MON-246 - Best practice applied

This commit is contained in:
Alexandre Gaillet 2018-09-28 14:04:50 +02:00 committed by Quentin Manfroi
parent bdfb100cbc
commit be59a0030c
3 changed files with 142 additions and 54 deletions

View File

@ -16,52 +16,65 @@ module "datadog-monitors-cloud-azure-mysql" {
Creates DataDog monitors with the following checks:
- Mysql Server compute consumption is high
- Mysql Server CPU usage is high
- Mysql Server compute consumption
- Mysql Server CPU usage
- Mysql Server has no connection
- Mysql Server IO consumption is high
- Mysql Server memory usage is high
- Mysql Server storage is running low
- Mysql Server IO consumption
- Mysql Server memory usage
- Mysql Server storage
## Inputs
| Name | Description | Type | Default | Required |
|------|-------------|:----:|:-----:|:-----:|
| compute_consumption_enabled | Flag to enable Mysql status monitor | string | `true` | no |
| compute_consumption_extra_tags | Extra tags for Mysql status monitor | list | `[]` | no |
| compute_consumption_message | Custom message for Mysql compute consumption monitor | string | `` | no |
| compute_consumption_silenced | Groups to mute for Mysql compute consumption monitor | map | `{}` | no |
| compute_consumption_threshold_critical | Mysql compute consumption in percent (critical threshold) | string | `90` | no |
| compute_consumption_threshold_warning | Mysql compute consumption in percent (warning threshold) | string | `80` | no |
| compute_consumption_time_aggregator | Monitor aggregator for Mysql compute consumption [available values: min, max or avg] | string | `min` | no |
| compute_consumption_timeframe | Monitor timeframe for Mysql compute consumption [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_15m` | no |
| cpu_usage_enabled | Flag to enable Mysql status monitor | string | `true` | no |
| cpu_usage_extra_tags | Extra tags for Mysql status monitor | list | `[]` | no |
| cpu_usage_message | Custom message for Mysql CPU monitor | string | `` | no |
| cpu_usage_silenced | Groups to mute for Mysql CPU monitor | 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 |
| evaluation_delay | Delay in seconds for the metric evaluation | string | `900` | 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 |
| free_storage_enabled | Flag to enable Mysql status monitor | string | `true` | no |
| free_storage_extra_tags | Extra tags for Mysql status monitor | list | `[]` | 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 | `{}` | 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_enabled | Flag to enable Mysql status monitor | string | `true` | no |
| io_consumption_extra_tags | Extra tags for Mysql status monitor | list | `[]` | 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 | `{}` | 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_enabled | Flag to enable Mysql status monitor | string | `true` | no |
| memory_usage_extra_tags | Extra tags for Mysql status monitor | list | `[]` | no |
| memory_usage_message | Custom message for Mysql memory monitor | string | `` | no |
| memory_usage_silenced | Groups to mute for Mysql memory monitor | 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 |
| message | Message sent when an alert is triggered | string | - | yes |
| new_host_delay | Delay in seconds before monitor new resource | string | `300` | no |
| no_connection_enabled | Flag to enable Mysql status monitor | string | `true` | no |
| no_connection_extra_tags | Extra tags for Mysql status monitor | list | `[]` | no |
| no_connection_message | Custom message for Mysql no connection monitor | string | `` | no |
| no_connection_silenced | Groups to mute for Mysql no connection monitor | map | `{}` | no |
| no_connection_time_aggregator | Monitor aggregator for Mysql no connection [available values: min, max or avg] | string | `min` | no |

View File

@ -5,15 +5,20 @@ variable "environment" {
}
# Global DataDog
variable "message" {
description = "Message sent when a Redis monitor is triggered"
}
variable "delay" {
variable "evaluation_delay" {
description = "Delay in seconds for the metric evaluation"
default = 900
}
variable "new_host_delay" {
description = "Delay in seconds before monitor new resource"
default = 300
}
variable "message" {
description = "Message sent when an alert is triggered"
}
variable "filter_tags_use_defaults" {
description = "Use default filter tags convention"
default = "true"
@ -31,6 +36,18 @@ variable "cpu_usage_silenced" {
default = {}
}
variable "cpu_usage_enabled" {
description = "Flag to enable Mysql status monitor"
type = "string"
default = "true"
}
variable "cpu_usage_extra_tags" {
description = "Extra tags for Mysql status monitor"
type = "list"
default = []
}
variable "cpu_usage_message" {
description = "Custom message for Mysql CPU monitor"
type = "string"
@ -65,6 +82,18 @@ variable "no_connection_silenced" {
default = {}
}
variable "no_connection_enabled" {
description = "Flag to enable Mysql status monitor"
type = "string"
default = "true"
}
variable "no_connection_extra_tags" {
description = "Extra tags for Mysql status monitor"
type = "list"
default = []
}
variable "no_connection_message" {
description = "Custom message for Mysql no connection monitor"
type = "string"
@ -89,6 +118,18 @@ variable "free_storage_silenced" {
default = {}
}
variable "free_storage_enabled" {
description = "Flag to enable Mysql status monitor"
type = "string"
default = "true"
}
variable "free_storage_extra_tags" {
description = "Extra tags for Mysql status monitor"
type = "list"
default = []
}
variable "free_storage_message" {
description = "Custom message for Mysql Free Storage monitor"
type = "string"
@ -123,6 +164,18 @@ variable "io_consumption_silenced" {
default = {}
}
variable "io_consumption_enabled" {
description = "Flag to enable Mysql status monitor"
type = "string"
default = "true"
}
variable "io_consumption_extra_tags" {
description = "Extra tags for Mysql status monitor"
type = "list"
default = []
}
variable "io_consumption_message" {
description = "Custom message for Mysql IO consumption monitor"
type = "string"
@ -157,6 +210,18 @@ variable "compute_consumption_silenced" {
default = {}
}
variable "compute_consumption_enabled" {
description = "Flag to enable Mysql status monitor"
type = "string"
default = "true"
}
variable "compute_consumption_extra_tags" {
description = "Extra tags for Mysql status monitor"
type = "list"
default = []
}
variable "compute_consumption_message" {
description = "Custom message for Mysql compute consumption monitor"
type = "string"
@ -191,6 +256,18 @@ variable "memory_usage_silenced" {
default = {}
}
variable "memory_usage_enabled" {
description = "Flag to enable Mysql status monitor"
type = "string"
default = "true"
}
variable "memory_usage_extra_tags" {
description = "Extra tags for Mysql status monitor"
type = "list"
default = []
}
variable "memory_usage_message" {
description = "Custom message for Mysql memory monitor"
type = "string"

View File

@ -1,18 +1,11 @@
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 = "[${var.environment}] Mysql Server CPU usage is high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
count = "${var.cpu_usage_enabled ? 1 : 0}"
name = "[${var.environment}] Mysql Server CPU usage {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = "${coalesce(var.cpu_usage_message, var.message)}"
query = <<EOF
${var.cpu_usage_time_aggregator}(${var.cpu_usage_timeframe}): (
avg:azure.dbformysql_servers.cpu_percent{${data.template_file.filter.rendered}} by {resource_group,region,name}
avg:azure.dbformysql_servers.cpu_percent${module.filter-tags.query_alert} by {resource_group,region,name}
) > ${var.cpu_usage_threshold_critical}
EOF
@ -25,26 +18,27 @@ resource "datadog_monitor" "mysql_cpu_usage" {
silenced = "${var.cpu_usage_silenced}"
notify_no_data = true
evaluation_delay = "${var.delay}"
notify_no_data = false
evaluation_delay = "${var.evaluation_delay}"
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = "${var.delay}"
new_host_delay = "${var.new_host_delay}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", "${var.cpu_usage_extra_tags}"]
}
resource "datadog_monitor" "mysql_no_connection" {
count = "${var.no_connection_enabled ? 1 : 0}"
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{${data.template_file.filter.rendered}} by {resource_group,region,name}
avg:azure.dbformysql_servers.active_connections${module.filter-tags.query_alert} by {resource_group,region,name}
) < 1
EOF
@ -53,25 +47,26 @@ resource "datadog_monitor" "mysql_no_connection" {
silenced = "${var.no_connection_silenced}"
notify_no_data = true
evaluation_delay = "${var.delay}"
evaluation_delay = "${var.evaluation_delay}"
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = "${var.delay}"
new_host_delay = "${var.new_host_delay}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", "${var.no_connection_extra_tags}"]
}
resource "datadog_monitor" "mysql_free_storage" {
name = "[${var.environment}] Mysql Server storage is running low {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
count = "${var.free_storage_enabled ? 1 : 0}"
name = "[${var.environment}] Mysql Server storage {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = "${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{${data.template_file.filter.rendered}} by {resource_group,region,name}
100 - avg:azure.dbformysql_servers.storage_percent${module.filter-tags.query_alert} by {resource_group,region,name}
) < ${var.free_storage_threshold_critical}
EOF
@ -84,26 +79,27 @@ resource "datadog_monitor" "mysql_free_storage" {
silenced = "${var.free_storage_silenced}"
notify_no_data = true
evaluation_delay = "${var.delay}"
notify_no_data = false
evaluation_delay = "${var.evaluation_delay}"
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = "${var.delay}"
new_host_delay = "${var.new_host_delay}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", "${var.free_storage_extra_tags}"]
}
resource "datadog_monitor" "mysql_io_consumption" {
name = "[${var.environment}] Mysql Server IO consumption is high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
count = "${var.io_consumption_enabled ? 1 : 0}"
name = "[${var.environment}] Mysql Server IO consumption {{#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{${data.template_file.filter.rendered}} by {resource_group,region,name}
avg:azure.dbformysql_servers.io_consumption_percent${module.filter-tags.query_alert} by {resource_group,region,name}
) > ${var.io_consumption_threshold_critical}
EOF
@ -116,26 +112,27 @@ resource "datadog_monitor" "mysql_io_consumption" {
silenced = "${var.io_consumption_silenced}"
notify_no_data = true
evaluation_delay = "${var.delay}"
notify_no_data = false
evaluation_delay = "${var.evaluation_delay}"
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = "${var.delay}"
new_host_delay = "${var.new_host_delay}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", "${var.io_consumption_extra_tags}"]
}
resource "datadog_monitor" "mysql_compute_consumption" {
name = "[${var.environment}] Mysql Server compute consumption is high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
count = "${var.compute_consumption_enabled ? 1 : 0}"
name = "[${var.environment}] Mysql Server compute consumption {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = "${coalesce(var.compute_consumption_message, var.message)}"
query = <<EOF
${var.compute_consumption_time_aggregator}(${var.compute_consumption_timeframe}): (
avg:azure.dbformysql_servers.compute_consumption_percent{${data.template_file.filter.rendered}} by {resource_group,region,name}
avg:azure.dbformysql_servers.compute_consumption_percent${module.filter-tags.query_alert} by {resource_group,region,name}
) > ${var.compute_consumption_threshold_critical}
EOF
@ -148,26 +145,27 @@ resource "datadog_monitor" "mysql_compute_consumption" {
silenced = "${var.compute_consumption_silenced}"
notify_no_data = true
evaluation_delay = "${var.delay}"
notify_no_data = false
evaluation_delay = "${var.evaluation_delay}"
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = "${var.delay}"
new_host_delay = "${var.new_host_delay}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", "${var.compute_consumption_extra_tags}"]
}
resource "datadog_monitor" "mysql_memory_usage" {
name = "[${var.environment}] Mysql Server memory usage is high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
count = "${var.memory_usage_enabled ? 1 : 0}"
name = "[${var.environment}] Mysql Server memory usage {{#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{${data.template_file.filter.rendered}} by {resource_group,region,name}
avg:azure.dbformysql_servers.memory_percent${module.filter-tags.query_alert} by {resource_group,region,name}
) > ${var.memory_usage_threshold_critical}
EOF
@ -180,15 +178,15 @@ resource "datadog_monitor" "mysql_memory_usage" {
silenced = "${var.memory_usage_silenced}"
notify_no_data = true
evaluation_delay = "${var.delay}"
notify_no_data = false
evaluation_delay = "${var.evaluation_delay}"
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = "${var.delay}"
new_host_delay = "${var.new_host_delay}"
tags = ["env:${var.environment}", "resource:mysql", "team:azure", "provider:azure"]
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", "${var.memory_usage_extra_tags}"]
}