Merged in MON-80-azure-hub-iot-monitors (pull request #8)
MON-80: Azure IoT Hubs Monitors Approved-by: Alexandre Gaillet <alexandre.gaillet@fr.clara.net> Approved-by: Quentin Manfroi <quentin.manfroi@yahoo.fr> Approved-by: Laurent Piroelle <laurent.piroelle@fr.clara.net>
This commit is contained in:
Alexandre Gaillet
Jérôme Respaut
commit
266c7cae89
76
cloud/azure/iothubs/README.md
Normal file
76
cloud/azure/iothubs/README.md
Normal file
@ -0,0 +1,76 @@
|
|||||||
|
Azure IOT Hubs DataDog monitors
|
||||||
|
===============================
|
||||||
|
|
||||||
|
How to use this module
|
||||||
|
----------------------
|
||||||
|
|
||||||
|
```
|
||||||
|
module "iothubs" {
|
||||||
|
source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/azure/iothubs?ref=MON-80-azure-hub-iot-monitors"
|
||||||
|
|
||||||
|
message = "${module.datadog-message-alerting.alerting-message}"
|
||||||
|
environment = "${var.environment}"
|
||||||
|
}
|
||||||
|
```
|
||||||
|
|
||||||
|
Purpose
|
||||||
|
-------
|
||||||
|
Creates a DataDog monitors with the following checks :
|
||||||
|
|
||||||
|
* Service status check
|
||||||
|
* Jobs failed average check
|
||||||
|
* Query Jobs failed average check
|
||||||
|
* List Jobs failed average check
|
||||||
|
* Total devices count check
|
||||||
|
* C2D methods failed average check
|
||||||
|
* C2D twin read failed average check
|
||||||
|
* C2D twin update failed average check
|
||||||
|
* D2C twin read failed average check
|
||||||
|
* D2C twin update failed average check
|
||||||
|
* D2C telemetry egress dropped count check
|
||||||
|
* D2C telemetry egress orphaned count check
|
||||||
|
* D2C telemetry egress invalid count check
|
||||||
|
* D2C telemetry egress fallback count check
|
||||||
|
* D2C telemetry ingress no sent count check
|
||||||
|
|
||||||
|
Inputs
|
||||||
|
------
|
||||||
|
|
||||||
|
| Name | Description | Type | Default | Required |
|
||||||
|
|------|-------------|:----:|:-----:|:-----:|
|
||||||
|
| delay | Delay in seconds for the metric evaluation | string | `600` | no |
|
||||||
|
| dropped_d2c_telemetry_egress_threshold_critical | D2C Telemetry Dropped limit (critical threshold) | string | `1000` | no |
|
||||||
|
| dropped_d2c_telemetry_egress_threshold_warning | D2C Telemetry Dropped limit (warning threshold) | string | `500` | no |
|
||||||
|
| environment | Architecture Environment | string | - | yes |
|
||||||
|
| failed_c2d_methods_rate_threshold_critical | C2D Methods Failed rate limit (critical threshold) | string | `10` | no |
|
||||||
|
| failed_c2d_methods_rate_threshold_warning | C2D Methods Failed rate limit (warning threshold) | string | `0` | no |
|
||||||
|
| failed_c2d_twin_read_rate_threshold_critical | C2D Twin Read Failed rate limit (critical threshold) | string | `10` | no |
|
||||||
|
| failed_c2d_twin_read_rate_threshold_warning | C2D Twin Read Failed rate limit (warning threshold) | string | `0` | no |
|
||||||
|
| failed_c2d_twin_update_rate_threshold_critical | C2D Twin Update Failed rate limit (critical threshold) | string | `10` | no |
|
||||||
|
| failed_c2d_twin_update_rate_threshold_warning | C2D Twin Update Failed rate limit (warning threshold) | string | `0` | no |
|
||||||
|
| failed_d2c_twin_read_rate_threshold_critical | D2C Twin Read Failed rate limit (critical threshold) | string | `10` | no |
|
||||||
|
| failed_d2c_twin_read_rate_threshold_warning | D2C Twin Read Failed rate limit (warning threshold) | string | `0` | no |
|
||||||
|
| failed_d2c_twin_update_rate_threshold_critical | D2C Twin Update Failed rate limit (critical threshold) | string | `10` | no |
|
||||||
|
| failed_d2c_twin_update_rate_threshold_warning | D2C Twin Update Failed rate limit (warning threshold) | string | `0` | no |
|
||||||
|
| failed_jobs_rate_threshold_critical | Jobs Failed rate limit (critical threshold) | string | `10` | no |
|
||||||
|
| failed_jobs_rate_threshold_warning | Jobs Failed rate limit (warning threshold) | string | `0` | no |
|
||||||
|
| failed_listjobs_rate_threshold_critical | ListJobs Failed rate limit (critical threshold) | string | `10` | no |
|
||||||
|
| failed_listjobs_rate_threshold_warning | ListJobs Failed rate limit (warning threshold) | string | `0` | no |
|
||||||
|
| failed_queryjobs_rate_threshold_critical | QueryJobs Failed rate limit (critical threshold) | string | `10` | no |
|
||||||
|
| failed_queryjobs_rate_threshold_warning | QueryJobs Failed rate limit (warning threshold) | string | `0` | no |
|
||||||
|
| fallback_d2c_telemetry_egress_threshold_critical | D2C Telemetry Fallback limit (critical threshold) | string | `1000` | no |
|
||||||
|
| fallback_d2c_telemetry_egress_threshold_warning | D2C Telemetry Fallback limit (warning threshold) | string | `500` | 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 |
|
||||||
|
| invalid_d2c_telemetry_egress_threshold_critical | D2C Telemetry Invalid limit (critical threshold) | string | `1000` | no |
|
||||||
|
| invalid_d2c_telemetry_egress_threshold_warning | D2C Telemetry Invalid limit (warning threshold) | string | `500` | no |
|
||||||
|
| message | Message sent when an alert is triggered | string | - | yes |
|
||||||
|
| orphaned_d2c_telemetry_egress_threshold_critical | D2C Telemetry Orphaned limit (critical threshold) | string | `1000` | no |
|
||||||
|
| orphaned_d2c_telemetry_egress_threshold_warning | D2C Telemetry Orphaned limit (warning threshold) | string | `500` | no |
|
||||||
|
|
||||||
|
Related documentation
|
||||||
|
---------------------
|
||||||
|
|
||||||
|
DataDog documentation: [https://docs.datadoghq.com/integrations/azure_iot_hub](https://docs.datadoghq.com/integrations/azure_iot_hub)
|
||||||
|
|
||||||
|
Azure IOT Hubs metrics documentation: [https://docs.microsoft.com/en-us/azure/iot-hub/iot-hub-monitor-resource-health](https://docs.microsoft.com/en-us/azure/iot-hub/iot-hub-monitor-resource-health)
|
||||||
146
cloud/azure/iothubs/inputs.tf
Normal file
146
cloud/azure/iothubs/inputs.tf
Normal file
@ -0,0 +1,146 @@
|
|||||||
|
# Global Terraform
|
||||||
|
variable "environment" {
|
||||||
|
description = "Architecture Environment"
|
||||||
|
type = "string"
|
||||||
|
}
|
||||||
|
|
||||||
|
# Global DataDog
|
||||||
|
variable "delay" {
|
||||||
|
description = "Delay in seconds for the metric evaluation"
|
||||||
|
default = 600
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "message" {
|
||||||
|
description = "Message sent when an alert is triggered"
|
||||||
|
}
|
||||||
|
|
||||||
|
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 IOT hubs specific
|
||||||
|
variable "failed_jobs_rate_threshold_warning" {
|
||||||
|
description = "Jobs Failed rate limit (warning threshold)"
|
||||||
|
default = 0
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_jobs_rate_threshold_critical" {
|
||||||
|
description = "Jobs Failed rate limit (critical threshold)"
|
||||||
|
default = 10
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_listjobs_rate_threshold_warning" {
|
||||||
|
description = "ListJobs Failed rate limit (warning threshold)"
|
||||||
|
default = 0
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_listjobs_rate_threshold_critical" {
|
||||||
|
description = "ListJobs Failed rate limit (critical threshold)"
|
||||||
|
default = 10
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_queryjobs_rate_threshold_warning" {
|
||||||
|
description = "QueryJobs Failed rate limit (warning threshold)"
|
||||||
|
default = 0
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_queryjobs_rate_threshold_critical" {
|
||||||
|
description = "QueryJobs Failed rate limit (critical threshold)"
|
||||||
|
default = 10
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_c2d_methods_rate_threshold_warning" {
|
||||||
|
description = "C2D Methods Failed rate limit (warning threshold)"
|
||||||
|
default = 0
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_c2d_methods_rate_threshold_critical" {
|
||||||
|
description = "C2D Methods Failed rate limit (critical threshold)"
|
||||||
|
default = 10
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_c2d_twin_read_rate_threshold_warning" {
|
||||||
|
description = "C2D Twin Read Failed rate limit (warning threshold)"
|
||||||
|
default = 0
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_c2d_twin_read_rate_threshold_critical" {
|
||||||
|
description = "C2D Twin Read Failed rate limit (critical threshold)"
|
||||||
|
default = 10
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_c2d_twin_update_rate_threshold_warning" {
|
||||||
|
description = "C2D Twin Update Failed rate limit (warning threshold)"
|
||||||
|
default = 0
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_c2d_twin_update_rate_threshold_critical" {
|
||||||
|
description = "C2D Twin Update Failed rate limit (critical threshold)"
|
||||||
|
default = 10
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_d2c_twin_read_rate_threshold_warning" {
|
||||||
|
description = "D2C Twin Read Failed rate limit (warning threshold)"
|
||||||
|
default = 0
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_d2c_twin_read_rate_threshold_critical" {
|
||||||
|
description = "D2C Twin Read Failed rate limit (critical threshold)"
|
||||||
|
default = 10
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_d2c_twin_update_rate_threshold_warning" {
|
||||||
|
description = "D2C Twin Update Failed rate limit (warning threshold)"
|
||||||
|
default = 0
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "failed_d2c_twin_update_rate_threshold_critical" {
|
||||||
|
description = "D2C Twin Update Failed rate limit (critical threshold)"
|
||||||
|
default = 10
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "dropped_d2c_telemetry_egress_threshold_warning" {
|
||||||
|
description = "D2C Telemetry Dropped limit (warning threshold)"
|
||||||
|
default = 500
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "dropped_d2c_telemetry_egress_threshold_critical" {
|
||||||
|
description = "D2C Telemetry Dropped limit (critical threshold)"
|
||||||
|
default = 1000
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "orphaned_d2c_telemetry_egress_threshold_warning" {
|
||||||
|
description = "D2C Telemetry Orphaned limit (warning threshold)"
|
||||||
|
default = 500
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "orphaned_d2c_telemetry_egress_threshold_critical" {
|
||||||
|
description = "D2C Telemetry Orphaned limit (critical threshold)"
|
||||||
|
default = 1000
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "invalid_d2c_telemetry_egress_threshold_warning" {
|
||||||
|
description = "D2C Telemetry Invalid limit (warning threshold)"
|
||||||
|
default = 500
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "invalid_d2c_telemetry_egress_threshold_critical" {
|
||||||
|
description = "D2C Telemetry Invalid limit (critical threshold)"
|
||||||
|
default = 1000
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "fallback_d2c_telemetry_egress_threshold_warning" {
|
||||||
|
description = "D2C Telemetry Fallback limit (warning threshold)"
|
||||||
|
default = 500
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "fallback_d2c_telemetry_egress_threshold_critical" {
|
||||||
|
description = "D2C Telemetry Fallback limit (critical threshold)"
|
||||||
|
default = 1000
|
||||||
|
}
|
||||||
470
cloud/azure/iothubs/monitors-iothubs.tf
Normal file
470
cloud/azure/iothubs/monitors-iothubs.tf
Normal file
@ -0,0 +1,470 @@
|
|||||||
|
data "template_file" "filter" {
|
||||||
|
template = "$${filter}"
|
||||||
|
|
||||||
|
vars {
|
||||||
|
filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_azure_iothub:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}"
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_jobs_failed" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many jobs failed on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):(
|
||||||
|
avg:azure.devices_iothubs.jobs.failed{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
|
||||||
|
( avg:azure.devices_iothubs.jobs.failed{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
|
||||||
|
avg:azure.devices_iothubs.jobs.completed{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
|
||||||
|
) * 100 > ${var.failed_jobs_rate_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.failed_jobs_rate_threshold_warning}"
|
||||||
|
critical = "${var.failed_jobs_rate_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_list_jobs_failed" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many list_jobs failure on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):(
|
||||||
|
avg:azure.devices_iothubs.jobs.list_jobs.failure{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() /
|
||||||
|
( avg:azure.devices_iothubs.jobs.list_jobs.success{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() +
|
||||||
|
avg:azure.devices_iothubs.jobs.list_jobs.failure{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() )
|
||||||
|
) * 100 > ${var.failed_listjobs_rate_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.failed_listjobs_rate_threshold_warning}"
|
||||||
|
critical = "${var.failed_listjobs_rate_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_query_jobs_failed" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many query_jobs failed on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):(
|
||||||
|
avg:azure.devices_iothubs.jobs.query_jobs.failure{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() /
|
||||||
|
( avg:azure.devices_iothubs.jobs.query_jobs.success{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() +
|
||||||
|
avg:azure.devices_iothubs.jobs.query_jobs.failure{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() )
|
||||||
|
) * 100 > ${var.failed_queryjobs_rate_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.failed_queryjobs_rate_threshold_warning}"
|
||||||
|
critical = "${var.failed_queryjobs_rate_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "status" {
|
||||||
|
name = "[${var.environment}] IOT Hub Status is not ok on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):avg:azure.devices_iothubs.status{${data.template_file.filter.rendered}} by {name,resource_group} < 1
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
notify_no_data = true
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "total_devices" {
|
||||||
|
name = "[${var.environment}] IOT Hub Total devices is wrong on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):avg:azure.devices_iothubs.devices.total_devices{${data.template_file.filter.rendered}} by {name,resource_group} == 0
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
notify_no_data = true
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_c2d_methods_failed" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many c2d methods failure on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):(
|
||||||
|
avg:azure.devices_iothubs.c2d.methods.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
|
||||||
|
( avg:azure.devices_iothubs.c2d.methods.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
|
||||||
|
avg:azure.devices_iothubs.c2d.methods.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
|
||||||
|
) * 100 > ${var.failed_c2d_methods_rate_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.failed_c2d_methods_rate_threshold_warning}"
|
||||||
|
critical = "${var.failed_c2d_methods_rate_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many c2d twin read failure on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):(
|
||||||
|
avg:azure.devices_iothubs.c2d.twin.read.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
|
||||||
|
( avg:azure.devices_iothubs.c2d.twin.read.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
|
||||||
|
avg:azure.devices_iothubs.c2d.twin.read.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
|
||||||
|
) * 100 > ${var.failed_c2d_twin_read_rate_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.failed_c2d_twin_read_rate_threshold_warning}"
|
||||||
|
critical = "${var.failed_c2d_twin_read_rate_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many c2d twin update failure on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):(
|
||||||
|
avg:azure.devices_iothubs.c2d.twin.update.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
|
||||||
|
( avg:azure.devices_iothubs.c2d.twin.update.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
|
||||||
|
avg:azure.devices_iothubs.c2d.twin.update.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
|
||||||
|
) * 100 > ${var.failed_c2d_twin_update_rate_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.failed_c2d_twin_update_rate_threshold_warning}"
|
||||||
|
critical = "${var.failed_c2d_twin_update_rate_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many d2c twin read failure on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):(
|
||||||
|
avg:azure.devices_iothubs.d2c.twin.read.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
|
||||||
|
( avg:azure.devices_iothubs.d2c.twin.read.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
|
||||||
|
avg:azure.devices_iothubs.d2c.twin.read.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
|
||||||
|
) * 100 > ${var.failed_d2c_twin_read_rate_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.failed_d2c_twin_read_rate_threshold_warning}"
|
||||||
|
critical = "${var.failed_d2c_twin_read_rate_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many d2c twin update failure on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
avg(last_5m):(
|
||||||
|
avg:azure.devices_iothubs.d2c.twin.update.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
|
||||||
|
( avg:azure.devices_iothubs.d2c.twin.update.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
|
||||||
|
avg:azure.devices_iothubs.d2c.twin.update.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
|
||||||
|
) * 100 > ${var.failed_d2c_twin_update_rate_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.failed_d2c_twin_update_rate_threshold_warning}"
|
||||||
|
critical = "${var.failed_d2c_twin_update_rate_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many d2c telemetry egress dropped on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
sum(last_5m): (
|
||||||
|
avg:azure.devices_iothubs.d2c.telemetry.egress.dropped{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
|
||||||
|
) > ${var.dropped_d2c_telemetry_egress_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.dropped_d2c_telemetry_egress_threshold_warning}"
|
||||||
|
critical = "${var.dropped_d2c_telemetry_egress_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many d2c telemetry egress orphaned on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
sum(last_5m): (
|
||||||
|
avg:azure.devices_iothubs.d2c.telemetry.egress.orphaned{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
|
||||||
|
) > ${var.orphaned_d2c_telemetry_egress_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.orphaned_d2c_telemetry_egress_threshold_warning}"
|
||||||
|
critical = "${var.orphaned_d2c_telemetry_egress_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many d2c telemetry egress invalid on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
sum(last_5m): (
|
||||||
|
avg:azure.devices_iothubs.d2c.telemetry.egress.invalid{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
|
||||||
|
) > ${var.invalid_d2c_telemetry_egress_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.invalid_d2c_telemetry_egress_threshold_warning}"
|
||||||
|
critical = "${var.invalid_d2c_telemetry_egress_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many d2c telemetry egress fallback on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
sum(last_5m): (
|
||||||
|
avg:azure.devices_iothubs.d2c.telemetry.egress.fallback{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
|
||||||
|
) > ${var.fallback_d2c_telemetry_egress_threshold_critical}
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
thresholds {
|
||||||
|
warning = "${var.fallback_d2c_telemetry_egress_threshold_warning}"
|
||||||
|
critical = "${var.fallback_d2c_telemetry_egress_threshold_critical}"
|
||||||
|
}
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
|
|
||||||
|
resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" {
|
||||||
|
name = "[${var.environment}] IOT Hub Too many d2c telemetry ingress no sent on {{name}}"
|
||||||
|
message = "${var.message}"
|
||||||
|
|
||||||
|
query = <<EOF
|
||||||
|
sum(last_5m): (
|
||||||
|
avg:azure.devices_iothubs.d2c.telemetry.ingress.all_protocol{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() -
|
||||||
|
avg:azure.devices_iothubs.d2c.telemetry.ingress.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
|
||||||
|
) > 0
|
||||||
|
EOF
|
||||||
|
|
||||||
|
type = "metric alert"
|
||||||
|
|
||||||
|
notify_no_data = false
|
||||||
|
evaluation_delay = "${var.delay}"
|
||||||
|
renotify_interval = 0
|
||||||
|
notify_audit = false
|
||||||
|
timeout_h = 0
|
||||||
|
include_tags = true
|
||||||
|
locked = false
|
||||||
|
require_full_window = true
|
||||||
|
new_host_delay = "${var.delay}"
|
||||||
|
no_data_timeframe = 20
|
||||||
|
|
||||||
|
tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
|
||||||
|
}
|
||||||
Loading…
x
Reference in New Issue
Block a user