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

MON-309 fix default usage

Browse Source
This commit is contained in:
Quentin Manfroi 2018-10-03 16:27:32 +02:00
parent d55c0fb468
commit 9aa4364cfd
11 changed files with 58 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
caas/kubernetes/ingress/inputs.tf
View File

@ -53,6 +53,12 @@ variable "ingress_5xx_message" {
default = ""
}
variable "ingress_5xx_time_aggregator" {
description = "Monitor aggregator for Ingress 5xx errors [available values: min, max or avg]"
type = "string"
default = "min"
}
variable "ingress_5xx_timeframe" {
description = "Monitor timeframe for Ingress 5xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
type = "string"
@ -94,6 +100,12 @@ variable "ingress_4xx_message" {
default = ""
}
variable "ingress_4xx_time_aggregator" {
description = "Monitor aggregator for Ingress 4xx errors [available values: min, max or avg]"
type = "string"
default = "min"
}
variable "ingress_4xx_timeframe" {
description = "Monitor timeframe for Ingress 4xx errors [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
type = "string"

20
caas/kubernetes/ingress/monitors-ingress.tf
View File

@ -4,12 +4,10 @@ resource "datadog_monitor" "nginx_ingress_too_many_5xx" {
message = "${coalesce(var.ingress_5xx_message, var.message)}"
query = <<EOF
sum(${var.ingress_5xx_timeframe}): (
default(
avg:nginx_ingress.nginx_upstream_responses_total${module.filter-tags-5xx.query_alert} by {upstream,ingress_class} /
(avg:nginx_ingress.nginx_upstream_requests_total${module.filter-tags.query_alert} by {upstream,ingress_class} + ${var.artificial_requests_count}),
0) * 100
) > ${var.ingress_5xx_threshold_critical}
${var.ingress_5xx_time_aggregator}(${var.ingress_5xx_timeframe}): (
default(avg:nginx_ingress.nginx_upstream_responses_total{module.filter-tags-5xx.query_alert} by {upstream,ingress_class}.as_rate(), 0) / (
default(avg:nginx_ingress.nginx_upstream_requests_total${module.filter-tags.query_alert} by {upstream,ingress_class}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.ingress_5xx_threshold_critical}
EOF
type = "metric alert"
@ -40,12 +38,10 @@ resource "datadog_monitor" "nginx_ingress_too_many_4xx" {
message = "${coalesce(var.ingress_4xx_message, var.message)}"
query = <<EOF
sum(${var.ingress_4xx_timeframe}): (
default(
avg:nginx_ingress.nginx_upstream_responses_total${module.filter-tags-4xx.query_alert} by {upstream,ingress_class} /
(avg:nginx_ingress.nginx_upstream_requests_total${module.filter-tags.query_alert} by {upstream,ingress_class} + ${var.artificial_requests_count}),
0) * 100
) > ${var.ingress_4xx_threshold_critical}
${var.ingress_4xx_time_aggregator}(${var.ingress_4xx_timeframe}): (
default(avg:nginx_ingress.nginx_upstream_responses_total{module.filter-tags-4xx.query_alert} by {upstream,ingress_class}.as_rate(), 0) / (
default(avg:nginx_ingress.nginx_upstream_requests_total${module.filter-tags.query_alert} by {upstream,ingress_class}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.ingress_4xx_threshold_critical}
EOF
type = "metric alert"

8
cloud/aws/alb/monitors-alb.tf
View File

@ -71,7 +71,7 @@ resource "datadog_monitor" "ALB_httpcode_5xx" {
query = <<EOF
${var.httpcode_alb_5xx_time_aggregator}(${var.httpcode_alb_5xx_timeframe}):
default(avg:aws.applicationelb.httpcode_elb_5xx${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate(), 0) / (
default(avg:aws.applicationelb.request_count${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate(), 1) + ${var.artificial_requests_count})
default(avg:aws.applicationelb.request_count${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.httpcode_alb_5xx_threshold_critical}
EOF
@ -103,7 +103,7 @@ resource "datadog_monitor" "ALB_httpcode_4xx" {
query = <<EOF
${var.httpcode_alb_4xx_time_aggregator}(${var.httpcode_alb_4xx_timeframe}):
default(avg:aws.applicationelb.httpcode_elb_4xx${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate(), 0) / (
default(avg:aws.applicationelb.request_count${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate(), 1) + ${var.artificial_requests_count})
default(avg:aws.applicationelb.request_count${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.httpcode_alb_4xx_threshold_critical}
EOF
@ -135,7 +135,7 @@ resource "datadog_monitor" "ALB_httpcode_target_5xx" {
query = <<EOF
${var.httpcode_target_5xx_time_aggregator}(${var.httpcode_target_5xx_timeframe}):
default(avg:aws.applicationelb.httpcode_target_5xx${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate(), 0) / (
default(avg:aws.applicationelb.request_count${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate(), 1) + ${var.artificial_requests_count})
default(avg:aws.applicationelb.request_count${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.httpcode_target_5xx_threshold_critical}
EOF
@ -167,7 +167,7 @@ resource "datadog_monitor" "ALB_httpcode_target_4xx" {
query = <<EOF
${var.httpcode_target_4xx_time_aggregator}(${var.httpcode_target_4xx_timeframe}):
default(avg:aws.applicationelb.httpcode_target_4xx${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate(), 0) / (
default(avg:aws.applicationelb.request_count${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate(), 1) + ${var.artificial_requests_count})
default(avg:aws.applicationelb.request_count${module.filter-tags.query_alert} by {region,loadbalancer}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.httpcode_target_4xx_threshold_critical}
EOF

4
cloud/aws/apigateway/monitors-api.tf
View File

@ -40,7 +40,7 @@ resource "datadog_monitor" "API_http_5xx_errors_count" {
query = <<EOF
${var.http_5xx_requests_time_aggregator}(${var.http_5xx_requests_timeframe}):
default(avg:aws.apigateway.5xxerror{${var.filter_tags}} by {region,apiname,stage}.as_rate(), 0) / (
default(avg:aws.apigateway.count{${var.filter_tags}} by {region,apiname,stage}.as_rate(), 1) + ${var.artificial_requests_count})
default(avg:aws.apigateway.count{${var.filter_tags}} by {region,apiname,stage}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.http_5xx_requests_threshold_critical}
EOF
@ -73,7 +73,7 @@ resource "datadog_monitor" "API_http_4xx_errors_count" {
query = <<EOF
${var.http_4xx_requests_time_aggregator}(${var.http_4xx_requests_timeframe}):
default(avg:aws.apigateway.4xxerror{${var.filter_tags}} by {region,apiname,stage}.as_rate(), 0) / (
default(avg:aws.apigateway.count{${var.filter_tags}} by {region,apiname,stage}.as_rate(), 1) + ${var.artificial_requests_count})
default(avg:aws.apigateway.count{${var.filter_tags}} by {region,apiname,stage}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.http_4xx_requests_threshold_critical}
EOF

2
cloud/aws/elasticache/memcached/monitors-memcached.tf
View File

@ -8,7 +8,7 @@ resource "datadog_monitor" "memcached_get_hits" {
query = <<EOF
${var.get_hits_time_aggregator}(${var.get_hits_timeframe}): (
default(avg:aws.elasticache.get_hits${module.filter-tags.query_alert} by {region,cacheclusterid,cachenodeid}.as_rate(), 0) / (
default(avg:aws.elasticache.get_hits${module.filter-tags.query_alert} by {region,cacheclusterid,cachenodeid}.as_rate(), 1) +
default(avg:aws.elasticache.get_hits${module.filter-tags.query_alert} by {region,cacheclusterid,cachenodeid}.as_rate(), 0) +
default(avg:aws.elasticache.get_misses${module.filter-tags.query_alert} by {region,cacheclusterid,cachenodeid}.as_rate(), 0))
) * 100 < ${var.get_hits_threshold_critical}
EOF

2
cloud/aws/elasticache/redis/monitors-redis.tf
View File

@ -8,7 +8,7 @@ resource "datadog_monitor" "redis_cache_hits" {
query = <<EOF
${var.cache_hits_time_aggregator}(${var.cache_hits_timeframe}): (
default(avg:aws.elasticache.cache_hits${module.filter-tags.query_alert} by {region,cacheclusterid,cachenodeid}.as_rate(), 0) / (
default(avg:aws.elasticache.cache_hits${module.filter-tags.query_alert} by {region,cacheclusterid,cachenodeid}.as_rate(), 1) +
default(avg:aws.elasticache.cache_hits${module.filter-tags.query_alert} by {region,cacheclusterid,cachenodeid}.as_rate(), 0) +
default(avg:aws.elasticache.cache_misses${module.filter-tags.query_alert} by {region,cacheclusterid,cachenodeid}.as_rate(), 0))
) * 100 < ${var.cache_hits_threshold_critical}
EOF

16
cloud/aws/elb/monitors-elb.tf
View File

@ -40,8 +40,8 @@ resource "datadog_monitor" "ELB_too_much_4xx" {
query = <<EOF
sum(${var.elb_4xx_timeframe}):
default(avg:aws.elb.httpcode_elb_4xx${module.filter-tags.query_alert} by {region,loadbalancername}.as_count(), 0) / (
default(avg:aws.elb.request_count${module.filter-tags.query_alert} by {region,loadbalancername}.as_count(), 0) + ${var.artificial_requests_count})
default(avg:aws.elb.httpcode_elb_4xx${module.filter-tags.query_alert} by {region,loadbalancername}.as_rate(), 0) / (
default(avg:aws.elb.request_count${module.filter-tags.query_alert} by {region,loadbalancername}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.elb_4xx_threshold_critical}
EOF
@ -74,8 +74,8 @@ resource "datadog_monitor" "ELB_too_much_5xx" {
query = <<EOF
sum(${var.elb_5xx_timeframe}):
default(avg:aws.elb.httpcode_elb_5xx${module.filter-tags.query_alert} by {region,loadbalancername}.as_count(), 0) / (
default(avg:aws.elb.request_count${module.filter-tags.query_alert} by {region,loadbalancername}.as_count(), 0) + ${var.artificial_requests_count})
default(avg:aws.elb.httpcode_elb_5xx${module.filter-tags.query_alert} by {region,loadbalancername}.as_rate(), 0) / (
default(avg:aws.elb.request_count${module.filter-tags.query_alert} by {region,loadbalancername}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.elb_5xx_threshold_critical}
EOF
@ -108,8 +108,8 @@ resource "datadog_monitor" "ELB_too_much_4xx_backend" {
query = <<EOF
sum(${var.elb_backend_4xx_timeframe}):
default(avg:aws.elb.httpcode_backend_4xx${module.filter-tags.query_alert} by {region,loadbalancername}.as_count(), 0) / (
default(avg:aws.elb.request_count${module.filter-tags.query_alert} by {region,loadbalancername}.as_count(), 0) + ${var.artificial_requests_count})
default(avg:aws.elb.httpcode_backend_4xx${module.filter-tags.query_alert} by {region,loadbalancername}.as_rate(), 0) / (
default(avg:aws.elb.request_count${module.filter-tags.query_alert} by {region,loadbalancername}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.elb_backend_4xx_threshold_critical}
EOF
@ -142,8 +142,8 @@ resource "datadog_monitor" "ELB_too_much_5xx_backend" {
query = <<EOF
sum(${var.elb_backend_5xx_timeframe}):
default(avg:aws.elb.httpcode_backend_5xx${module.filter-tags.query_alert} by {region,loadbalancername}.as_count(), 0) / (
default(avg:aws.elb.request_count${module.filter-tags.query_alert} by {region,loadbalancername}.as_count(), 0) + ${var.artificial_requests_count})
default(avg:aws.elb.httpcode_backend_5xx${module.filter-tags.query_alert} by {region,loadbalancername}.as_rate(), 0) / (
default(avg:aws.elb.request_count${module.filter-tags.query_alert} by {region,loadbalancername}.as_rate() + ${var.artificial_requests_count}, 1))
* 100 > ${var.elb_backend_5xx_threshold_critical}
EOF

2
cloud/azure/app-services/monitors-app_services.tf
View File

@ -137,7 +137,7 @@ resource "datadog_monitor" "appservices_http_success_status_rate" {
query = <<EOF
${var.http_successful_requests_time_aggregator}(${var.http_successful_requests_timeframe}): ( (
default(avg:azure.app_services.http2xx${module.filter-tags.query_alert} by {resource_group,region,name}.as_rate(), 1) +
default(avg:azure.app_services.http2xx${module.filter-tags.query_alert} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.app_services.http3xx${module.filter-tags.query_alert} by {resource_group,region,name}.as_rate(), 0) ) /
default(avg:azure.app_services.requests${module.filter-tags.query_alert} by {resource_group,region,name}.as_rate(), 1)
) * 100 < ${var.http_successful_requests_threshold_critical}

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

@ -61,8 +61,8 @@ resource "datadog_monitor" "cosmos_db_4xx_requests" {
default(sum:azure.documentdb_databaseaccounts.total_requests${format(module.filter-tags-statuscode.query_alert, "413")} by {resource_group,region,name,collectionname}.as_rate(), 0) +
default(sum:azure.documentdb_databaseaccounts.total_requests${format(module.filter-tags-statuscode.query_alert, "429")} by {resource_group,region,name,collectionname}.as_rate(), 0) +
default(sum:azure.documentdb_databaseaccounts.total_requests${format(module.filter-tags-statuscode.query_alert, "449")} by {resource_group,region,name,collectionname}.as_rate(), 0) ) / (
default(sum:azure.cosmosdb.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 1) +
default(sum:azure.documentdb_databaseaccounts.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 1) )
default(sum:azure.cosmosdb.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 0) +
default(sum:azure.documentdb_databaseaccounts.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 0) )
) * 100 > ${var.cosmos_db_4xx_request_rate_threshold_critical}
EOF
@ -100,8 +100,8 @@ resource "datadog_monitor" "cosmos_db_5xx_requests" {
default(sum:azure.cosmosdb.total_requests${format(module.filter-tags-statuscode.query_alert, "503")} by {resource_group,region,name,collectionname}.as_rate(), 0) +
default(sum:azure.documentdb_databaseaccounts.total_requests${format(module.filter-tags-statuscode.query_alert, "500")} by {resource_group,region,name,collectionname}.as_rate(), 0) +
default(sum:azure.documentdb_databaseaccounts.total_requests${format(module.filter-tags-statuscode.query_alert, "503")} by {resource_group,region,name,collectionname}.as_rate(), 0) ) / (
default(sum:azure.cosmosdb.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 1) +
default(sum:azure.documentdb_databaseaccounts.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 1) )
default(sum:azure.cosmosdb.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 0) +
default(sum:azure.documentdb_databaseaccounts.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 0) )
) * 100 > ${var.cosmos_db_5xx_request_rate_threshold_critical}
EOF
@ -138,8 +138,8 @@ resource "datadog_monitor" "cosmos_db_scaling" {
${var.cosmos_db_scaling_time_aggregator}(${var.cosmos_db_scaling_timeframe}): ( (
default(sum:azure.cosmosdb.total_requests${format(module.filter-tags-statuscode.query_alert, "429")} by {resource_group,region,name,collectionname}.as_rate(), 0) +
default(sum:azure.documentdb_databaseaccounts.total_requests${format(module.filter-tags-statuscode.query_alert, "429")} by {resource_group,region,name,collectionname}.as_rate(), 0) ) / (
default(sum:azure.cosmosdb.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 1) +
default(sum:azure.documentdb_databaseaccounts.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 1) )
default(sum:azure.cosmosdb.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 0) +
default(sum:azure.documentdb_databaseaccounts.total_requests${module.filter-tags.query_alert} by {resource_group,region,name,collectionname}.as_rate(), 0) )
) * 100 > ${var.cosmos_db_scaling_error_rate_threshold_critical}
EOF

22
cloud/azure/iothubs/monitors-iothubs.tf
View File

@ -7,7 +7,7 @@ resource "datadog_monitor" "too_many_jobs_failed" {
${var.failed_jobs_rate_time_aggregator}(${var.failed_jobs_rate_timeframe}):(
default(avg:azure.devices_iothubs.jobs.failed{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) / (
default(avg:azure.devices_iothubs.jobs.failed{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.jobs.completed{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.jobs.completed{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.failed_jobs_rate_threshold_critical}
EOF
@ -41,7 +41,7 @@ resource "datadog_monitor" "too_many_list_jobs_failed" {
query = <<EOF
${var.failed_listjobs_rate_time_aggregator}(${var.failed_listjobs_rate_timeframe}):(
default(avg:azure.devices_iothubs.jobs.list_jobs.failure{${var.filter_tags}} by {resource_group,name}.as_rate(), 0) / (
default(avg:azure.devices_iothubs.jobs.list_jobs.success{${var.filter_tags}} by {resource_group,name}.as_rate(), 1) +
default(avg:azure.devices_iothubs.jobs.list_jobs.success{${var.filter_tags}} by {resource_group,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.jobs.list_jobs.failure{${var.filter_tags}} by {resource_group,name}.as_rate(), 0) )
) * 100 > ${var.failed_listjobs_rate_threshold_critical}
EOF
@ -76,7 +76,7 @@ resource "datadog_monitor" "too_many_query_jobs_failed" {
query = <<EOF
${var.failed_queryjobs_rate_time_aggregator}(${var.failed_queryjobs_rate_timeframe}):(
default(avg:azure.devices_iothubs.jobs.query_jobs.failure{${var.filter_tags}} by {resource_group,name}.as_rate(), 0) / (
default(avg:azure.devices_iothubs.jobs.query_jobs.success{${var.filter_tags}} by {resource_group,name}.as_rate(), 1) +
default(avg:azure.devices_iothubs.jobs.query_jobs.success{${var.filter_tags}} by {resource_group,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.jobs.query_jobs.failure{${var.filter_tags}} by {resource_group,name}.as_rate(), 0) )
) * 100 > ${var.failed_queryjobs_rate_threshold_critical}
EOF
@ -168,7 +168,7 @@ resource "datadog_monitor" "too_many_c2d_methods_failed" {
${var.failed_c2d_methods_rate_time_aggregator}(${var.failed_c2d_methods_rate_timeframe}):(
default(avg:azure.devices_iothubs.c2d.methods.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) / (
default(avg:azure.devices_iothubs.c2d.methods.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.c2d.methods.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.c2d.methods.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.failed_c2d_methods_rate_threshold_critical}
EOF
@ -203,7 +203,7 @@ resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
${var.failed_c2d_twin_read_rate_time_aggregator}(${var.failed_c2d_twin_read_rate_timeframe}):(
default(avg:azure.devices_iothubs.c2d.twin.read.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) / (
default(avg:azure.devices_iothubs.c2d.twin.read.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.c2d.twin.read.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.c2d.twin.read.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.failed_c2d_twin_read_rate_threshold_critical}
EOF
@ -238,7 +238,7 @@ resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
${var.failed_c2d_twin_update_rate_time_aggregator}(${var.failed_c2d_twin_update_rate_timeframe}):(
default(avg:azure.devices_iothubs.c2d.twin.update.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) / (
default(avg:azure.devices_iothubs.c2d.twin.update.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.c2d.twin.update.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.c2d.twin.update.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.failed_c2d_twin_update_rate_threshold_critical}
EOF
@ -273,7 +273,7 @@ resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
${var.failed_d2c_twin_read_rate_time_aggregator}(${var.failed_d2c_twin_read_rate_timeframe}):(
default(avg:azure.devices_iothubs.d2c.twin.read.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) / (
default(avg:azure.devices_iothubs.d2c.twin.read.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.twin.read.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.d2c.twin.read.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.failed_d2c_twin_read_rate_threshold_critical}
EOF
@ -308,7 +308,7 @@ resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
${var.failed_d2c_twin_update_rate_time_aggregator}(${var.failed_d2c_twin_update_rate_timeframe}):(
default(avg:azure.devices_iothubs.d2c.twin.update.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) / (
default(avg:azure.devices_iothubs.d2c.twin.update.failure{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.twin.update.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.d2c.twin.update.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.failed_d2c_twin_update_rate_threshold_critical}
EOF
@ -345,7 +345,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
default(avg:azure.devices_iothubs.d2c.telemetry.egress.dropped{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.orphaned{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.invalid{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.d2c.telemetry.egress.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.dropped_d2c_telemetry_egress_rate_threshold_critical}
EOF
@ -382,7 +382,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
default(avg:azure.devices_iothubs.d2c.telemetry.egress.dropped{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.orphaned{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.invalid{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.d2c.telemetry.egress.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.orphaned_d2c_telemetry_egress_rate_threshold_critical}
EOF
@ -419,7 +419,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
default(avg:azure.devices_iothubs.d2c.telemetry.egress.dropped{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.orphaned{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.invalid{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) +
default(avg:azure.devices_iothubs.d2c.telemetry.egress.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 1) )
default(avg:azure.devices_iothubs.d2c.telemetry.egress.success{${var.filter_tags}} by {resource_group,region,name}.as_rate(), 0) )
) * 100 > ${var.invalid_d2c_telemetry_egress_rate_threshold_critical}
EOF

8
cloud/gcp/lb/monitors-lb.tf
View File

@ -11,8 +11,8 @@ resource "datadog_monitor" "error_rate_4xx" {
query = <<EOF
${var.error_rate_4xx_time_aggregator}(${var.error_rate_4xx_timeframe}):
default(sum:gcp.loadbalancing.https.request_count{${var.filter_tags},response_code_class:400} by {forwarding_rule_name}.as_rate(), 0) / (
default(sum:gcp.loadbalancing.https.request_count{${var.filter_tags}} by {forwarding_rule_name}.as_rate(), 1) + ${var.error_rate_4xx_artificial_request}) * 100
> ${var.error_rate_4xx_threshold_critical}
default(sum:gcp.loadbalancing.https.request_count{${var.filter_tags}} by {forwarding_rule_name}.as_rate() + ${var.error_rate_4xx_artificial_request}, 1))
* 100 > ${var.error_rate_4xx_threshold_critical}
EOF
thresholds {
@ -49,8 +49,8 @@ resource "datadog_monitor" "error_rate_5xx" {
query = <<EOF
${var.error_rate_5xx_time_aggregator}(${var.error_rate_5xx_timeframe}):
default(sum:gcp.loadbalancing.https.request_count{${var.filter_tags},response_code_class:500} by {forwarding_rule_name}.as_rate(), 0) / (
default(sum:gcp.loadbalancing.https.request_count{${var.filter_tags}} by {forwarding_rule_name}.as_rate(), 1) + ${var.error_rate_5xx_artificial_request}) * 100
> ${var.error_rate_5xx_threshold_critical}
default(sum:gcp.loadbalancing.https.request_count{${var.filter_tags}} by {forwarding_rule_name}.as_rate() + ${var.error_rate_5xx_artificial_request}, 1))
* 100 > ${var.error_rate_5xx_threshold_critical}
EOF
thresholds {