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

MON-459 fix undesired changes

Browse Source
This commit is contained in:
Quentin Manfroi 2019-06-14 12:39:34 +02:00
parent b093f2f1e4
commit 2215481f71
43 changed files with 327 additions and 388 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
caas/kubernetes/ingress/vts/monitors-ingress.tf
View File

@ -2,6 +2,7 @@ resource "datadog_monitor" "nginx_ingress_too_many_5xx" {
count = var.ingress_5xx_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Nginx Ingress 5xx errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.ingress_5xx_message, var.message)
type = "query alert"
query = <<EOQ
${var.ingress_5xx_time_aggregator}(${var.ingress_5xx_timeframe}): default(
@ -10,16 +11,14 @@ resource "datadog_monitor" "nginx_ingress_too_many_5xx" {
* 100, 0) > ${var.ingress_5xx_threshold_critical}
EOQ
type = "query alert"
thresholds = {
warning = var.ingress_5xx_threshold_warning
critical = var.ingress_5xx_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
@ -34,6 +33,7 @@ resource "datadog_monitor" "nginx_ingress_too_many_4xx" {
count = var.ingress_4xx_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Nginx Ingress 4xx errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.ingress_4xx_message, var.message)
type = "query alert"
query = <<EOQ
${var.ingress_4xx_time_aggregator}(${var.ingress_4xx_timeframe}): default(
@ -42,16 +42,14 @@ resource "datadog_monitor" "nginx_ingress_too_many_4xx" {
* 100, 0) > ${var.ingress_4xx_threshold_critical}
EOQ
type = "query alert"
thresholds = {
warning = var.ingress_4xx_threshold_warning
critical = var.ingress_4xx_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0

18
caas/kubernetes/node/monitors-k8s-node.tf
View File

@ -2,7 +2,6 @@ resource "datadog_monitor" "disk_pressure" {
count = var.disk_pressure_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node Disk pressure"
message = coalesce(var.disk_pressure_message, var.message)
type = "service check"
query = <<EOQ
@ -30,7 +29,6 @@ resource "datadog_monitor" "disk_out" {
count = var.disk_out_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node Out of disk"
message = coalesce(var.disk_out_message, var.message)
type = "service check"
query = <<EOQ
@ -58,7 +56,6 @@ resource "datadog_monitor" "memory_pressure" {
count = var.memory_pressure_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node Memory pressure"
message = coalesce(var.memory_pressure_message, var.message)
type = "service check"
query = <<EOQ
@ -86,7 +83,6 @@ resource "datadog_monitor" "ready" {
count = var.ready_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node not ready"
message = coalesce(var.ready_message, var.message)
type = "service check"
query = <<EOQ
@ -114,7 +110,6 @@ resource "datadog_monitor" "kubelet_ping" {
count = var.kubelet_ping_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node Kubelet API does not respond"
message = coalesce(var.kubelet_ping_message, var.message)
type = "service check"
query = <<EOQ
@ -142,7 +137,6 @@ resource "datadog_monitor" "kubelet_syncloop" {
count = var.kubelet_syncloop_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node Kubelet sync loop that updates containers does not work"
message = coalesce(var.kubelet_syncloop_message, var.message)
type = "service check"
query = <<EOQ
@ -169,15 +163,14 @@ tags = ["env:${var.environment}", "type:caas", "provider:kubernetes", "resource:
resource "datadog_monitor" "unregister_net_device" {
count = var.unregister_net_device_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node Frequent unregister net device"
type = "event alert"
message = coalesce(var.unregister_net_device_message, var.message)
type = "event alert"
query = <<EOQ
events('sources:kubernetes priority:all ${module.filter-tags.event_alert} \"UnregisterNetDevice\"').rollup('count').last('${var.unregister_net_device_timeframe}') > ${var.unregister_net_device_threshold_critical}
EOQ
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
@ -191,8 +184,8 @@ tags = ["env:${var.environment}", "type:caas", "provider:kubernetes", "resource:
resource "datadog_monitor" "node_unschedulable" {
count = var.node_unschedulable_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node unschedulable"
type = "metric alert"
message = coalesce(var.node_unschedulable_message, var.message)
type = "metric alert"
query = <<EOQ
${var.node_unschedulable_time_aggregator}(${var.node_unschedulable_timeframe}):
@ -206,7 +199,6 @@ EOQ
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
@ -221,8 +213,8 @@ EOQ
resource "datadog_monitor" "volume_space" {
count = var.volume_space_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node volume space usage {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.volume_space_message, var.message)
type = "query alert"
query = <<EOQ
${var.volume_space_time_aggregator}(${var.volume_space_timeframe}):
@ -238,7 +230,6 @@ EOQ
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
@ -253,8 +244,8 @@ EOQ
resource "datadog_monitor" "volume_inodes" {
count = var.volume_inodes_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Node volume inodes usage {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.volume_inodes_message, var.message)
type = "query alert"
query = <<EOQ
${var.volume_inodes_time_aggregator}(${var.volume_inodes_timeframe}):
@ -270,7 +261,6 @@ warning = var.volume_inodes_threshold_warning
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false

5
caas/kubernetes/pod/monitors-k8s-pod.tf
View File

@ -1,8 +1,8 @@
resource "datadog_monitor" "pod_phase_status" {
count = var.pod_phase_status_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Pod phase status failed"
type = "metric alert"
message = coalesce(var.pod_phase_status_message, var.message)
type = "metric alert"
query = <<EOQ
${var.pod_phase_status_time_aggregator}(${var.pod_phase_status_timeframe}):
@ -30,8 +30,8 @@ EOQ
resource "datadog_monitor" "error" {
count = var.error_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Pod waiting errors"
type = "query alert"
message = coalesce(var.error_message, var.message)
type = "query alert"
query = <<EOQ
${var.error_time_aggregator}(${var.error_timeframe}):
@ -46,7 +46,6 @@ warning = var.error_threshold_warning
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false

11
caas/kubernetes/workload/monitors-k8s-workload.tf
View File

@ -2,7 +2,6 @@ resource "datadog_monitor" "job" {
count = var.job_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes job failed"
message = coalesce(var.job_message, var.message)
type = "service check"
query = <<EOQ
@ -30,7 +29,6 @@ resource "datadog_monitor" "cronjob" {
count = var.cronjob_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes cronjob scheduling failed"
message = coalesce(var.cronjob_message, var.message)
type = "service check"
query = <<EOQ
@ -57,8 +55,8 @@ tags = ["env:${var.environment}", "type:caas", "provider:kubernetes", "resource:
resource "datadog_monitor" "replica_available" {
count = var.replica_available_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Available replicas {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.replica_available_message, var.message)
type = "query alert"
query = <<EOQ
${var.replica_available_time_aggregator}(${var.replica_available_timeframe}):
@ -73,7 +71,6 @@ critical = var.replica_available_threshold_critical
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
@ -88,8 +85,8 @@ tags = ["env:${var.environment}", "type:caas", "provider:kubernetes", "resource:
resource "datadog_monitor" "replica_ready" {
count = var.replica_ready_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Ready replicas {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.replica_ready_message, var.message)
type = "query alert"
query = <<EOQ
${var.replica_available_time_aggregator}(${var.replica_available_timeframe}):
@ -104,7 +101,6 @@ EOQ
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
@ -119,8 +115,8 @@ EOQ
resource "datadog_monitor" "replica_current" {
count = var.replica_current_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Kubernetes Current replicas {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.replica_current_message, var.message)
type = "query alert"
query = <<EOQ
${var.replica_available_time_aggregator}(${var.replica_available_timeframe}):
@ -135,7 +131,6 @@ EOQ
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false

4
cloud/aws/alb/monitors-alb.tf
View File

@ -60,7 +60,7 @@ resource "datadog_monitor" "ALB_httpcode_5xx" {
count = var.httpcode_alb_5xx_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] ALB HTTP code 5xx {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.httpcode_alb_5xx_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.httpcode_alb_5xx_time_aggregator}(${var.httpcode_alb_5xx_timeframe}):
@ -89,7 +89,7 @@ resource "datadog_monitor" "ALB_httpcode_4xx" {
count = var.httpcode_alb_4xx_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] ALB HTTP code 4xx {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.httpcode_alb_4xx_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.httpcode_alb_4xx_time_aggregator}(${var.httpcode_alb_4xx_timeframe}):

21
cloud/aws/apigateway/monitors-api.tf
View File

@ -2,8 +2,8 @@
resource "datadog_monitor" "API_Gateway_latency" {
count = var.latency_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] API Gateway latency {{#is_alert}}{{{comparator}}} {{threshold}}ms ({{value}}ms){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}ms ({{value}}ms){{/is_warning}}"
type = "query alert"
message = coalesce(var.latency_message, var.message)
type = "query alert"
query = <<EOQ
${var.latency_time_aggregator}(${var.latency_timeframe}):
@ -11,14 +11,13 @@ resource "datadog_monitor" "API_Gateway_latency" {
> ${var.latency_threshold_critical}
EOQ
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
thresholds = {
warning = var.latency_threshold_warning
critical = var.latency_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
require_full_window = false
@ -32,8 +31,8 @@ EOQ
resource "datadog_monitor" "API_http_5xx_errors_count" {
count = var.http_5xx_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] API Gateway HTTP 5xx errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.http_5xx_requests_message, var.message)
type = "query alert"
query = <<EOQ
${var.http_5xx_requests_time_aggregator}(${var.http_5xx_requests_timeframe}):
@ -42,14 +41,13 @@ resource "datadog_monitor" "API_http_5xx_errors_count" {
* 100 > ${var.http_5xx_requests_threshold_critical}
EOQ
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
thresholds = {
warning = var.http_5xx_requests_threshold_warning
critical = var.http_5xx_requests_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
require_full_window = false
@ -63,8 +61,8 @@ tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:apigat
resource "datadog_monitor" "API_http_4xx_errors_count" {
count = var.http_4xx_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] API Gateway HTTP 4xx errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.http_4xx_requests_message, var.message)
type = "query alert"
query = <<EOQ
${var.http_4xx_requests_time_aggregator}(${var.http_4xx_requests_timeframe}):
@ -73,14 +71,13 @@ query = <<EOQ
* 100 > ${var.http_4xx_requests_threshold_critical}
EOQ
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
thresholds = {
warning = var.http_4xx_requests_threshold_warning
critical = var.http_4xx_requests_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
require_full_window = false

10
cloud/aws/elasticache/memcached/monitors-memcached.tf
View File

@ -2,7 +2,6 @@ resource "datadog_monitor" "memcached_get_hits" {
count = var.get_hits_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Elasticache memcached cache hit ratio {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.get_hits_message, var.message)
type = "query alert"
query = <<EOQ
@ -18,15 +17,15 @@ EOQ
critical = var.get_hits_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticache-memcached", "team:claranet", "created-by:terraform", "engine:memcached", var.get_hits_extra_tags]
}
@ -35,7 +34,6 @@ resource "datadog_monitor" "memcached_cpu_high" {
count = var.cpu_high_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Elasticache memcached CPU {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.cpu_high_message, var.message)
type = "query alert"
query = <<EOQ
@ -49,15 +47,15 @@ warning = var.cpu_high_threshold_warning
critical = var.cpu_high_threshold_critical
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticache-memcached", "team:claranet", "created-by:terraform", "engine:memcached", var.cpu_high_extra_tags]
}

23
cloud/aws/elasticache/redis/monitors-redis.tf
View File

@ -2,7 +2,6 @@ resource "datadog_monitor" "redis_cache_hits" {
count = var.cache_hits_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Elasticache redis cache hit ratio {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.cache_hits_message, var.message)
type = "query alert"
query = <<EOQ
@ -18,15 +17,15 @@ EOQ
critical = var.cache_hits_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticache-redis", "team:claranet", "created-by:terraform", "engine:redis", var.cache_hits_extra_tags]
}
@ -35,7 +34,6 @@ resource "datadog_monitor" "redis_cpu_high" {
count = var.cpu_high_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Elasticache redis CPU {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.cpu_high_message, var.message)
type = "query alert"
query = <<EOQ
@ -44,15 +42,15 @@ resource "datadog_monitor" "redis_cpu_high" {
) > ${var.cpu_high_threshold_critical}
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticache-redis", "team:claranet", "created-by:terraform", "engine:redis", var.cpu_high_extra_tags]
}
@ -61,8 +59,7 @@ resource "datadog_monitor" "redis_replication_lag" {
count = var.replication_lag_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Elasticache redis replication lag {{#is_alert}}{{{comparator}}} {{threshold}}s ({{value}}s){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}s ({{value}}s){{/is_warning}}"
message = coalesce(var.replication_lag_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.replication_lag_time_aggregator}(${var.replication_lag_timeframe}): (
@ -75,15 +72,15 @@ warning = var.replication_lag_threshold_warning
critical = var.replication_lag_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticache-redis", "team:claranet", "created-by:terraform", "engine:redis", var.replication_lag_extra_tags]
}
@ -92,8 +89,8 @@ resource "datadog_monitor" "redis_commands" {
count = var.commands_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Elasticache redis is receiving no commands"
message = coalesce(var.commands_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
sum(${var.commands_timeframe}): (
@ -102,15 +99,15 @@ query = <<EOQ
) <= 0
EOQ
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticache-redis", "team:claranet", "created-by:terraform", "engine:redis", var.commands_extra_tags]
}

15
cloud/aws/elasticsearch/monitors-elasticsearch.tf
View File

@ -21,15 +21,15 @@ EOQ
critical = 2
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticsearch", "team:claranet", "created-by:terraform", var.es_cluster_status_extra_tags]
}
@ -53,15 +53,15 @@ warning = var.diskspace_threshold_warning
critical = var.diskspace_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticsearch", "team:claranet", "created-by:terraform", var.diskspace_extra_tags]
}
@ -71,8 +71,7 @@ resource "datadog_monitor" "es_cpu_90_15min" {
count = var.cpu_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] ElasticSearch cluster CPU high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.cpu_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.cpu_time_aggregator}(${var.cpu_timeframe}): (
@ -85,15 +84,15 @@ warning = var.cpu_threshold_warning
critical = var.cpu_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elasticsearch", "team:claranet", "created-by:terraform", var.cpu_extra_tags]
}

28
cloud/aws/elb/monitors-elb.tf
View File

@ -17,15 +17,15 @@ EOQ
warning = 100
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elb", "team:claranet", "created-by:terraform", var.elb_no_healthy_instance_extra_tags]
}
@ -48,15 +48,15 @@ warning = var.elb_4xx_threshold_warning
critical = var.elb_4xx_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elb", "team:claranet", "created-by:terraform", var.elb_4xx_extra_tags]
}
@ -65,7 +65,7 @@ resource "datadog_monitor" "ELB_too_much_5xx" {
count = var.elb_5xx_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] ELB 5xx errors too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.elb_5xx_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
sum(${var.elb_5xx_timeframe}):
@ -79,15 +79,15 @@ warning = var.elb_5xx_threshold_warning
critical = var.elb_5xx_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elb", "team:claranet", "created-by:terraform", var.elb_5xx_extra_tags]
}
@ -96,7 +96,7 @@ resource "datadog_monitor" "ELB_too_much_4xx_backend" {
count = var.elb_backend_4xx_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] ELB backend 4xx errors too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.elb_backend_4xx_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
sum(${var.elb_backend_4xx_timeframe}):
@ -110,15 +110,15 @@ EOQ
critical = var.elb_backend_4xx_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elb", "team:claranet", "created-by:terraform", var.elb_backend_4xx_extra_tags]
}
@ -141,15 +141,15 @@ EOQ
critical = var.elb_backend_5xx_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elb", "team:claranet", "created-by:terraform", var.elb_backend_5xx_extra_tags]
}
@ -171,15 +171,15 @@ warning = var.elb_backend_latency_warning
critical = var.elb_backend_latency_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:elb", "team:claranet", "created-by:terraform", var.elb_backend_latency_extra_tags]
}

4
cloud/aws/kinesis-firehose/monitors-kinesis-firehose.tf
View File

@ -15,15 +15,15 @@ EOQ
critical = 0
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:kinesis-firehose", "team:claranet", "created-by:terraform", var.incoming_records_extra_tags]
}

4
cloud/aws/rds/aurora/mysql/monitors-rds-aurora-mysql.tf
View File

@ -16,14 +16,14 @@ EOQ
critical = var.aurora_replicalag_threshold_critical
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:rds-aurora-mysql", "team:claranet", "created-by:terraform", var.aurora_replicalag_extra_tags]
}

4
cloud/aws/rds/aurora/postgresql/monitors-rds-aurora-postgresql.tf
View File

@ -16,14 +16,14 @@ EOQ
critical = var.aurora_replicalag_threshold_critical
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:rds-aurora-postgresql", "team:claranet", "created-by:terraform", var.aurora_replicalag_extra_tags]
}

14
cloud/aws/rds/common/monitors-rds-common.tf
View File

@ -16,14 +16,14 @@ EOQ
critical = var.cpu_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:rds", "team:claranet", "created-by:terraform", var.cpu_extra_tags]
}
@ -47,14 +47,14 @@ warning = var.diskspace_threshold_warning
critical = var.diskspace_threshold_critical
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:rds", "team:claranet", "created-by:terraform", var.diskspace_extra_tags]
}
@ -64,7 +64,7 @@ resource "datadog_monitor" "rds_replica_lag" {
count = var.replicalag_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] RDS replica lag {{#is_alert}}{{{comparator}}} {{threshold}} ms ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ms ({{value}}%){{/is_warning}}"
message = coalesce(var.replicalag_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
avg(${var.replicalag_timeframe}): (
@ -77,14 +77,14 @@ warning = var.replicalag_threshold_warning
critical = var.replicalag_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:aws", "resource:rds", "team:claranet", "created-by:terraform", var.replicalag_extra_tags]
}

4
cloud/aws/vpn/monitors-vpn.tf
View File

@ -10,10 +10,10 @@ resource "datadog_monitor" "VPN_status" {
) < 1
EOQ
notify_no_data = true
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true

24
cloud/azure/apimanagement/monitors-azure-apimanagement.tf
View File

@ -12,15 +12,15 @@ EOQ
critical = 1
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:apimanagement", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -43,14 +43,14 @@ critical = var.failed_requests_threshold_critical
warning = var.failed_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:apimanagement", "team:claranet", "created-by:terraform", var.failed_requests_extra_tags]
@ -60,7 +60,7 @@ resource "datadog_monitor" "apimgt_other_requests" {
count = var.other_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] API Management too many other requests {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.other_requests_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.other_requests_time_aggregator}(${var.other_requests_timeframe}): (
@ -74,14 +74,14 @@ critical = var.other_requests_threshold_critical
warning = var.other_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:apimanagement", "team:claranet", "created-by:terraform", var.other_requests_extra_tags]
@ -91,7 +91,7 @@ resource "datadog_monitor" "apimgt_unauthorized_requests" {
count = var.unauthorized_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] API Management too many unauthorized requests {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.unauthorized_requests_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.unauthorized_requests_time_aggregator}(${var.unauthorized_requests_timeframe}): (
@ -105,14 +105,14 @@ EOQ
warning = var.unauthorized_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:apimanagement", "team:claranet", "created-by:terraform", var.unauthorized_requests_extra_tags]
@ -138,14 +138,14 @@ EOQ
warning = var.successful_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:apimanagement", "team:claranet", "created-by:terraform", var.successful_requests_extra_tags]

6
cloud/azure/app-services/monitors-app_services.tf
View File

@ -31,8 +31,8 @@ EOQ
resource "datadog_monitor" "appservices_memory_usage_count" {
count = var.memory_usage_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] App Services memory usage {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
type = "query alert"
message = coalesce(var.memory_usage_message, var.message)
type = "query alert"
query = <<EOQ
${var.memory_usage_time_aggregator}(${var.memory_usage_timeframe}): (
@ -61,7 +61,7 @@ resource "datadog_monitor" "appservices_http_5xx_errors_count" {
count = var.http_5xx_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] App Services HTTP 5xx errors too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.http_5xx_requests_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.http_5xx_requests_time_aggregator}(${var.http_5xx_requests_timeframe}): (
@ -90,8 +90,8 @@ tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:app-
resource "datadog_monitor" "appservices_http_4xx_errors_count" {
count = var.http_4xx_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] App Services HTTP 4xx errors too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.http_4xx_requests_message, var.message)
type = "query alert"
query = <<EOQ
${var.http_4xx_requests_time_aggregator}(${var.http_4xx_requests_timeframe}): (

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

@ -14,15 +14,15 @@ EOQ
critical = 1
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:cosmos_db", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -55,15 +55,15 @@ critical = var.cosmos_db_4xx_request_rate_threshold_critical
warning = var.cosmos_db_4xx_request_rate_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:cosmos_db", "team:claranet", "created-by:terraform", var.cosmos_db_4xx_request_extra_tags]
}
@ -72,7 +72,7 @@ resource "datadog_monitor" "cosmos_db_5xx_requests" {
count = var.cosmos_db_5xx_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Cosmos DB 5xx requests rate is high {{#is_alert}}{{comparator}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{comparator}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.cosmos_db_5xx_requests_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.cosmos_db_5xx_request_time_aggregator}(${var.cosmos_db_5xx_request_timeframe}): default( (
@ -87,15 +87,15 @@ critical = var.cosmos_db_5xx_request_rate_threshold_critical
warning = var.cosmos_db_5xx_request_rate_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:cosmos_db", "team:claranet", "created-by:terraform", var.cosmos_db_5xx_request_rate_extra_tags]
}
@ -104,7 +104,7 @@ resource "datadog_monitor" "cosmos_db_scaling" {
count = var.cosmos_db_scaling_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Cosmos DB max scaling reached for collection {{#is_alert}}{{comparator}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{comparator}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.cosmos_db_scaling_message, var.message)
type = "query alert"
type = "query alert"
# List of available status codes : https://docs.microsoft.com/en-us/rest/api/cosmos-db/http-status-codes-for-cosmosdb
query = <<EOQ
@ -119,15 +119,15 @@ EOQ
warning = var.cosmos_db_scaling_error_rate_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:cosmos_db", "team:claranet", "created-by:terraform", var.cosmos_db_scaling_extra_tags]
}

4
cloud/azure/datalakestore/monitors-datalakestore.tf
View File

@ -10,15 +10,15 @@ resource "datadog_monitor" "datalakestore_status" {
) < 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:datalakestore", "team:claranet", "created-by:terraform", var.status_extra_tags]
}

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

@ -10,15 +10,15 @@ resource "datadog_monitor" "eventgrid_no_successful_message" {
avg:azure.eventgrid_topics.publish_success_count${module.filter-tags.query_alert} by {resource_group,region,name} < 0
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:eventgrid", "team:claranet", "created-by:terraform", var.no_successful_message_rate_extra_tags]
}
@ -44,15 +44,15 @@ critical = var.failed_messages_rate_thresold_critical
warning = var.failed_messages_rate_thresold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:eventgrid", "team:claranet", "created-by:terraform", var.failed_messages_rate_extra_tags]
}
@ -61,7 +61,7 @@ resource "datadog_monitor" "eventgrid_unmatched_events" {
count = var.unmatched_events_rate_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Event Grid too many unmatched events {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.unmatched_events_rate_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.unmatched_events_rate_time_aggregator}(${var.unmatched_events_rate_timeframe}): (default(
@ -78,15 +78,15 @@ critical = var.unmatched_events_rate_thresold_critical
warning = var.unmatched_events_rate_thresold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:eventgrid", "team:claranet", "created-by:terraform", var.unmatched_events_rate_extra_tags]
}

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

@ -10,15 +10,15 @@ resource "datadog_monitor" "eventhub_status" {
) != 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:eventhub", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -41,15 +41,15 @@ critical = var.failed_requests_rate_thresold_critical
warning = var.failed_requests_rate_thresold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:eventhub", "team:claranet", "created-by:terraform", var.failed_requests_rate_extra_tags]
}
@ -58,7 +58,7 @@ resource "datadog_monitor" "eventhub_errors" {
count = var.errors_rate_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Event Hub too many errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.errors_rate_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.errors_rate_time_aggregator}(${var.errors_rate_timeframe}): ( (
@ -74,15 +74,15 @@ critical = var.errors_rate_thresold_critical
warning = var.errors_rate_thresold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:eventhub", "team:claranet", "created-by:terraform", var.errors_rate_extra_tags]
}

6
cloud/azure/functions/monitors-functions.tf
View File

@ -1,8 +1,8 @@
resource "datadog_monitor" "function_http_5xx_errors_rate" {
count = var.http_5xx_errors_rate_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Function App HTTP 5xx errors too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
type = "query alert"
message = coalesce(var.http_5xx_errors_rate_message, var.message)
type = "query alert"
query = <<EOQ
${var.http_5xx_errors_rate_time_aggregator}(${var.http_5xx_errors_rate_timeframe}): default(
@ -30,8 +30,8 @@ EOQ
resource "datadog_monitor" "function_high_connections_count" {
count = var.high_connections_count_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Function App connections count too high {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
type = "query alert"
message = coalesce(var.high_connections_count_message, var.message)
type = "query alert"
query = <<EOQ
${var.high_connections_count_time_aggregator}(${var.high_connections_count_timeframe}):
@ -58,8 +58,8 @@ tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:azur
resource "datadog_monitor" "function_high_threads_count" {
count = var.high_threads_count_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Function App threads count too high {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
type = "query alert"
message = coalesce(var.high_threads_count_message, var.message)
type = "query alert"
query = <<EOQ
${var.high_threads_count_time_aggregator}(${var.high_threads_count_timeframe}):

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

@ -18,15 +18,15 @@ EOQ
critical = var.failed_jobs_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.failed_jobs_rate_extra_tags]
}
@ -51,15 +51,15 @@ warning = var.failed_listjobs_rate_threshold_warning
critical = var.failed_listjobs_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.failed_listjobs_rate_extra_tags]
}
@ -68,7 +68,7 @@ resource "datadog_monitor" "too_many_query_jobs_failed" {
count = var.failed_queryjobs_rate_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] IOT Hub Too many query_jobs failed {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.failed_queryjobs_rate_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.failed_queryjobs_rate_time_aggregator}(${var.failed_queryjobs_rate_timeframe}):
@ -84,15 +84,15 @@ warning = var.failed_queryjobs_rate_threshold_warning
critical = var.failed_queryjobs_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.failed_queryjobs_rate_extra_tags]
}
@ -101,7 +101,7 @@ resource "datadog_monitor" "status" {
count = var.status_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] IOT Hub is down"
message = coalesce(var.status_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.status_time_aggregator}(${var.status_timeframe}): (
@ -109,15 +109,15 @@ query = <<EOQ
) < 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -134,15 +134,15 @@ resource "datadog_monitor" "total_devices" {
) == 0
EOQ
notify_no_data = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.total_devices_extra_tags]
}
@ -167,15 +167,15 @@ warning = var.failed_c2d_methods_rate_threshold_warning
critical = var.failed_c2d_methods_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.failed_c2d_methods_rate_extra_tags]
}
@ -184,7 +184,7 @@ resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
count = var.failed_c2d_twin_read_rate_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] IOT Hub Too many c2d twin read failure {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.failed_c2d_twin_read_rate_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.failed_c2d_twin_read_rate_time_aggregator}(${var.failed_c2d_twin_read_rate_timeframe}):
@ -200,15 +200,15 @@ warning = var.failed_c2d_twin_read_rate_threshold_warning
critical = var.failed_c2d_twin_read_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.failed_c2d_twin_read_rate_extra_tags]
}
@ -217,7 +217,7 @@ resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
count = var.failed_c2d_twin_update_rate_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] IOT Hub Too many c2d twin update failure {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.failed_c2d_twin_update_rate_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.failed_c2d_twin_update_rate_time_aggregator}(${var.failed_c2d_twin_update_rate_timeframe}):
@ -233,15 +233,15 @@ EOQ
critical = var.failed_c2d_twin_update_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.failed_c2d_twin_update_rate_extra_tags]
}
@ -266,15 +266,15 @@ EOQ
critical = var.failed_d2c_twin_read_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.failed_d2c_twin_read_rate_extra_tags]
}
@ -299,15 +299,15 @@ warning = var.failed_d2c_twin_update_rate_threshold_warning
critical = var.failed_d2c_twin_update_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.failed_d2c_twin_update_rate_extra_tags]
}
@ -316,7 +316,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
count = var.dropped_d2c_telemetry_egress_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] IOT Hub Too many d2c telemetry egress dropped {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
message = coalesce(var.dropped_d2c_telemetry_egress_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.dropped_d2c_telemetry_egress_time_aggregator}(${var.dropped_d2c_telemetry_egress_timeframe}):
@ -334,15 +334,15 @@ warning = var.dropped_d2c_telemetry_egress_rate_threshold_warning
critical = var.dropped_d2c_telemetry_egress_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.dropped_d2c_telemetry_egress_extra_tags]
}
@ -351,7 +351,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
count = var.orphaned_d2c_telemetry_egress_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] IOT Hub Too many d2c telemetry egress orphaned {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
message = coalesce(var.orphaned_d2c_telemetry_egress_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.orphaned_d2c_telemetry_egress_time_aggregator}(${var.orphaned_d2c_telemetry_egress_timeframe}):
@ -369,15 +369,15 @@ EOQ
critical = var.orphaned_d2c_telemetry_egress_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.orphaned_d2c_telemetry_egress_extra_tags]
}
@ -404,15 +404,15 @@ EOQ
critical = var.invalid_d2c_telemetry_egress_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.invalid_d2c_telemetry_egress_extra_tags]
}
@ -431,15 +431,15 @@ resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" {
, 0) > 0
EOQ
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:iothubs", "team:claranet", "created-by:terraform", var.too_many_d2c_telemetry_ingress_nosent_extra_tags]
}

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

@ -10,15 +10,15 @@ resource "datadog_monitor" "keyvault_status" {
) < 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:keyvault", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -43,15 +43,15 @@ critical = var.api_result_threshold_critical
warning = var.api_result_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:keyvault", "team:claranet", "created-by:terraform", var.api_result_extra_tags]
}
@ -73,15 +73,15 @@ critical = var.api_latency_threshold_critical
warning = var.api_latency_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:keyvault", "team:claranet", "created-by:terraform", var.api_latency_extra_tags]
}

4
cloud/azure/load-balancer/monitors-load-balancer.tf
View File

@ -10,15 +10,15 @@ resource "datadog_monitor" "loadbalancer_status" {
) < 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:load-balancer", "team:claranet", "created-by:terraform", var.status_extra_tags]
}

19
cloud/azure/mysql/monitors-mysql.tf
View File

@ -15,15 +15,15 @@ EOQ
warning = var.cpu_usage_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", var.cpu_usage_extra_tags]
}
@ -38,7 +38,6 @@ resource "datadog_monitor" "mysql_free_storage" {
${var.free_storage_time_aggregator}(${var.free_storage_timeframe}): (
100 - avg:azure.dbformysql_servers.storage_percent${module.filter-tags.query_alert} by {resource_group,region,name}
) < ${var.free_storage_threshold_critical}
EOQ
thresholds = {
@ -46,15 +45,15 @@ critical = var.free_storage_threshold_critical
warning = var.free_storage_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", var.free_storage_extra_tags]
}
@ -76,15 +75,15 @@ critical = var.io_consumption_threshold_critical
warning = var.io_consumption_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", var.io_consumption_extra_tags]
}
@ -93,7 +92,7 @@ resource "datadog_monitor" "mysql_memory_usage" {
count = var.memory_usage_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Mysql Server memory usage {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.memory_usage_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.memory_usage_time_aggregator}(${var.memory_usage_timeframe}): (
@ -106,15 +105,15 @@ EOQ
warning = var.memory_usage_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:mysql", "team:claranet", "created-by:terraform", var.memory_usage_extra_tags]
}

25
cloud/azure/postgresql/monitors-postegresql.tf
View File

@ -15,15 +15,15 @@ EOQ
warning = var.cpu_usage_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:postgresql", "team:claranet", "created-by:terraform", var.cpu_usage_extra_tags]
}
@ -40,15 +40,15 @@ resource "datadog_monitor" "postgresql_no_connection" {
) < 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:postgresql", "team:claranet", "created-by:terraform", var.no_connection_extra_tags]
}
@ -57,13 +57,12 @@ resource "datadog_monitor" "postgresql_free_storage" {
count = var.free_storage_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Postgresql Server storage {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.free_storage_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.free_storage_time_aggregator}(${var.free_storage_timeframe}): (
100 - avg:azure.dbforpostgresql_servers.storage_percent${module.filter-tags.query_alert} by {resource_group,region,name}
) < ${var.free_storage_threshold_critical}
EOQ
thresholds = {
@ -71,15 +70,15 @@ critical = var.free_storage_threshold_critical
warning = var.free_storage_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:postgresql", "team:claranet", "created-by:terraform", var.free_storage_extra_tags]
}
@ -88,7 +87,7 @@ resource "datadog_monitor" "postgresql_io_consumption" {
count = var.io_consumption_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Postgresql Server IO consumption {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.io_consumption_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.io_consumption_time_aggregator}(${var.io_consumption_timeframe}): (
@ -101,15 +100,15 @@ EOQ
warning = var.io_consumption_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:postgresql", "team:claranet", "created-by:terraform", var.io_consumption_extra_tags]
}
@ -131,15 +130,15 @@ EOQ
warning = var.memory_usage_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:postgresql", "team:claranet", "created-by:terraform", var.memory_usage_extra_tags]
}

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

@ -10,15 +10,15 @@ resource "datadog_monitor" "status" {
) != 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:redis", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -40,15 +40,15 @@ warning = var.evictedkeys_limit_threshold_warning
critical = var.evictedkeys_limit_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:redis", "team:claranet", "created-by:terraform", var.evictedkeys_limit_extra_tags]
}
@ -57,7 +57,7 @@ resource "datadog_monitor" "percent_processor_time" {
count = var.percent_processor_time_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Redis processor time too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.percent_processor_time_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.percent_processor_time_time_aggregator}(${var.percent_processor_time_timeframe}): (
@ -70,15 +70,15 @@ warning = var.percent_processor_time_threshold_warning
critical = var.percent_processor_time_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:redis", "team:claranet", "created-by:terraform", var.percent_processor_time_extra_tags]
}
@ -87,7 +87,7 @@ resource "datadog_monitor" "server_load" {
count = var.server_load_rate_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Redis server load too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.server_load_rate_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.server_load_rate_time_aggregator}(${var.server_load_rate_timeframe}): (
@ -100,15 +100,15 @@ EOQ
critical = var.server_load_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:redis", "team:claranet", "created-by:terraform", var.server_load_rate_extra_tags]
}

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

@ -10,15 +10,15 @@ resource "datadog_monitor" "status" {
) != 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:serverfarms", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -40,15 +40,15 @@ warning = var.cpu_percentage_threshold_warning
critical = var.cpu_percentage_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:serverfarms", "team:claranet", "created-by:terraform", var.cpu_percentage_extra_tags]
}
@ -70,15 +70,15 @@ warning = var.memory_percentage_threshold_warning
critical = var.memory_percentage_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:serverfarms", "team:claranet", "created-by:terraform", var.memory_percentage_extra_tags]
}

20
cloud/azure/servicebus/monitors-service-bus.tf
View File

@ -10,15 +10,15 @@ resource "datadog_monitor" "servicebus_status" {
) != 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:servicebus", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -35,15 +35,15 @@ resource "datadog_monitor" "service_bus_no_active_connections" {
) < 1
EOQ
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "resource:servicebus", "team:azure", "provider:azure"]
}
@ -52,7 +52,7 @@ resource "datadog_monitor" "service_bus_user_errors" {
count = var.user_errors_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Service Bus user errors rate is high {{#is_alert}}{{comparator}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{comparator}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.user_errors_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.user_errors_time_aggregator}(${var.user_errors_timeframe}): (
@ -66,15 +66,15 @@ critical = var.user_errors_threshold_critical
warning = var.user_errors_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "resource:servicebus", "team:azure", "provider:azure"]
}
@ -83,7 +83,7 @@ resource "datadog_monitor" "service_bus_server_errors" {
count = var.server_errors_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Service Bus server errors rate is high {{#is_alert}}{{comparator}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{comparator}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.server_errors_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.server_errors_time_aggregator}(${var.server_errors_timeframe}): (
@ -97,15 +97,15 @@ EOQ
warning = var.server_errors_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "resource:servicebus", "team:azure", "provider:azure"]
}

24
cloud/azure/sql-database/monitors-sql-database.tf
View File

@ -10,15 +10,15 @@ resource "datadog_monitor" "status" {
) != 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:sql-database", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -40,15 +40,15 @@ critical = var.cpu_threshold_critical
warning = var.cpu_threshold_warning
}
notify_no_data = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:sql-database", "team:claranet", "created-by:terraform", var.cpu_extra_tags]
}
@ -57,7 +57,7 @@ resource "datadog_monitor" "sql-database_free_space_low" {
count = var.diskspace_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] SQL Database high disk usage {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.diskspace_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.diskspace_time_aggregator}(${var.diskspace_timeframe}): (
@ -70,15 +70,15 @@ warning = var.diskspace_threshold_warning
critical = var.diskspace_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:sql-database", "team:claranet", "created-by:terraform", var.diskspace_extra_tags]
}
@ -87,7 +87,7 @@ resource "datadog_monitor" "sql-database_dtu_consumption_high" {
count = var.dtu_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] SQL Database DTU Consumption too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.dtu_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.dtu_time_aggregator}(${var.dtu_timeframe}): (
@ -100,15 +100,15 @@ EOQ
critical = var.dtu_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:sql-database", "team:claranet", "created-by:terraform", var.dtu_extra_tags]
}
@ -129,15 +129,15 @@ EOQ
critical = var.deadlock_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:sql-database", "team:claranet", "created-by:terraform", var.deadlock_extra_tags]
}

14
cloud/azure/sql-elasticpool/monitors-sql-elasticpool.tf
View File

@ -15,15 +15,15 @@ EOQ
warning = var.cpu_threshold_warning
}
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:sql-elasticpool", "team:claranet", "created-by:terraform", var.cpu_extra_tags]
}
@ -45,15 +45,15 @@ warning = var.diskspace_threshold_warning
critical = var.diskspace_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:sql-elasticpool", "team:claranet", "created-by:terraform", var.diskspace_extra_tags]
}
@ -62,7 +62,7 @@ resource "datadog_monitor" "sql_elasticpool_dtu_consumption_high" {
count = var.dtu_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] SQL Elastic Pool DTU Consumption too high {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.dtu_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.dtu_time_aggregator}(${var.dtu_timeframe}): (
@ -75,15 +75,15 @@ warning = var.dtu_threshold_warning
critical = var.dtu_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:sql-elasticpool", "team:claranet", "created-by:terraform", var.dtu_extra_tags]
}

45
cloud/azure/storage/monitors-azure-storage.tf
View File

@ -15,15 +15,14 @@ EOQ
warning = var.availability_threshold_warning
}
type = "metric alert"
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.availability_extra_tags]
@ -46,14 +45,14 @@ critical = var.successful_requests_threshold_critical
warning = var.successful_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.successful_requests_extra_tags]
@ -63,7 +62,7 @@ resource "datadog_monitor" "latency" {
count = var.latency_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Azure Storage too high end to end latency {{#is_alert}}{{{comparator}}} {{threshold}}ms ({{value}}ms){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}ms ({{value}}ms){{/is_warning}}"
message = coalesce(var.latency_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.latency_time_aggregator}(${var.latency_timeframe}): (default(
@ -76,14 +75,14 @@ critical = var.latency_threshold_critical
warning = var.latency_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.latency_extra_tags]
@ -93,7 +92,7 @@ resource "datadog_monitor" "timeout_error_requests" {
count = var.timeout_error_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Azure Storage too many timeout errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.timeout_error_requests_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.timeout_error_requests_time_aggregator}(${var.timeout_error_requests_timeframe}): (default(
@ -106,14 +105,14 @@ EOQ
warning = var.timeout_error_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.timeout_error_requests_extra_tags]
@ -136,14 +135,14 @@ EOQ
warning = var.network_error_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.network_error_requests_extra_tags]
@ -166,14 +165,14 @@ critical = var.throttling_error_requests_threshold_critical
warning = var.throttling_error_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.throttling_error_requests_extra_tags]
@ -183,7 +182,7 @@ resource "datadog_monitor" "server_other_error_requests" {
count = var.server_other_error_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Azure Storage too many server_other errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.server_other_error_requests_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.server_other_error_requests_time_aggregator}(${var.server_other_error_requests_timeframe}): (default(
@ -196,14 +195,14 @@ critical = var.server_other_error_requests_threshold_critical
warning = var.server_other_error_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.server_other_error_requests_extra_tags]
@ -213,7 +212,7 @@ resource "datadog_monitor" "client_other_error_requests" {
count = var.client_other_error_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Azure Storage too many client_other errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.client_other_error_requests_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.client_other_error_requests_time_aggregator}(${var.client_other_error_requests_timeframe}): (default(
@ -226,14 +225,14 @@ EOQ
warning = var.client_other_error_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.client_other_error_requests_extra_tags]
@ -256,14 +255,14 @@ EOQ
warning = var.authorization_error_requests_threshold_warning
}
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
notify_no_data = false
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
evaluation_delay = var.evaluation_delay
renotify_interval = 0
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:storage", "team:claranet", "created-by:terraform", var.authorization_error_requests_extra_tags]

25
cloud/azure/stream-analytics/monitors-stream-analytics.tf
View File

@ -10,15 +10,15 @@ resource "datadog_monitor" "status" {
) < 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:stream-analytics", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -35,15 +35,15 @@ resource "datadog_monitor" "su_utilization" {
) > ${var.su_utilization_threshold_critical}
EOQ
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
thresholds = {
warning = var.su_utilization_threshold_warning
@ -57,7 +57,7 @@ resource "datadog_monitor" "failed_function_requests" {
count = var.failed_function_requests_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Stream Analytics too many failed requests {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.failed_function_requests_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.failed_function_requests_time_aggregator}(${var.failed_function_requests_timeframe}): (
@ -69,17 +69,16 @@ EOQ
thresholds = {
warning = var.failed_function_requests_threshold_warning
critical = var.failed_function_requests_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 60
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:stream-analytics", "team:claranet", "created-by:terraform", var.failed_function_requests_extra_tags]
}
@ -88,7 +87,7 @@ resource "datadog_monitor" "conversion_errors" {
count = var.conversion_errors_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Stream Analytics too many conversion errors {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.conversion_errors_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.conversion_errors_time_aggregator}(${var.conversion_errors_timeframe}): (
@ -101,15 +100,15 @@ EOQ
critical = var.conversion_errors_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:stream-analytics", "team:claranet", "created-by:terraform", var.conversion_errors_extra_tags]
}
@ -131,15 +130,15 @@ EOQ
critical = var.runtime_errors_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:stream-analytics", "team:claranet", "created-by:terraform", var.runtime_errors_extra_tags]
}

14
cloud/azure/virtual-machine/monitors-virtual-machine.tf
View File

@ -10,15 +10,15 @@ resource "datadog_monitor" "virtualmachine_status" {
) < 1
EOQ
notify_no_data = true
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:virtualmachine", "team:claranet", "created-by:terraform", var.status_extra_tags]
}
@ -40,15 +40,15 @@ critical = var.cpu_usage_threshold_critical
warning = var.cpu_usage_threshold_warning
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 0
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:virtualmachine", "team:claranet", "created-by:terraform", var.cpu_usage_extra_tags]
}
@ -57,7 +57,7 @@ resource "datadog_monitor" "virtualmachine_credit_cpu_remaining_too_low" {
count = var.cpu_remaining_rate_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Virtual Machine credit CPU {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.cpu_remaining_rate_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.cpu_remaining_rate_time_aggregator}(${var.cpu_remaining_rate_timeframe}):
@ -73,15 +73,15 @@ warning = var.cpu_remaining_rate_threshold_warning
critical = var.cpu_remaining_rate_threshold_critical
}
notify_no_data = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = false
renotify_interval = 0
notify_audit = false
timeout_h = 1
include_tags = true
locked = false
require_full_window = false
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:azure", "resource:virtualmachine", "team:claranet", "created-by:terraform", var.cpu_remaining_rate_extra_tags]
}

52
cloud/gcp/big-query/monitors-big-query.tf
View File

@ -18,6 +18,8 @@ EOQ
critical = var.concurrent_queries_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -26,9 +28,6 @@ EOQ
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.concurrent_queries_extra_tags]
}
@ -52,6 +51,8 @@ warning = var.execution_time_threshold_warning
critical = var.execution_time_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -60,9 +61,6 @@ notify_audit = false
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.execution_time_extra_tags]
}
@ -73,7 +71,7 @@ resource "datadog_monitor" "scanned_bytes" {
count = var.scanned_bytes_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] GCP Big Query Scanned Bytes {{#is_alert}}{{{comparator}}} {{threshold}}B/mn ({{value}}B/mn){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}B/mn ({{value}}B/mn){{/is_warning}}"
message = coalesce(var.scanned_bytes_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
avg(${var.scanned_bytes_timeframe}):
@ -86,6 +84,8 @@ warning = var.scanned_bytes_threshold_warning
critical = var.scanned_bytes_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -94,9 +94,6 @@ notify_audit = false
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.scanned_bytes_extra_tags]
}
@ -107,7 +104,7 @@ resource "datadog_monitor" "scanned_bytes_billed" {
count = var.scanned_bytes_billed_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] GCP Big Query Scanned Bytes Billed {{#is_alert}}{{{comparator}}} {{threshold}}B/mn ({{value}}B/mn){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}B/mn ({{value}}B/mn){{/is_warning}}"
message = coalesce(var.scanned_bytes_billed_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
avg(${var.scanned_bytes_billed_timeframe}):
@ -120,6 +117,8 @@ EOQ
critical = var.scanned_bytes_billed_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -128,9 +127,6 @@ EOQ
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.scanned_bytes_billed_extra_tags]
}
@ -154,6 +150,8 @@ EOQ
critical = var.available_slots_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -162,9 +160,6 @@ EOQ
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.available_slots_extra_tags]
}
@ -183,12 +178,13 @@ resource "datadog_monitor" "stored_bytes" {
> ${var.stored_bytes_threshold_critical}
EOQ
thresholds = {
warning = var.stored_bytes_threshold_warning
critical = var.stored_bytes_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -197,9 +193,6 @@ notify_audit = false
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.stored_bytes_extra_tags]
}
@ -223,6 +216,8 @@ warning = var.table_count_threshold_warning
critical = var.table_count_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -231,9 +226,6 @@ notify_audit = false
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.table_count_extra_tags]
}
@ -244,7 +236,7 @@ resource "datadog_monitor" "uploaded_bytes" {
count = var.uploaded_bytes_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] GCP Big Query Uploaded Bytes {{#is_alert}}{{{comparator}}} {{threshold}}B/mn ({{value}}B/mn){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}B/mn ({{value}}B/mn){{/is_warning}}"
message = coalesce(var.uploaded_bytes_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
avg(${var.uploaded_bytes_timeframe}):
@ -257,6 +249,8 @@ EOQ
critical = var.uploaded_bytes_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -265,9 +259,6 @@ EOQ
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.uploaded_bytes_extra_tags]
}
@ -291,6 +282,8 @@ EOQ
critical = var.uploaded_bytes_billed_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
include_tags = true
notify_no_data = false
require_full_window = false
@ -299,9 +292,6 @@ EOQ
timeout_h = 0
locked = false
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:big-query", "team:claranet", "created-by:terraform", var.uploaded_bytes_billed_extra_tags]
}

31
cloud/gcp/cloud-sql/common/monitors-cloud-sql-common.tf
View File

@ -18,6 +18,8 @@ EOQ
critical = var.cpu_utilization_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -26,9 +28,6 @@ EOQ
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:cloud-sql", "team:claranet", "created-by:terraform", var.cpu_utilization_extra_tags]
}
@ -52,6 +51,8 @@ warning = var.disk_utilization_threshold_warning
critical = var.disk_utilization_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -60,9 +61,6 @@ require_full_window = false
notify_no_data = true
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:cloud-sql", "team:claranet", "created-by:terraform", var.disk_utilization_extra_tags]
}
@ -100,6 +98,8 @@ EOQ
critical_recovery = var.disk_utilization_forecast_threshold_critical_recovery
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -108,9 +108,6 @@ EOQ
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:cloud-sql", "team:claranet", "created-by:terraform", var.disk_utilization_forecast_extra_tags]
}
@ -134,6 +131,8 @@ EOQ
critical = var.memory_utilization_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -142,9 +141,6 @@ EOQ
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:cloud-sql", "team:claranet", "created-by:terraform", var.memory_utilization_extra_tags]
}
@ -182,6 +178,8 @@ EOQ
critical_recovery = var.memory_utilization_forecast_threshold_critical_recovery
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -190,9 +188,6 @@ EOQ
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:cloud-sql", "team:claranet", "created-by:terraform", var.memory_utilization_forecast_extra_tags]
}
@ -203,7 +198,6 @@ resource "datadog_monitor" "failover_unavailable" {
count = var.failover_unavailable_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Cloud SQL Failover Unavailable"
message = coalesce(var.failover_unavailable_message, var.message)
type = "metric alert"
query = <<EOQ
@ -217,6 +211,8 @@ thresholds = {
critical = var.failover_unavailable_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -225,9 +221,6 @@ require_full_window = false
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:cloud-sql", "team:claranet", "created-by:terraform", var.failover_unavailable_extra_tags]
}

5
cloud/gcp/cloud-sql/mysql/monitors-cloudsql-mysql.tf
View File

@ -18,6 +18,8 @@ EOQ
warning = var.replication_lag_threshold_warning
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -26,9 +28,6 @@ EOQ
notify_no_data = true
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:cloud-sql", "team:claranet", "created-by:terraform", "engine:mysql", var.replication_lag_extra_tags]
}

19
cloud/gcp/gce/instance/monitors-gce-instance.tf
View File

@ -18,6 +18,8 @@ EOQ
critical = var.cpu_utilization_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -26,9 +28,6 @@ EOQ
notify_no_data = true
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:gce-instance", "team:claranet", "created-by:terraform", var.cpu_utilization_extra_tags]
}
@ -58,6 +57,8 @@ warning = var.disk_throttled_bps_threshold_warning
critical = var.disk_throttled_bps_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -66,9 +67,6 @@ require_full_window = false
notify_no_data = var.disk_throttled_bps_notify_no_data
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:gce-instance", "team:claranet", "created-by:terraform", var.disk_throttled_bps_extra_tags]
}
@ -79,7 +77,7 @@ resource "datadog_monitor" "disk_throttled_ops" {
count = var.disk_throttled_ops_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Compute Engine instance Disk Throttled OPS {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
message = coalesce(var.disk_throttled_ops_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.disk_throttled_ops_time_aggregator}(${var.disk_throttled_ops_timeframe}):
@ -98,17 +96,16 @@ warning = var.disk_throttled_ops_threshold_warning
critical = var.disk_throttled_ops_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = var.disk_throttled_ops_notify_no_data
notify_audit = false
locked = false
timeout_h = 0
include_tags = true
require_full_window = false
notify_no_data = var.disk_throttled_ops_notify_no_data
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:gce-instance", "team:claranet", "created-by:terraform", var.disk_throttled_ops_extra_tags]
}

29
cloud/gcp/lb/monitors-lb.tf
View File

@ -19,6 +19,8 @@ EOQ
critical = var.error_rate_4xx_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -27,9 +29,6 @@ EOQ
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:lb", "team:claranet", "created-by:terraform", var.error_rate_4xx_extra_tags]
}
@ -54,6 +53,8 @@ warning = var.error_rate_5xx_threshold_warning
critical = var.error_rate_5xx_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -62,9 +63,6 @@ require_full_window = false
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:lb", "team:claranet", "created-by:terraform", var.error_rate_5xx_extra_tags]
}
@ -75,7 +73,7 @@ resource "datadog_monitor" "backend_latency_service" {
count = var.backend_latency_service_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] GCP LB service backend latency {{#is_alert}}{{{comparator}}} {{threshold}}ms ({{value}}ms){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}ms ({{value}}ms){{/is_warning}}"
message = coalesce(var.backend_latency_service_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.backend_latency_service_time_aggregator}(${var.backend_latency_service_timeframe}):
@ -88,6 +86,8 @@ warning = var.backend_latency_service_threshold_warning
critical = var.backend_latency_service_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -96,9 +96,6 @@ require_full_window = false
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:lb", "team:claranet", "created-by:terraform", var.backend_latency_service_extra_tags]
}
@ -109,7 +106,7 @@ resource "datadog_monitor" "backend_latency_bucket" {
count = var.backend_latency_bucket_enabled == "true" ? 1 : 0
name = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] GCP LB bucket backend latency {{#is_alert}}{{{comparator}}} {{threshold}}ms ({{value}}ms){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}ms ({{value}}ms){{/is_warning}}"
message = coalesce(var.backend_latency_bucket_message, var.message)
type = "query alert"
type = "query alert"
query = <<EOQ
${var.backend_latency_bucket_time_aggregator}(${var.backend_latency_bucket_timeframe}):
@ -122,6 +119,8 @@ EOQ
critical = var.backend_latency_bucket_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -130,9 +129,6 @@ EOQ
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:lb", "team:claranet", "created-by:terraform", var.backend_latency_bucket_extra_tags]
}
@ -156,6 +152,8 @@ EOQ
critical = var.request_count_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -164,9 +162,6 @@ EOQ
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:lb", "team:claranet", "created-by:terraform", var.request_count_extra_tags]
}

10
cloud/gcp/pubsub/monitors-pubsub.tf
View File

@ -17,6 +17,8 @@ EOQ
critical = var.sending_operations_count_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -25,9 +27,6 @@ EOQ
notify_no_data = true
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:pubsub", "team:claranet", "created-by:terraform", var.sending_operations_count_extra_tags]
}
@ -51,6 +50,8 @@ warning = var.unavailable_sending_operations_count_threshold_warning
critical = var.unavailable_sending_operations_count_threshold_critical
}
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_audit = false
locked = false
timeout_h = 0
@ -59,9 +60,6 @@ require_full_window = false
notify_no_data = false
renotify_interval = 0
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
tags = ["env:${var.environment}", "type:cloud", "provider:gcp", "resource:pubsub", "team:claranet", "created-by:terraform", var.unavailable_sending_operations_count_extra_tags]
}

2
database/mongodb/monitors-mongo.tf
View File

@ -8,7 +8,6 @@ resource "datadog_monitor" "mongodb_primary" {
${var.mongodb_primary_aggregator}(${var.mongodb_primary_timeframe}):
min:mongodb.replset.state${module.filter-tags.query_alert} by {replset_name} >= 2
EOQ
evaluation_delay = var.evaluation_delay
new_host_delay = var.new_host_delay
notify_no_data = true
@ -61,7 +60,6 @@ query = <<EOQ
${var.mongodb_server_count_aggregator}(${var.mongodb_server_count_timeframe}):
sum:mongodb.replset.health${module.filter-tags.query_alert} by {replset_name}
> 99
EOQ
thresholds = {

2
database/mysql/monitors-mysql.tf
View File

@ -13,9 +13,9 @@ EOQ
critical = 5
}
notify_no_data = true
no_data_timeframe = var.mysql_availability_no_data_timeframe
new_host_delay = var.new_host_delay
notify_no_data = true
renotify_interval = 0
notify_audit = false
timeout_h = 0