Merged in MON-78_Azure_Stream_analytics_monitor (pull request #13)
MON-78: Azure Stream analytics monitor Approved-by: Quentin Manfroi <quentin.manfroi@yahoo.fr> Approved-by: Alexandre Gaillet <alexandre.gaillet@fr.clara.net> Approved-by: Laurent Piroelle <laurent.piroelle@fr.clara.net>
This commit is contained in:
jeremy.nancel@fr.clara.net
Jérôme Respaut
commit
cec5e7ef4b
41
cloud/azure/stream-analytics/README.md
Normal file
41
cloud/azure/stream-analytics/README.md
Normal file
@ -0,0 +1,41 @@
|
||||
Azure Stream Analytics DataDog monitors
|
||||
=======================================
|
||||
|
||||
How to use this module
|
||||
----------------------
|
||||
|
||||
```
|
||||
module "datadog-monitors-azure-redis" {
|
||||
source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/azure/stream-analytics?ref={revision}"
|
||||
|
||||
message = "${module.datadog-message-alerting.alerting-message}"
|
||||
environment = "${var.environment}"
|
||||
subscription_id = "${var.subscription_id}"
|
||||
}
|
||||
```
|
||||
|
||||
Inputs
|
||||
------
|
||||
|
||||
| Name | Description | Type | Default | Required |
|
||||
|------|-------------|:----:|:-----:|:-----:|
|
||||
| conversion_errors_threshold_critical | Conversion errors limit (critical threshold) | string | `10` | no |
|
||||
| conversion_errors_threshold_warning | Conversion errors limit (warning threshold) | string | `0` | no |
|
||||
| delay | Delay in seconds for the metric evaluation | string | `600` | no |
|
||||
| environment | Architecture environment | string | - | yes |
|
||||
| function_requests_threshold_critical | Failed Function Request rate limit (critical threshold) | string | `10` | no |
|
||||
| function_requests_threshold_warning | Failed Function Request rate limit (warning threshold) | string | `0` | no |
|
||||
| message | Message sent when a monitor is triggered | string | - | yes |
|
||||
| provider | What is the monitored provider | string | azure | no |
|
||||
| runtime_errors_threshold_critical | | string | `10` | no |
|
||||
| runtime_errors_threshold_warning | | string | `0` | no |
|
||||
| su_utilization_threshold_critical | | string | `80` | no |
|
||||
| su_utilization_threshold_warning | Monitor specific | string | `60` | no |
|
||||
| service | What is the monitored service | string | storage | no |
|
||||
| subscription_id | Azure account id used as filter for monitors | string | - | yes |
|
||||
| use_filter_tags | Filter the data with service tags if true | string | `true` | no |
|
||||
|
||||
Related documentation
|
||||
---------------------
|
||||
|
||||
DataDog documentation: [https://docs.datadoghq.com/integrations/azure/](https://docs.datadoghq.com/integrations/azure/)
|
||||
66
cloud/azure/stream-analytics/inputs.tf
Normal file
66
cloud/azure/stream-analytics/inputs.tf
Normal file
@ -0,0 +1,66 @@
|
||||
# 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 = 600
|
||||
}
|
||||
|
||||
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 Stream Analytics specific
|
||||
variable "su_utilization_threshold_warning" {
|
||||
description = "Streaming Unit utilization rate limit (warning threshold)"
|
||||
default = 60
|
||||
}
|
||||
|
||||
variable "su_utilization_threshold_critical" {
|
||||
description = "Streaming Unit utilization rate limit (critical threshold)"
|
||||
default = 80
|
||||
}
|
||||
|
||||
variable "function_requests_threshold_warning" {
|
||||
description = "Failed Function Request rate limit (warning threshold)"
|
||||
default = 0
|
||||
}
|
||||
|
||||
variable "failed_function_requests_threshold_critical" {
|
||||
description = "Failed Function Request rate limit (critical threshold)"
|
||||
default = 10
|
||||
}
|
||||
|
||||
variable "conversion_errors_threshold_warning" {
|
||||
description = "Conversion errors limit (warning threshold)"
|
||||
default = 0
|
||||
}
|
||||
|
||||
variable "conversion_errors_threshold_critical" {
|
||||
description = "Conversion errors limit (critical threshold)"
|
||||
default = 10
|
||||
}
|
||||
|
||||
variable "runtime_errors_threshold_warning" {
|
||||
description = "Runtime errors limit (warning threshold)"
|
||||
default = 0
|
||||
}
|
||||
|
||||
variable "runtime_errors_threshold_critical" {
|
||||
description = "Runtime errors limit (critical threshold)"
|
||||
default = 10
|
||||
}
|
||||
147
cloud/azure/stream-analytics/monitors-stream-analytics.tf
Normal file
147
cloud/azure/stream-analytics/monitors-stream-analytics.tf
Normal file
@ -0,0 +1,147 @@
|
||||
data "template_file" "filter" {
|
||||
template = "$${filter}"
|
||||
|
||||
vars {
|
||||
filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_azure_streamanalytics:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}"
|
||||
}
|
||||
}
|
||||
|
||||
resource "datadog_monitor" "status" {
|
||||
name = "[${var.environment}] Stream Analytics Status is not ok on {{name}}"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
avg(last_5m):avg:azure.streamanalytics_streamingjobs.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:streamanalytics", "team:azure", "provider:azure"]
|
||||
}
|
||||
|
||||
resource "datadog_monitor" "su_utilization" {
|
||||
name = "[${var.environment}] Stream Analytics streaming Units utilization at more than ${var.su_utilization_threshold_critical}% on {{name}}"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
avg(last_5m): (
|
||||
avg:azure.streamanalytics_streamingjobs.resource_utilization{${data.template_file.filter.rendered}} by {name,resource_group}
|
||||
) > ${var.su_utilization_threshold_critical}
|
||||
EOF
|
||||
type = "metric alert"
|
||||
|
||||
notify_no_data = false
|
||||
evaluation_delay = "${var.delay}"
|
||||
renotify_interval = 0
|
||||
notify_audit = false
|
||||
timeout_h = 0
|
||||
include_tags = true
|
||||
locked = false
|
||||
require_full_window = true
|
||||
new_host_delay = "${var.delay}"
|
||||
no_data_timeframe = 20
|
||||
thresholds {
|
||||
warning = "${var.su_utilization_threshold_warning}"
|
||||
critical = "${var.su_utilization_threshold_critical}"
|
||||
}
|
||||
|
||||
tags = ["env:${var.environment}", "resource:streamanalytics", "team:azure", "provider:azure"]
|
||||
}
|
||||
|
||||
resource "datadog_monitor" "failed_function_requests" {
|
||||
name = "[${var.environment}] Stream Analytics more than ${var.failed_function_requests_threshold_critical} failed function requests on {{name}}"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
avg(last_5m): (
|
||||
avg:azure.streamanalytics_streamingjobs.aml_callout_failed_requests{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
|
||||
avg:azure.streamanalytics_streamingjobs.aml_callout_requests{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
|
||||
) * 100 > ${var.failed_function_requests_threshold_critical}
|
||||
EOF
|
||||
type = "metric alert"
|
||||
|
||||
notify_no_data = false
|
||||
evaluation_delay = "${var.delay}"
|
||||
renotify_interval = 60
|
||||
notify_audit = false
|
||||
timeout_h = 0
|
||||
include_tags = true
|
||||
locked = false
|
||||
require_full_window = true
|
||||
new_host_delay = "${var.delay}"
|
||||
no_data_timeframe = 20
|
||||
thresholds {
|
||||
warning = "${var.function_requests_threshold_warning}"
|
||||
critical = "${var.failed_function_requests_threshold_critical}"
|
||||
}
|
||||
|
||||
tags = ["env:${var.environment}", "resource:streamanalytics", "team:azure", "provider:azure"]
|
||||
}
|
||||
|
||||
resource "datadog_monitor" "conversion_errors" {
|
||||
name = "[${var.environment}] Stream Analytics more than ${var.conversion_errors_threshold_critical} conversion errors on {{name}}"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
avg(last_5m): (
|
||||
avg:azure.streamanalytics_streamingjobs.conversion_errors{${data.template_file.filter.rendered}} by {name,resource_group}
|
||||
) > ${var.conversion_errors_threshold_critical}
|
||||
EOF
|
||||
type = "metric alert"
|
||||
|
||||
notify_no_data = false
|
||||
evaluation_delay = "${var.delay}"
|
||||
renotify_interval = 0
|
||||
notify_audit = false
|
||||
timeout_h = 0
|
||||
include_tags = true
|
||||
locked = false
|
||||
require_full_window = true
|
||||
new_host_delay = "${var.delay}"
|
||||
no_data_timeframe = 20
|
||||
thresholds {
|
||||
warning = "${var.conversion_errors_threshold_warning}"
|
||||
critical = "${var.conversion_errors_threshold_critical}"
|
||||
}
|
||||
|
||||
tags = ["env:${var.environment}", "resource:streamanalytics", "team:azure", "provider:azure"]
|
||||
}
|
||||
|
||||
resource "datadog_monitor" "runtime_errors" {
|
||||
name = "[${var.environment}] Stream Analytics more than ${var.runtime_errors_threshold_critical} runtime errors on {{name}}"
|
||||
message = "${var.message}"
|
||||
|
||||
query = <<EOF
|
||||
avg(last_5m): (
|
||||
avg:azure.streamanalytics_streamingjobs.errors{${data.template_file.filter.rendered}} by {name,resource_group}
|
||||
) > ${var.runtime_errors_threshold_critical}
|
||||
EOF
|
||||
type = "metric alert"
|
||||
|
||||
notify_no_data = false
|
||||
evaluation_delay = "${var.delay}"
|
||||
renotify_interval = 0
|
||||
notify_audit = false
|
||||
timeout_h = 0
|
||||
include_tags = true
|
||||
locked = false
|
||||
require_full_window = true
|
||||
new_host_delay = "${var.delay}"
|
||||
no_data_timeframe = 20
|
||||
thresholds {
|
||||
warning = "${var.runtime_errors_threshold_warning}"
|
||||
critical = "${var.runtime_errors_threshold_critical}"
|
||||
}
|
||||
|
||||
tags = ["env:${var.environment}", "resource:streamanalytics", "team:azure", "provider:azure"]
|
||||
}
|
||||
Loading…
x
Reference in New Issue
Block a user