > ${var.cpu_usage_threshold_critical}
EOF
type = "metric alert"
thresholds {
critical = "${var.cpu_usage_threshold_critical}"
warning = "${var.cpu_usage_threshold_warning}"
}
silenced = "${var.cpu_usage_silenced}"
notify_no_data = true
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}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
}
resource "datadog_monitor" "mysql_no_connection" {
name = "[${var.environment}] Mysql Server has no connection"
message = "${coalesce(var.no_connection_message, var.message)}"
query = <<EOF
${var.no_connection_time_aggregator}(${var.no_connection_timeframe}): (
avg:azure.dbformysql_servers.active_connections{${data.template_file.filter.rendered}} by {resource_group,region,name}
) < 1
EOF
type = "metric alert"
silenced = "${var.no_connection_silenced}"
notify_no_data = true
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}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
}
resource "datadog_monitor" "mysql_free_storage" {
name = "[${var.environment}] Mysql Server storage is running low {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = "${coalesce(var.free_storage_message, var.message)}"
query = <<EOF
${var.free_storage_time_aggregator}(${var.free_storage_timeframe}): (
100 - avg:azure.dbformysql_servers.storage_percent{${data.template_file.filter.rendered}} by {resource_group,region,name}
) < ${var.free_storage_threshold_critical}
EOF
type = "metric alert"
thresholds {
critical = "${var.free_storage_threshold_critical}"
warning = "${var.free_storage_threshold_warning}"
}
silenced = "${var.free_storage_silenced}"
notify_no_data = true
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}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
}
resource "datadog_monitor" "mysql_io_consumption" {
name = "[${var.environment}] Mysql Server IO consumption is high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = "${coalesce(var.io_consumption_message, var.message)}"
query = <<EOF
${var.io_consumption_time_aggregator}(${var.io_consumption_timeframe}): (
avg:azure.dbformysql_servers.io_consumption_percent{${data.template_file.filter.rendered}} by {resource_group,region,name}
) > ${var.io_consumption_threshold_critical}
EOF
type = "metric alert"
thresholds {
critical = "${var.io_consumption_threshold_critical}"
warning = "${var.io_consumption_threshold_warning}"
}
silenced = "${var.io_consumption_silenced}"
notify_no_data = true
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}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
}
resource "datadog_monitor" "mysql_memory_usage" {
name = "[${var.environment}] Mysql Server memory usage is high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = "${coalesce(var.memory_usage_message, var.message)}"
query = <<EOF
${var.memory_usage_time_aggregator}(${var.memory_usage_timeframe}): (
avg:azure.dbformysql_servers.memory_percent{${data.template_file.filter.rendered}} by {resource_group,region,name}
) > ${var.memory_usage_threshold_critical}
EOF
type = "metric alert"
thresholds {
critical = "${var.memory_usage_threshold_critical}"
warning = "${var.memory_usage_threshold_warning}"
}
silenced = "${var.memory_usage_silenced}"
notify_no_data = true
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}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
}