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

Merged in MON-135-use-default-on-event-hub-query-a (pull request #60)

Browse Source 
MON-135 - Updated requests queries for Event Hub monitors

Approved-by: Jérôme Respaut <shr3ps@gmail.com>
Approved-by: Laurent Piroelle <laurent.piroelle@fr.clara.net>
Approved-by: Alexandre Gaillet <alexandre.gaillet@fr.clara.net>
Approved-by: Quentin Manfroi <quentin.manfroi@yahoo.fr>
This commit is contained in:
Alexandre Gaillet 2018-03-19 16:21:00 +00:00 committed by Quentin Manfroi
commit aae29be75a
2 changed files with 19 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -13,7 +13,7 @@ resource "datadog_monitor" "appservices_response_time" {
message = "${coalesce(var.response_time_message, var.message)}"
query = <<EOF
avg(last_5m): (
min(last_5m): (
avg:azure.app_services.average_response_time{${data.template_file.filter.rendered}} by {resource_group,region,name}
) > ${var.response_time_threshold_critical}
EOF

27
cloud/azure/eventhub/monitors-eventhub.tf
View File

@ -38,11 +38,11 @@ resource "datadog_monitor" "eventhub_failed_requests" {
query = <<EOF
sum(last_5m): (
avg:azure.eventhub_namespaces.failed_requests{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() ) /
(
avg:azure.eventhub_namespaces.successful_requests{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.eventhub_namespaces.failed_requests{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
) * 100 > ${var.failed_requests_rate_thresold_critical}
default(
avg:azure.eventhub_namespaces.failed_requests{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() /
avg:azure.eventhub_namespaces.incoming_requests{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count(),
0) * 100
) > ${var.failed_requests_rate_thresold_critical}
EOF
type = "metric alert"
@ -58,7 +58,7 @@ resource "datadog_monitor" "eventhub_failed_requests" {
evaluation_delay = "${var.delay}"
renotify_interval = 0
notify_audit = false
timeout_h = 1
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
@ -69,20 +69,21 @@ resource "datadog_monitor" "eventhub_failed_requests" {
}
resource "datadog_monitor" "eventhub_errors" {
name = "[${var.environment}] Event Hub too manny errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
name = "[${var.environment}] Event Hub too many errors {{comparator}} {{#is_alert}}{{threshold}}%{{/is_alert}}{{#is_warning}}{{warn_threshold}}%{{/is_warning}} ({{value}}%)"
message = "${coalesce(var.errors_rate_message, var.message)}"
query = <<EOF
sum(last_5m): (
default(
(
avg:azure.eventhub_namespaces.internal_server_errors{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.eventhub_namespaces.server_busy_errors{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.eventhub_namespaces.other_errors{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
) / (
avg:azure.eventhub_namespaces.successful_requests{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.eventhub_namespaces.internal_server_errors{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.eventhub_namespaces.server_busy_errors{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.eventhub_namespaces.other_errors{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
) * 100 > ${var.errors_rate_thresold_critical}
avg:eventhub_namespaces.incoming_requests{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
),
0) * 100
) > ${var.errors_rate_thresold_critical}
EOF
type = "metric alert"
@ -98,7 +99,7 @@ resource "datadog_monitor" "eventhub_errors" {
evaluation_delay = "${var.delay}"
renotify_interval = 0
notify_audit = false
timeout_h = 1
timeout_h = 0
include_tags = true
locked = false
require_full_window = false