MON-80 Add c2d and d2c monitors

This commit is contained in:
Alexandre Gaillet 2017-10-30 17:13:42 +01:00
parent 4c474be541
commit effaaf0e12
2 changed files with 140 additions and 4 deletions

View File

@ -43,8 +43,6 @@ variable "status_message" {}
variable "total_devices_message" {}
variable "c2d_methods_failed_message" {}
variable "c2d_methods_failed_threshold_warning" {
default = 0
}
@ -52,3 +50,45 @@ variable "c2d_methods_failed_threshold_warning" {
variable "c2d_methods_failed_threshold_critical" {
default = 10
}
variable "c2d_methods_failed_message" {}
variable "c2d_twin_read_failed_threshold_warning" {
default = 0
}
variable "c2d_twin_read_failed_threshold_critical" {
default = 10
}
variable "c2d_twin_read_failed_message" {}
variable "c2d_twin_update_failed_threshold_warning" {
default = 0
}
variable "c2d_twin_update_failed_threshold_critical" {
default = 10
}
variable "c2d_twin_update_failed_message" {}
variable "d2c_twin_read_failed_threshold_warning" {
default = 0
}
variable "d2c_twin_read_failed_threshold_critical" {
default = 10
}
variable "d2c_twin_read_failed_message" {}
variable "d2c_twin_update_failed_threshold_warning" {
default = 0
}
variable "d2c_twin_update_failed_threshold_critical" {
default = 10
}
variable "d2c_twin_update_failed_message" {}

View File

@ -131,3 +131,99 @@ resource "datadog_monitor" "too_many_c2d_methods_failed" {
new_host_delay = "${var.delay}"
no_data_timeframe = 20
}
resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
name = "[${var.environment}] Too many c2d twin read failure on {{name}} "
message = "${var.c2d_twin_read_failed_message}"
query = "avg(last_5m):( avg:azure.devices_iothubs.c2d.twin.read.failure{*} by {name,resource_group}.as_count() / ( avg:azure.devices_iothubs.c2d.twin.read.failure{*} by {name,resource_group}.as_count() + avg:azure.devices_iothubs.c2d.twin.read.success{*} by {name,resource_group}.as_count() ) ) * 100 > ${var.c2d_twin_read_failed_threshold_critical}"
type = "query alert"
thresholds {
warning = "${var.c2d_twin_read_failed_threshold_warning}"
critical = "${var.c2d_twin_read_failed_threshold_critical}"
}
notify_no_data = false
evaluation_delay = "${var.delay}"
renotify_interval = 60
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
no_data_timeframe = 20
}
resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
name = "[${var.environment}] Too many c2d twin update failure on {{name}} "
message = "${var.c2d_twin_update_failed_message}"
query = "avg(last_5m):( avg:azure.devices_iothubs.c2d.twin.update.failure{*} by {name,resource_group}.as_count() / ( avg:azure.devices_iothubs.c2d.twin.update.failure{*} by {name,resource_group}.as_count() + avg:azure.devices_iothubs.c2d.twin.update.success{*} by {name,resource_group}.as_count() ) ) * 100 > ${var.c2d_twin_update_failed_threshold_critical}"
type = "query alert"
thresholds {
warning = "${var.c2d_twin_update_failed_threshold_warning}"
critical = "${var.c2d_twin_update_failed_threshold_critical}"
}
notify_no_data = false
evaluation_delay = "${var.delay}"
renotify_interval = 60
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
no_data_timeframe = 20
}
resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
name = "[${var.environment}] Too many d2c twin read failure on {{name}} "
message = "${var.d2c_twin_read_failed_message}"
query = "avg(last_5m):( avg:azure.devices_iothubs.d2c.twin.read.failure{*} by {name,resource_group}.as_count() / ( avg:azure.devices_iothubs.d2c.twin.read.failure{*} by {name,resource_group}.as_count() + avg:azure.devices_iothubs.d2c.twin.read.success{*} by {name,resource_group}.as_count() ) ) * 100 > ${var.d2c_twin_read_failed_threshold_critical}"
type = "query alert"
thresholds {
warning = "${var.d2c_twin_read_failed_threshold_warning}"
critical = "${var.d2c_twin_read_failed_threshold_critical}"
}
notify_no_data = false
evaluation_delay = "${var.delay}"
renotify_interval = 60
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = true
new_host_delay = "${var.delay}"
no_data_timeframe = 20
}
resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
name = "[${var.environment}] Too many d2c twin update failure on {{name}} "
message = "${var.d2c_twin_update_failed_message}"
query = "avg(last_5m):( avg:azure.devices_iothubs.d2c.twin.update.failure{*} by {name,resource_group}.as_count() / ( avg:azure.devices_iothubs.d2c.twin.update.failure{*} by {name,resource_group}.as_count() + avg:azure.devices_iothubs.d2c.twin.update.success{*} by {name,resource_group}.as_count() ) ) * 100 > ${var.d2c_twin_update_failed_threshold_critical}"
type = "query alert"
thresholds {
warning = "${var.d2c_twin_update_failed_threshold_warning}"
critical = "${var.d2c_twin_update_failed_threshold_critical}"
}
notify_no_data = false
evaluation_delay = "${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
}