wbyc/terraform-datadog-old-monitors
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

MON-79 azure storage minitors

Browse Source
This commit is contained in:
Quentin Manfroi 2017-10-31 20:32:44 +01:00
parent ac6e0d69e1
commit fdfe302a34
3 changed files with 386 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
cloud/azure/storage/README.md Normal file
View File

@ -0,0 +1,60 @@
Azure Storage DataDog monitors
============================
How to use this module
----------------------
```
module "datadog-monitors-azure-storage" {
source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/azure/storage?ref={revision}"
message = "${module.datadog-message-alerting.alerting-message}"
environment = "${var.environment}"
client_name = "${var.client_name}"
}
```
Purpose
-------
Creates a DataDog monitors with the following checks :
* Service availability
* End to end latency
* Minimum successful requests
* Maximum timeout error requests
* Maximum network error requests
* Maximum throttling error requests
* Maximum server other error requests
* Maximum client other error requests
* Maximum authorization error requests
Inputs
------
| Name | Description | Type | Default | Required |
|------|-------------|:----:|:-----:|:-----:|
| client_name | Client name | string | - | yes |
| delay | Delay in seconds for the metric evaluation | string | `600` | no |
| environment | Architecture environment | string | - | yes |
| message | Message sent when a monitor is triggered | string | - | yes |
| use_filter_tags | Filter the data with service tags if true | string | `true` | no |
| availability_threshold_critical | Minimum threshold of availability | string | `90` | no |
| successful_requests_threshold_critical | Minimum threshold of successful requests | string | `90` | no |
| latency_threshold_critical | Maximum threshold of latency in ms | string | `1000` | no |
| timeout_error_requests_threshold_critical | Maximum threshold of timeout error requests in percent | string | `35` | no |
| network_error_requests_threshold_critical | Maximum threshold of network error requests in percent | string | `35` | no |
| throttling_error_requests_threshold_critical | Maximum threshold of throttling error requests in percent | string | `50` | no |
| server_other_error_requests_threshold_critical | Maximum threshold of server other error requests in percent | string | `50` | no |
| client_other_error_requests_threshold_critical | Maximum threshold of client other error requests in percent | string | `75` | no |
| authorization_error_requests_threshold_critical | Maximum threshold of authorization error requests in percent | string | `75` | no |
Related documentation
---------------------
DataDog documentation: [https://docs.datadoghq.com/integrations/azure_storage/](https://docs.datadoghq.com/integrations/azure_storage/)
DataDog blog: [https://www.datadoghq.com/blog/monitor-azure-storage-datadog/](https://www.datadoghq.com/blog/monitor-azure-storage-datadog/)
Azure Redis metrics documentation: [https://docs.microsoft.com/en-us/azure/storage/common/storage-monitor-storage-account](https://docs.microsoft.com/en-us/azure/storage/common/storage-monitor-storage-account)

71
cloud/azure/storage/inputs.tf Normal file
View File

@ -0,0 +1,71 @@
# Global Terraform
variable "client_name" {
type = "string"
}
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 = 600
}
variable "use_filter_tags" {
description = "Filter the data with service tags if true"
default = "true"
}
# Azure Storage specific
variable "availability_threshold_critical" {
description = "Minimum acceptable percent of availability for a storage"
default = 90
}
variable "successful_requests_threshold_critical" {
description = "Minimum acceptable percent of successful requests for a storage"
default = 90
}
variable "latency_threshold_critical" {
description = "Maximum acceptable end to end latency (ms) for a storage"
default = 1000
}
variable "timeout_error_requests_threshold_critical" {
description = "Maximum acceptable percent of timeout error requests for a storage"
default = 35
}
variable "network_error_requests_threshold_critical" {
description = "Maximum acceptable percent of network error requests for a storage"
default = 35
}
variable "throttling_error_requests_threshold_critical" {
description = "Maximum acceptable percent of throttling error requests for a storage"
default = 50
}
variable "server_other_error_requests_threshold_critical" {
description = "Maximum acceptable percent of server other error requests for a storage"
default = 50
}
variable "client_other_error_requests_threshold_critical" {
description = "Maximum acceptable percent of client other error requests for a storage"
default = 75
}
variable "authorization_error_requests_threshold_critical" {
description = "Maximum acceptable percent of authorization error requests for a storage"
default = 75
}

255
cloud/azure/storage/monitors-azure-storage.tf Normal file
View File

@ -0,0 +1,255 @@
data "template_file" "filter" {
template = "$${filter}"
vars {
filter = "${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_storage:enabled,env:%s", var.environment) : "*"}"
}
}
resource "datadog_monitor" "availability" {
name = "[${var.environment}] Azure Storage {{name}} unvailability detected"
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.storage.availability{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} < ${var.availability_threshold_critical}
EOF
thresholds {
critical = "${var.availability_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}
resource "datadog_monitor" "successful_requests" {
name = "[${var.environment}] Azure Storage {{name}} too much failed requests"
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.storage.percent_success{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} < ${var.successful_requests_threshold_critical}
EOF
thresholds {
critical = "${var.successful_requests_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}
resource "datadog_monitor" "latency" {
name = "[${var.environment}] Azure Storage {{name}} too high end to end latency"
message = "${var.message}"
query = <<EOF
max(last_5m):avg:azure.storage.average_e2_e_latency{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} > ${var.latency_threshold_critical}
EOF
thresholds {
critical = "${var.latency_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}
resource "datadog_monitor" "timeout_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of timeout error requests on {{name}}"
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.storage.percent_timeout_error{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} > ${var.timeout_error_requests_threshold_critical}
EOF
thresholds {
critical = "${var.timeout_error_requests_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}
resource "datadog_monitor" "network_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of network error requests on {{name}}"
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.storage.percent_network_error{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} > ${var.network_error_requests_threshold_critical}
EOF
thresholds {
critical = "${var.network_error_requests_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}
resource "datadog_monitor" "throttling_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of throttling error requests on {{name}}"
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.storage.percent_throttling_error{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} > ${var.throttling_error_requests_threshold_critical}
EOF
thresholds {
critical = "${var.throttling_error_requests_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}
resource "datadog_monitor" "server_other_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of server_other error requests on {{name}}"
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.storage.percent_server_other_error{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} > ${var.server_other_error_requests_threshold_critical}
EOF
thresholds {
critical = "${var.server_other_error_requests_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}
resource "datadog_monitor" "client_other_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of client_other error requests on {{name}}"
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.storage.percent_client_other_error{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} > ${var.client_other_error_requests_threshold_critical}
EOF
thresholds {
critical = "${var.client_other_error_requests_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}
resource "datadog_monitor" "authorization_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of authorization error requests on {{name}}"
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.storage.percent_authorization_error{${data.template_file.filter.rendered}} by {resource_group,storage_type,name} > ${var.authorization_error_requests_threshold_critical}
EOF
thresholds {
critical = "${var.authorization_error_requests_threshold_critical}"
}
type = "query alert"
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}"
renotify_interval = 0
no_data_timeframe = 20
tags = ["env:${var.environment}"]
}