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

MON-73 Best practices : rename Azure monitors and set require full window to false

Browse Source
This commit is contained in:
Laurent Piroelle 2018-02-13 16:49:37 +01:00 committed by Quentin Manfroi
parent fed0d592e9
commit e43f07d913
8 changed files with 101 additions and 101 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
cloud/azure/apimanagement/monitors-azure-apimanagement.tf
View File

@ -9,7 +9,7 @@ data "template_file" "filter" {
} }
resource "datadog_monitor" "apimgt_status" { resource "datadog_monitor" "apimgt_status" {
name = "[${var.environment}] API Management status is not ok on {{name}}" name = "[${var.environment}] API Management is down"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -29,7 +29,7 @@ resource "datadog_monitor" "apimgt_status" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -37,7 +37,7 @@ resource "datadog_monitor" "apimgt_status" {
} }
resource "datadog_monitor" "apimgt_failed_requests" { resource "datadog_monitor" "apimgt_failed_requests" {
name = "[${var.environment}] API Management {{name}} too much failed requests" name = "[${var.environment}] API Management too many failed requests {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -58,7 +58,7 @@ resource "datadog_monitor" "apimgt_failed_requests" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -68,7 +68,7 @@ resource "datadog_monitor" "apimgt_failed_requests" {
} }
resource "datadog_monitor" "apimgt_other_requests" { resource "datadog_monitor" "apimgt_other_requests" {
name = "[${var.environment}] API Management {{name}} too much other requests" name = "[${var.environment}] API Management too many other requests {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -89,7 +89,7 @@ resource "datadog_monitor" "apimgt_other_requests" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -99,7 +99,7 @@ resource "datadog_monitor" "apimgt_other_requests" {
} }
resource "datadog_monitor" "apimgt_unauthorized_requests" { resource "datadog_monitor" "apimgt_unauthorized_requests" {
name = "[${var.environment}] API Management {{name}} too much unauthorized requests" name = "[${var.environment}] API Management too many unauthorized requests {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -120,7 +120,7 @@ resource "datadog_monitor" "apimgt_unauthorized_requests" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -130,7 +130,7 @@ resource "datadog_monitor" "apimgt_unauthorized_requests" {
} }
resource "datadog_monitor" "apimgt_successful_requests" { resource "datadog_monitor" "apimgt_successful_requests" {
name = "[${var.environment}] API Management {{name}} successful requests rate too low" name = "[${var.environment}] API Management successful requests rate too low {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -151,7 +151,7 @@ resource "datadog_monitor" "apimgt_successful_requests" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0

24
cloud/azure/app-services/monitors-app_services.tf
View File

@ -8,7 +8,7 @@ data "template_file" "filter" {
# Monitoring App Services response time # Monitoring App Services response time
resource "datadog_monitor" "appservices_response_time" { resource "datadog_monitor" "appservices_response_time" {
name = "[${var.environment}] App Services response time of {{value}}s is to high on {{name}}" name = "[${var.environment}] App Services response time too high {{comparator}} {{#is_alert}}{{threshold}}s{{/is_alert}}{{#is_warning}}{{warn_threshold}}s{{/is_warning}} ({{value}}s)"
type = "metric alert" type = "metric alert"
message = "${var.message}" message = "${var.message}"
@ -26,9 +26,9 @@ resource "datadog_monitor" "appservices_response_time" {
critical = "${var.response_time_threshold_critical}" critical = "${var.response_time_threshold_critical}"
} }
notify_no_data = true # Will notify when no data is received notify_no_data = true # Will notify when no data is received
renotify_interval = 0 renotify_interval = 0
require_full_window = true require_full_window = false
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
@ -37,7 +37,7 @@ resource "datadog_monitor" "appservices_response_time" {
# Monitoring App Services memory usage # Monitoring App Services memory usage
resource "datadog_monitor" "appservices_memory_usage_count" { resource "datadog_monitor" "appservices_memory_usage_count" {
name = "[${var.environment}] App Services memory usage > ${ceil(var.memory_usage_threshold_critical/1000000)}MiB on {{name}}" name = "[${var.environment}] App Services memory usage {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
type = "metric alert" type = "metric alert"
message = "${var.message}" message = "${var.message}"
@ -55,9 +55,9 @@ resource "datadog_monitor" "appservices_memory_usage_count" {
critical = "${var.memory_usage_threshold_critical}" critical = "${var.memory_usage_threshold_critical}"
} }
notify_no_data = true # Will notify when no data is received notify_no_data = true # Will notify when no data is received
renotify_interval = 0 renotify_interval = 0
require_full_window = true require_full_window = false
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
@ -66,7 +66,7 @@ resource "datadog_monitor" "appservices_memory_usage_count" {
# Monitoring App Services 5xx errors percent # Monitoring App Services 5xx errors percent
resource "datadog_monitor" "appservices_http_5xx_errors_count" { resource "datadog_monitor" "appservices_http_5xx_errors_count" {
name = "[${var.environment}] App Services HTTP 5xx errors is {{value}}% above the limit on {{name}}" name = "[${var.environment}] App Services HTTP 5xx errors too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
type = "metric alert" type = "metric alert"
message = "${var.message}" message = "${var.message}"
@ -87,7 +87,7 @@ resource "datadog_monitor" "appservices_http_5xx_errors_count" {
notify_no_data = false # Will NOT notify when no data is received notify_no_data = false # Will NOT notify when no data is received
renotify_interval = 0 renotify_interval = 0
require_full_window = true require_full_window = false
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
@ -96,7 +96,7 @@ resource "datadog_monitor" "appservices_http_5xx_errors_count" {
# Monitoring App Services 4xx errors percent # Monitoring App Services 4xx errors percent
resource "datadog_monitor" "appservices_http_4xx_errors_count" { resource "datadog_monitor" "appservices_http_4xx_errors_count" {
name = "[${var.environment}] App Services HTTP 4xx errors is {{value}}% above the limit on {{name}}" name = "[${var.environment}] App Services HTTP 4xx errors too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
type = "metric alert" type = "metric alert"
message = "${var.message}" message = "${var.message}"
@ -117,7 +117,7 @@ resource "datadog_monitor" "appservices_http_4xx_errors_count" {
notify_no_data = false # Will NOT notify when no data is received notify_no_data = false # Will NOT notify when no data is received
renotify_interval = 0 renotify_interval = 0
require_full_window = true require_full_window = false
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
@ -126,7 +126,7 @@ resource "datadog_monitor" "appservices_http_4xx_errors_count" {
# Monitoring App Services HTTP 2xx & 3xx status pages percent # Monitoring App Services HTTP 2xx & 3xx status pages percent
resource "datadog_monitor" "appservices_http_success_status_rate" { resource "datadog_monitor" "appservices_http_success_status_rate" {
name = "[${var.environment}] App Services HTTP successful responses is {{value}}% below the limit on {{name}}" name = "[${var.environment}] App Services HTTP successful responses too low {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
type = "metric alert" type = "metric alert"
message = "${var.message}" message = "${var.message}"
@ -148,7 +148,7 @@ resource "datadog_monitor" "appservices_http_success_status_rate" {
notify_no_data = false # Will notify when no data is received notify_no_data = false # Will notify when no data is received
renotify_interval = 0 renotify_interval = 0
require_full_window = true require_full_window = false
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true

12
cloud/azure/eventhub/monitors-eventhub.tf
View File

@ -7,7 +7,7 @@ data "template_file" "filter" {
} }
resource "datadog_monitor" "eventhub_status" { resource "datadog_monitor" "eventhub_status" {
name = "[${var.environment}] Event Hub status is not ok on {{name}}" name = "[${var.environment}] Event Hub is down"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -23,7 +23,7 @@ resource "datadog_monitor" "eventhub_status" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -31,7 +31,7 @@ resource "datadog_monitor" "eventhub_status" {
} }
resource "datadog_monitor" "eventhub_failed_requests" { resource "datadog_monitor" "eventhub_failed_requests" {
name = "[${var.environment}] Event Hub too much failed requests on {{name}}" name = "[${var.environment}] Event Hub too many failed requests {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -57,7 +57,7 @@ resource "datadog_monitor" "eventhub_failed_requests" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -65,7 +65,7 @@ resource "datadog_monitor" "eventhub_failed_requests" {
} }
resource "datadog_monitor" "eventhub_errors" { resource "datadog_monitor" "eventhub_errors" {
name = "[${var.environment}] Event Hub too much errors on {{name}}" name = "[${var.environment}] Event Hub too manny errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -95,7 +95,7 @@ resource "datadog_monitor" "eventhub_errors" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20

60
cloud/azure/iothubs/monitors-iothubs.tf
View File

@ -1,5 +1,5 @@
resource "datadog_monitor" "too_many_jobs_failed" { resource "datadog_monitor" "too_many_jobs_failed" {
name = "[${var.environment}] IOT Hub Too many jobs failed on {{name}}" name = "[${var.environment}] IOT Hub Too many jobs failed {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -24,7 +24,7 @@ resource "datadog_monitor" "too_many_jobs_failed" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -32,7 +32,7 @@ resource "datadog_monitor" "too_many_jobs_failed" {
} }
resource "datadog_monitor" "too_many_list_jobs_failed" { resource "datadog_monitor" "too_many_list_jobs_failed" {
name = "[${var.environment}] IOT Hub Too many list_jobs failure on {{name}}" name = "[${var.environment}] IOT Hub Too many list_jobs failure {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -57,7 +57,7 @@ resource "datadog_monitor" "too_many_list_jobs_failed" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -65,7 +65,7 @@ resource "datadog_monitor" "too_many_list_jobs_failed" {
} }
resource "datadog_monitor" "too_many_query_jobs_failed" { resource "datadog_monitor" "too_many_query_jobs_failed" {
name = "[${var.environment}] IOT Hub Too many query_jobs failed on {{name}}" name = "[${var.environment}] IOT Hub Too many query_jobs failed {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -90,7 +90,7 @@ resource "datadog_monitor" "too_many_query_jobs_failed" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -98,7 +98,7 @@ resource "datadog_monitor" "too_many_query_jobs_failed" {
} }
resource "datadog_monitor" "status" { resource "datadog_monitor" "status" {
name = "[${var.environment}] IOT Hub Status is not ok on {{name}}" name = "[${var.environment}] IOT Hub is down"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -114,7 +114,7 @@ resource "datadog_monitor" "status" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -122,7 +122,7 @@ resource "datadog_monitor" "status" {
} }
resource "datadog_monitor" "total_devices" { resource "datadog_monitor" "total_devices" {
name = "[${var.environment}] IOT Hub Total devices is wrong on {{name}}" name = "[${var.environment}] IOT Hub Total devices is wrong {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -138,7 +138,7 @@ resource "datadog_monitor" "total_devices" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -146,7 +146,7 @@ resource "datadog_monitor" "total_devices" {
} }
resource "datadog_monitor" "too_many_c2d_methods_failed" { resource "datadog_monitor" "too_many_c2d_methods_failed" {
name = "[${var.environment}] IOT Hub Too many c2d methods failure on {{name}}" name = "[${var.environment}] IOT Hub Too many c2d methods failure {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -171,7 +171,7 @@ resource "datadog_monitor" "too_many_c2d_methods_failed" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -179,7 +179,7 @@ resource "datadog_monitor" "too_many_c2d_methods_failed" {
} }
resource "datadog_monitor" "too_many_c2d_twin_read_failed" { resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
name = "[${var.environment}] IOT Hub Too many c2d twin read failure on {{name}}" name = "[${var.environment}] IOT Hub Too many c2d twin read failure {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -204,7 +204,7 @@ resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -212,7 +212,7 @@ resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
} }
resource "datadog_monitor" "too_many_c2d_twin_update_failed" { resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
name = "[${var.environment}] IOT Hub Too many c2d twin update failure on {{name}}" name = "[${var.environment}] IOT Hub Too many c2d twin update failure {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -237,7 +237,7 @@ resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -245,7 +245,7 @@ resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
} }
resource "datadog_monitor" "too_many_d2c_twin_read_failed" { resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
name = "[${var.environment}] IOT Hub Too many d2c twin read failure on {{name}}" name = "[${var.environment}] IOT Hub Too many d2c twin read failure {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -270,7 +270,7 @@ resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -278,7 +278,7 @@ resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
} }
resource "datadog_monitor" "too_many_d2c_twin_update_failed" { resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
name = "[${var.environment}] IOT Hub Too many d2c twin update failure on {{name}}" name = "[${var.environment}] IOT Hub Too many d2c twin update failure {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -303,7 +303,7 @@ resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -311,7 +311,7 @@ resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
} }
resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" { resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
name = "[${var.environment}] IOT Hub Too many d2c telemetry egress dropped on {{name}}" name = "[${var.environment}] IOT Hub Too many d2c telemetry egress dropped {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -334,7 +334,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -342,7 +342,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
} }
resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" { resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
name = "[${var.environment}] IOT Hub Too many d2c telemetry egress orphaned on {{name}}" name = "[${var.environment}] IOT Hub Too many d2c telemetry egress orphaned {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -365,7 +365,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -373,7 +373,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
} }
resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" { resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
name = "[${var.environment}] IOT Hub Too many d2c telemetry egress invalid on {{name}}" name = "[${var.environment}] IOT Hub Too many d2c telemetry egress invalid {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -396,7 +396,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -404,7 +404,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
} }
resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" { resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" {
name = "[${var.environment}] IOT Hub Too many d2c telemetry egress fallback on {{name}}" name = "[${var.environment}] IOT Hub Too many d2c telemetry egress fallback {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -427,7 +427,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -435,7 +435,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" {
} }
resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" { resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" {
name = "[${var.environment}] IOT Hub Too many d2c telemetry ingress no sent on {{name}}" name = "[${var.environment}] IOT Hub Too many d2c telemetry ingress not sent {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -454,7 +454,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20

14
cloud/azure/redis/monitors-azure-redis.tf
View File

@ -23,7 +23,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -31,7 +31,7 @@ EOF
} }
resource "datadog_monitor" "evictedkeys" { resource "datadog_monitor" "evictedkeys" {
name = "[${var.environment}] Redis {{value}} evictedkeys on {{name}}" name = "[${var.environment}] Redis too many evictedkeys {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -54,7 +54,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -62,7 +62,7 @@ EOF
} }
resource "datadog_monitor" "percent_processor_time" { resource "datadog_monitor" "percent_processor_time" {
name = "[${var.environment}] Redis processor time {{value}}% on {{name}}" name = "[${var.environment}] Redis processor time too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -85,7 +85,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -93,7 +93,7 @@ EOF
} }
resource "datadog_monitor" "server_load" { resource "datadog_monitor" "server_load" {
name = "[${var.environment}] Redis processor server load {{value}}% on {{name}}" name = "[${var.environment}] Redis server load too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -116,7 +116,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20

16
cloud/azure/sql-database/monitors-sql-database-basics.tf
View File

@ -7,7 +7,7 @@ data "template_file" "filter" {
} }
resource "datadog_monitor" "sql-database_cpu_90_15min" { resource "datadog_monitor" "sql-database_cpu_90_15min" {
name = "[${var.environment}] SQL Database CPU high > ${var.cpu_threshold_critical}% on {{name}}" name = "[${var.environment}] SQL Database CPU too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -29,7 +29,7 @@ resource "datadog_monitor" "sql-database_cpu_90_15min" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -37,7 +37,7 @@ resource "datadog_monitor" "sql-database_cpu_90_15min" {
} }
resource "datadog_monitor" "sql-database_free_space_low" { resource "datadog_monitor" "sql-database_free_space_low" {
name = "[${var.environment}] SQL Database free space < ${var.diskspace_threshold_critical}% on {{name}}" name = "[${var.environment}] SQL Database low free space {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
type = "metric alert" type = "metric alert"
@ -60,7 +60,7 @@ resource "datadog_monitor" "sql-database_free_space_low" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -68,7 +68,7 @@ resource "datadog_monitor" "sql-database_free_space_low" {
} }
resource "datadog_monitor" "sql-database_dtu_consumption_high" { resource "datadog_monitor" "sql-database_dtu_consumption_high" {
name = "[${var.environment}] SQL Database DTU Consumption on {{name}} > ${var.dtu_threshold_critical}" name = "[${var.environment}] SQL Database DTU Consumption too high {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
type = "metric alert" type = "metric alert"
@ -91,7 +91,7 @@ resource "datadog_monitor" "sql-database_dtu_consumption_high" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -99,7 +99,7 @@ resource "datadog_monitor" "sql-database_dtu_consumption_high" {
} }
resource "datadog_monitor" "sql-database_deadlocks_count" { resource "datadog_monitor" "sql-database_deadlocks_count" {
name = "[${var.environment}] SQL Database Deadlocks too high on {{name}}" name = "[${var.environment}] SQL Database Deadlocks too high {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
type = "metric alert" type = "metric alert"
@ -121,7 +121,7 @@ resource "datadog_monitor" "sql-database_deadlocks_count" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20

36
cloud/azure/storage/monitors-azure-storage.tf
View File

@ -7,7 +7,7 @@ data "template_file" "filter" {
} }
resource "datadog_monitor" "availability" { resource "datadog_monitor" "availability" {
name = "[${var.environment}] Azure Storage {{name}} unavailability detected" name = "[${var.environment}] Azure Storage is down"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -27,7 +27,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -37,7 +37,7 @@ EOF
} }
resource "datadog_monitor" "successful_requests" { resource "datadog_monitor" "successful_requests" {
name = "[${var.environment}] Azure Storage {{name}} too much failed requests" name = "[${var.environment}] Azure Storage too few successful requests {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -57,7 +57,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -67,7 +67,7 @@ EOF
} }
resource "datadog_monitor" "latency" { resource "datadog_monitor" "latency" {
name = "[${var.environment}] Azure Storage {{name}} too high end to end latency" name = "[${var.environment}] Azure Storage too high end to end latency {{comparator}} {{#is_alert}}{{threshold}}ms{{/is_alert}}{{#is_warning}}{{warn_threshold}}ms{{/is_warning}} ({{value}}ms)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -87,7 +87,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -97,7 +97,7 @@ EOF
} }
resource "datadog_monitor" "timeout_error_requests" { resource "datadog_monitor" "timeout_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of timeout error requests on {{name}}" name = "[${var.environment}] Azure Storage too many timeout errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -117,7 +117,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -127,7 +127,7 @@ EOF
} }
resource "datadog_monitor" "network_error_requests" { resource "datadog_monitor" "network_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of network error requests on {{name}}" name = "[${var.environment}] Azure Storage too many network errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -147,7 +147,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -157,7 +157,7 @@ EOF
} }
resource "datadog_monitor" "throttling_error_requests" { resource "datadog_monitor" "throttling_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of throttling error requests on {{name}}" name = "[${var.environment}] Azure Storage too many throttling errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -177,7 +177,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -187,7 +187,7 @@ EOF
} }
resource "datadog_monitor" "server_other_error_requests" { resource "datadog_monitor" "server_other_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of server_other error requests on {{name}}" name = "[${var.environment}] Azure Storage too many server_other errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -207,7 +207,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -217,7 +217,7 @@ EOF
} }
resource "datadog_monitor" "client_other_error_requests" { resource "datadog_monitor" "client_other_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of client_other error requests on {{name}}" name = "[${var.environment}] Azure Storage too many client_other errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -237,7 +237,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0
@ -247,7 +247,7 @@ EOF
} }
resource "datadog_monitor" "authorization_error_requests" { resource "datadog_monitor" "authorization_error_requests" {
name = "[${var.environment}] Azure Storage {{value}}% of authorization error requests on {{name}}" name = "[${var.environment}] Azure Storage too many authorization errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -267,7 +267,7 @@ EOF
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
renotify_interval = 0 renotify_interval = 0

20
cloud/azure/stream-analytics/monitors-stream-analytics.tf
View File

@ -7,7 +7,7 @@ data "template_file" "filter" {
} }
resource "datadog_monitor" "status" { resource "datadog_monitor" "status" {
name = "[${var.environment}] Stream Analytics Status is not ok on {{name}}" name = "[${var.environment}] Stream Analytics is down"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -23,7 +23,7 @@ resource "datadog_monitor" "status" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -31,7 +31,7 @@ resource "datadog_monitor" "status" {
} }
resource "datadog_monitor" "su_utilization" { resource "datadog_monitor" "su_utilization" {
name = "[${var.environment}] Stream Analytics streaming Units utilization at more than ${var.su_utilization_threshold_critical}% on {{name}}" name = "[${var.environment}] Stream Analytics streaming units utilization too high {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -49,7 +49,7 @@ resource "datadog_monitor" "su_utilization" {
timeout_h = 0 timeout_h = 0
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -62,7 +62,7 @@ resource "datadog_monitor" "su_utilization" {
} }
resource "datadog_monitor" "failed_function_requests" { resource "datadog_monitor" "failed_function_requests" {
name = "[${var.environment}] Stream Analytics more than ${var.failed_function_requests_threshold_critical} failed function requests on {{name}}" name = "[${var.environment}] Stream Analytics too many failed requests {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -81,7 +81,7 @@ resource "datadog_monitor" "failed_function_requests" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -94,7 +94,7 @@ resource "datadog_monitor" "failed_function_requests" {
} }
resource "datadog_monitor" "conversion_errors" { resource "datadog_monitor" "conversion_errors" {
name = "[${var.environment}] Stream Analytics more than ${var.conversion_errors_threshold_critical} conversion errors on {{name}}" name = "[${var.environment}] Stream Analytics too many conversion errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -112,7 +112,7 @@ resource "datadog_monitor" "conversion_errors" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
@ -125,7 +125,7 @@ resource "datadog_monitor" "conversion_errors" {
} }
resource "datadog_monitor" "runtime_errors" { resource "datadog_monitor" "runtime_errors" {
name = "[${var.environment}] Stream Analytics more than ${var.runtime_errors_threshold_critical} runtime errors on {{name}}" name = "[${var.environment}] Stream Analytics too many runtime errors {{comparator}} {{#is_alert}}{{threshold}}{{/is_alert}}{{#is_warning}}{{warn_threshold}}{{/is_warning}} ({{value}})"
message = "${var.message}" message = "${var.message}"
query = <<EOF query = <<EOF
@ -143,7 +143,7 @@ resource "datadog_monitor" "runtime_errors" {
timeout_h = 1 timeout_h = 1
include_tags = true include_tags = true
locked = false locked = false
require_full_window = true require_full_window = false
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20