Merged in MON-74-azure-appservices-monitors (pull request #15)
MON-74: Azure App Services monitors Approved-by: Alexandre Gaillet <alexandre.gaillet@fr.clara.net> Approved-by: Laurent Piroelle <laurent.piroelle@fr.clara.net>
This commit is contained in:
Marc-Antoine
Jérôme Respaut
commit
a393cd2a60
50
cloud/azure/app-services/README.md
Normal file
50
cloud/azure/app-services/README.md
Normal file
@ -0,0 +1,50 @@
|
||||
Azure AppServices (Web, API, Functions) DataDog monitors
|
||||
========================================================
|
||||
|
||||
How to use this module
|
||||
----------------------
|
||||
|
||||
```
|
||||
module "datadog-monitors-azure-app-services" {
|
||||
source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/azure/app-services?ref={revision}"
|
||||
|
||||
message = "${module.datadog-message-alerting.alerting-message}"
|
||||
environment = "${var.environment}"
|
||||
}
|
||||
```
|
||||
|
||||
Purpose
|
||||
-------
|
||||
Creates a DataDog monitors with the following checks :
|
||||
|
||||
* Response time
|
||||
* Memory usage count
|
||||
* HTTP 404 errors
|
||||
* HTTP 50x errors
|
||||
* HTTP 20x rate
|
||||
|
||||
Inputs
|
||||
------
|
||||
|
||||
| Name | Description | Type | Default | Required |
|
||||
|------|-------------|:----:|:-----:|:-----:|
|
||||
| delay | Delay in seconds for the metric evaluation | string | `600` | 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 |
|
||||
| http_2xx_status_rate_limit | | string | `30` | no |
|
||||
| http_2xx_status_rate_threshold_critical | Alerting threshold (percentage) | string | `0.9` | no |
|
||||
| http_2xx_status_rate_threshold_warning | Warning threshold (percentage) | string | `0.95` | no |
|
||||
| http_404_errors_count_rate_limit | | string | `30` | no |
|
||||
| http_404_errors_count_rate_threshold_critical | Alerting threshold (number of requests) | string | `30` | no |
|
||||
| http_404_errors_count_rate_threshold_warning | Warning threshold (number of requests) | string | `10` | no |
|
||||
| memory_usage_threshold_critical | Alerting threshold in Mib | string | `52430000` | no |
|
||||
| memory_usage_threshold_warning | Warning threshold in MiB | string | `33550000` | no |
|
||||
| message | Message sent when a monitor is triggered | string | - | yes |
|
||||
| response_time_threshold_critical | Alerting threshold in seconds | string | `0.8` | no |
|
||||
| response_time_threshold_warning | Warning threshold in seconds | string | `0.4` | no |
|
||||
|
||||
Related documentation
|
||||
---------------------
|
||||
|
||||
DataDog documentation: [https://docs.datadoghq.com/integrations/azure_app_services](https://docs.datadoghq.com/integrations/azure_app_services)
|
||||
87
cloud/azure/app-services/inputs.tf
Normal file
87
cloud/azure/app-services/inputs.tf
Normal file
@ -0,0 +1,87 @@
|
||||
variable "environment" {
|
||||
description = "Architecture environment"
|
||||
type = "string"
|
||||
}
|
||||
|
||||
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 = "*"
|
||||
}
|
||||
|
||||
variable "message" {
|
||||
description = "Message sent when a monitor is triggered"
|
||||
}
|
||||
|
||||
variable "delay" {
|
||||
description = "Delay in seconds for the metric evaluation"
|
||||
default = 600
|
||||
}
|
||||
|
||||
###################################
|
||||
### RESPONSE TIME VARIABLES ###
|
||||
###################################
|
||||
|
||||
variable "response_time_threshold_critical" {
|
||||
default = 0.8
|
||||
description = "Alerting threshold in seconds"
|
||||
}
|
||||
|
||||
variable "response_time_threshold_warning" {
|
||||
default = 0.4
|
||||
description = "Warning threshold in seconds"
|
||||
}
|
||||
|
||||
###################################
|
||||
### MEMORY USAGE VARIABLES ###
|
||||
###################################
|
||||
|
||||
variable "memory_usage_threshold_critical" {
|
||||
default = 52430000
|
||||
description = "Alerting threshold in Mib"
|
||||
}
|
||||
|
||||
variable "memory_usage_threshold_warning" {
|
||||
default = 33550000
|
||||
description = "Warning threshold in MiB"
|
||||
}
|
||||
|
||||
#################################
|
||||
### HTTP 404 status pages ###
|
||||
#################################
|
||||
|
||||
variable "http_404_errors_count_rate_limit" {
|
||||
default = 30
|
||||
}
|
||||
|
||||
variable "http_404_errors_count_rate_threshold_critical" {
|
||||
default = 30
|
||||
description = "Alerting threshold (number of requests)"
|
||||
}
|
||||
|
||||
variable "http_404_errors_count_rate_threshold_warning" {
|
||||
default = 10
|
||||
description = "Warning threshold (number of requests)"
|
||||
}
|
||||
|
||||
#################################
|
||||
### HTTP 202 status pages ###
|
||||
#################################
|
||||
|
||||
variable "http_2xx_status_rate_limit" {
|
||||
default = 30
|
||||
}
|
||||
|
||||
variable "http_2xx_status_rate_threshold_critical" {
|
||||
default = 0.9
|
||||
description = "Alerting threshold (percentage)"
|
||||
}
|
||||
|
||||
variable "http_2xx_status_rate_threshold_warning" {
|
||||
default = 0.95
|
||||
description = "Warning threshold (percentage)"
|
||||
}
|
||||
124
cloud/azure/app-services/monitors-app_services.tf
Normal file
124
cloud/azure/app-services/monitors-app_services.tf
Normal file
@ -0,0 +1,124 @@
|
||||
data "template_file" "filter" {
|
||||
template = "$${filter}"
|
||||
|
||||
vars {
|
||||
filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_azure_storage:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}"
|
||||
}
|
||||
}
|
||||
|
||||
# Monitoring App Services response time
|
||||
resource "datadog_monitor" "appservices_response_time" {
|
||||
name = "[${var.environment}] App Services response time > ${var.response_time_threshold_critical}s on {{name}}"
|
||||
type = "metric alert"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
avg(last_5m): (
|
||||
avg:azure.app_services.average_response_time{${data.template_file.filter.rendered}}
|
||||
) >= ${var.response_time_threshold_critical}
|
||||
EOF
|
||||
|
||||
evaluation_delay = "${var.delay}"
|
||||
new_host_delay = "${var.delay}"
|
||||
|
||||
thresholds {
|
||||
warning = "${var.response_time_threshold_warning}"
|
||||
critical = "${var.response_time_threshold_critical}"
|
||||
}
|
||||
|
||||
notify_no_data = true # Will notify when no data is received
|
||||
renotify_interval = 0
|
||||
require_full_window = true
|
||||
timeout_h = 0
|
||||
include_tags = true
|
||||
|
||||
tags = ["env:${var.environment}", "resource:appservices", "team:azure", "provider:azure"]
|
||||
}
|
||||
|
||||
# Monitoring App Services memory usage
|
||||
resource "datadog_monitor" "appservices_memory_usage_count" {
|
||||
name = "[${var.environment}] App Services memory usage > ${ceil(var.memory_usage_threshold_critical/1000000)}MiB on {{name}}"
|
||||
type = "metric alert"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
avg(last_5m): (
|
||||
avg:azure.app_services.memory_working_set{${data.template_file.filter.rendered}}
|
||||
) >= ${var.memory_usage_threshold_critical}
|
||||
EOF
|
||||
|
||||
evaluation_delay = "${var.delay}"
|
||||
new_host_delay = "${var.delay}"
|
||||
|
||||
thresholds {
|
||||
warning = "${var.memory_usage_threshold_warning}"
|
||||
critical = "${var.memory_usage_threshold_critical}"
|
||||
}
|
||||
|
||||
notify_no_data = true # Will notify when no data is received
|
||||
renotify_interval = 0
|
||||
require_full_window = true
|
||||
timeout_h = 0
|
||||
include_tags = true
|
||||
|
||||
tags = ["env:${var.environment}", "resource:appservices", "team:azure", "provider:azure"]
|
||||
}
|
||||
|
||||
# Monitoring App Services 404 errors rate
|
||||
resource "datadog_monitor" "appservices_http_404_errors_count" {
|
||||
name = "[${var.environment}] App Services HTTP errors > ${var.http_404_errors_count_rate_limit} limit on {{name}}"
|
||||
type = "metric alert"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
max(last_5m): (
|
||||
per_minute(avg:azure.app_services.http404{${data.template_file.filter.rendered}}.as_rate())
|
||||
) > ${var.http_404_errors_count_rate_threshold_critical}
|
||||
EOF
|
||||
|
||||
evaluation_delay = "${var.delay}"
|
||||
new_host_delay = "${var.delay}"
|
||||
|
||||
thresholds {
|
||||
warning = "${var.http_404_errors_count_rate_threshold_warning}"
|
||||
critical = "${var.http_404_errors_count_rate_threshold_critical}"
|
||||
}
|
||||
|
||||
notify_no_data = false # Will NOT notify when no data is received
|
||||
renotify_interval = 0
|
||||
require_full_window = true
|
||||
timeout_h = 0
|
||||
include_tags = true
|
||||
|
||||
tags = ["env:${var.environment}", "resource:appservices", "team:azure", "provider:azure"]
|
||||
}
|
||||
|
||||
# Monitoring App Services HTTP 2xx status pages rate
|
||||
resource "datadog_monitor" "appservices_http_2xx_status_rate" {
|
||||
name = "[${var.environment}] App Services Too much non 2xx HTTP status in response to the requests on {{name}}"
|
||||
type = "metric alert"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
avg(last_5m): (
|
||||
avg:azure.app_services.http2xx{${data.template_file.filter.rendered}}.as_count() /
|
||||
avg:azure.app_services.http2xx{${data.template_file.filter.rendered}}.as_count()
|
||||
) < ${var.http_2xx_status_rate_threshold_critical}
|
||||
EOF
|
||||
|
||||
evaluation_delay = "${var.delay}"
|
||||
new_host_delay = "${var.delay}"
|
||||
|
||||
thresholds {
|
||||
warning = "${var.http_2xx_status_rate_threshold_warning}"
|
||||
critical = "${var.http_2xx_status_rate_threshold_critical}"
|
||||
}
|
||||
|
||||
notify_no_data = true # Will notify when no data is received
|
||||
renotify_interval = 0
|
||||
require_full_window = true
|
||||
timeout_h = 0
|
||||
include_tags = true
|
||||
|
||||
tags = ["env:${var.environment}", "resource:appservices", "team:azure", "provider:azure"]
|
||||
}
|
||||
Loading…
x
Reference in New Issue
Block a user