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

Merged in MON-80-azure-hub-iot-monitors (pull request #29)

Browse Source 
MON-80 azure hub iot monitors

Approved-by: Quentin Manfroi <quentin.manfroi@yahoo.fr>
Approved-by: Laurent Piroelle <laurent.piroelle@fr.clara.net>
This commit is contained in:
Quentin Manfroi 2017-12-15 11:55:44 +00:00
commit b687cacd4c
1 changed files with 34 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

68
cloud/azure/iothubs/monitors-iothubs.tf
View File

@ -11,10 +11,10 @@ resource "datadog_monitor" "too_many_jobs_failed" {
message = "${var.message}"
query = <<EOF
avg(last_5m):(
avg:azure.devices_iothubs.jobs.failed{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
( avg:azure.devices_iothubs.jobs.failed{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
avg:azure.devices_iothubs.jobs.completed{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
sum(last_5m):(
avg:azure.devices_iothubs.jobs.failed{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() /
( avg:azure.devices_iothubs.jobs.failed{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.devices_iothubs.jobs.completed{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() )
) * 100 > ${var.failed_jobs_rate_threshold_critical}
EOF
@ -44,7 +44,7 @@ resource "datadog_monitor" "too_many_list_jobs_failed" {
message = "${var.message}"
query = <<EOF
avg(last_5m):(
sum(last_5m):(
avg:azure.devices_iothubs.jobs.list_jobs.failure{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() /
( avg:azure.devices_iothubs.jobs.list_jobs.success{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() +
avg:azure.devices_iothubs.jobs.list_jobs.failure{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() )
@ -77,7 +77,7 @@ resource "datadog_monitor" "too_many_query_jobs_failed" {
message = "${var.message}"
query = <<EOF
avg(last_5m):(
sum(last_5m):(
avg:azure.devices_iothubs.jobs.query_jobs.failure{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() /
( avg:azure.devices_iothubs.jobs.query_jobs.success{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() +
avg:azure.devices_iothubs.jobs.query_jobs.failure{${data.template_file.filter.rendered}} by {resource_group,name}.as_count() )
@ -110,7 +110,7 @@ resource "datadog_monitor" "status" {
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.devices_iothubs.status{${data.template_file.filter.rendered}} by {name,resource_group} < 1
avg(last_5m):avg:azure.devices_iothubs.status{${data.template_file.filter.rendered}} by {resource_group,region,name} < 1
EOF
type = "metric alert"
@ -134,7 +134,7 @@ resource "datadog_monitor" "total_devices" {
message = "${var.message}"
query = <<EOF
avg(last_5m):avg:azure.devices_iothubs.devices.total_devices{${data.template_file.filter.rendered}} by {name,resource_group} == 0
avg(last_5m):avg:azure.devices_iothubs.devices.total_devices{${data.template_file.filter.rendered}} by {resource_group,region,name} == 0
EOF
type = "metric alert"
@ -158,10 +158,10 @@ resource "datadog_monitor" "too_many_c2d_methods_failed" {
message = "${var.message}"
query = <<EOF
avg(last_5m):(
avg:azure.devices_iothubs.c2d.methods.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
( avg:azure.devices_iothubs.c2d.methods.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
avg:azure.devices_iothubs.c2d.methods.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
sum(last_5m):(
avg:azure.devices_iothubs.c2d.methods.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() /
( avg:azure.devices_iothubs.c2d.methods.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.devices_iothubs.c2d.methods.success{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() )
) * 100 > ${var.failed_c2d_methods_rate_threshold_critical}
EOF
@ -191,10 +191,10 @@ resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
message = "${var.message}"
query = <<EOF
avg(last_5m):(
avg:azure.devices_iothubs.c2d.twin.read.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
( avg:azure.devices_iothubs.c2d.twin.read.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
avg:azure.devices_iothubs.c2d.twin.read.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
sum(last_5m):(
avg:azure.devices_iothubs.c2d.twin.read.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() /
( avg:azure.devices_iothubs.c2d.twin.read.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.devices_iothubs.c2d.twin.read.success{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() )
) * 100 > ${var.failed_c2d_twin_read_rate_threshold_critical}
EOF
@ -224,10 +224,10 @@ resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
message = "${var.message}"
query = <<EOF
avg(last_5m):(
avg:azure.devices_iothubs.c2d.twin.update.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
( avg:azure.devices_iothubs.c2d.twin.update.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
avg:azure.devices_iothubs.c2d.twin.update.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
sum(last_5m):(
avg:azure.devices_iothubs.c2d.twin.update.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() /
( avg:azure.devices_iothubs.c2d.twin.update.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.devices_iothubs.c2d.twin.update.success{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() )
) * 100 > ${var.failed_c2d_twin_update_rate_threshold_critical}
EOF
@ -257,10 +257,10 @@ resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
message = "${var.message}"
query = <<EOF
avg(last_5m):(
avg:azure.devices_iothubs.d2c.twin.read.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
( avg:azure.devices_iothubs.d2c.twin.read.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
avg:azure.devices_iothubs.d2c.twin.read.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
sum(last_5m):(
avg:azure.devices_iothubs.d2c.twin.read.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() /
( avg:azure.devices_iothubs.d2c.twin.read.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.devices_iothubs.d2c.twin.read.success{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() )
) * 100 > ${var.failed_d2c_twin_read_rate_threshold_critical}
EOF
@ -290,10 +290,10 @@ resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
message = "${var.message}"
query = <<EOF
avg(last_5m):(
avg:azure.devices_iothubs.d2c.twin.update.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() /
( avg:azure.devices_iothubs.d2c.twin.update.failure{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() +
avg:azure.devices_iothubs.d2c.twin.update.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() )
sum(last_5m):(
avg:azure.devices_iothubs.d2c.twin.update.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() /
( avg:azure.devices_iothubs.d2c.twin.update.failure{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() +
avg:azure.devices_iothubs.d2c.twin.update.success{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() )
) * 100 > ${var.failed_d2c_twin_update_rate_threshold_critical}
EOF
@ -324,7 +324,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
query = <<EOF
sum(last_5m): (
avg:azure.devices_iothubs.d2c.telemetry.egress.dropped{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
avg:azure.devices_iothubs.d2c.telemetry.egress.dropped{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
) > ${var.dropped_d2c_telemetry_egress_threshold_critical}
EOF
@ -355,7 +355,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
query = <<EOF
sum(last_5m): (
avg:azure.devices_iothubs.d2c.telemetry.egress.orphaned{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
avg:azure.devices_iothubs.d2c.telemetry.egress.orphaned{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
) > ${var.orphaned_d2c_telemetry_egress_threshold_critical}
EOF
@ -386,7 +386,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
query = <<EOF
sum(last_5m): (
avg:azure.devices_iothubs.d2c.telemetry.egress.invalid{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
avg:azure.devices_iothubs.d2c.telemetry.egress.invalid{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
) > ${var.invalid_d2c_telemetry_egress_threshold_critical}
EOF
@ -417,7 +417,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" {
query = <<EOF
sum(last_5m): (
avg:azure.devices_iothubs.d2c.telemetry.egress.fallback{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
avg:azure.devices_iothubs.d2c.telemetry.egress.fallback{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
) > ${var.fallback_d2c_telemetry_egress_threshold_critical}
EOF
@ -448,8 +448,8 @@ resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" {
query = <<EOF
sum(last_5m): (
avg:azure.devices_iothubs.d2c.telemetry.ingress.all_protocol{${data.template_file.filter.rendered}} by {name,resource_group}.as_count() -
avg:azure.devices_iothubs.d2c.telemetry.ingress.success{${data.template_file.filter.rendered}} by {name,resource_group}.as_count()
avg:azure.devices_iothubs.d2c.telemetry.ingress.all_protocol{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count() -
avg:azure.devices_iothubs.d2c.telemetry.ingress.success{${data.template_file.filter.rendered}} by {resource_group,region,name}.as_count()
) > 0
EOF