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

MON-80 Normalize monitors

Browse Source
This commit is contained in:
Laurent Piroelle 2017-11-23 15:12:54 +01:00
parent e0fa47008a
commit 279778ed88
3 changed files with 42 additions and 57 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
cloud/azure/iothubs/README.md
View File

@ -8,9 +8,8 @@ How to use this module
module "iothubs" { module "iothubs" {
source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/azure/iothubs?ref=MON-80-azure-hub-iot-monitors" source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/azure/iothubs?ref=MON-80-azure-hub-iot-monitors"
message = "${module.datadog-message-alerting.alerting-message}" message = "${module.datadog-message-alerting.alerting-message}"
environment = "${var.environment}" environment = "${var.environment}"
subscription_id = "${var.subscription_id}"
} }
``` ```
@ -61,15 +60,13 @@ Inputs
| failed_queryjobs_rate_threshold_warning | QueryJobs Failed rate limit (warning threshold) | string | `0` | no | | failed_queryjobs_rate_threshold_warning | QueryJobs Failed rate limit (warning threshold) | string | `0` | no |
| fallback_d2c_telemetry_egress_threshold_critical | D2C Telemetry Fallback limit (critical threshold) | string | `1000` | no | | fallback_d2c_telemetry_egress_threshold_critical | D2C Telemetry Fallback limit (critical threshold) | string | `1000` | no |
| fallback_d2c_telemetry_egress_threshold_warning | D2C Telemetry Fallback limit (warning threshold) | string | `500` | no | | fallback_d2c_telemetry_egress_threshold_warning | D2C Telemetry Fallback limit (warning threshold) | string | `500` | no |
| filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | `*` | no |
| filter_tags_use_defaults | Use default filter tags convention | string | `true` | no |
| invalid_d2c_telemetry_egress_threshold_critical | D2C Telemetry Invalid limit (critical threshold) | string | `1000` | no | | invalid_d2c_telemetry_egress_threshold_critical | D2C Telemetry Invalid limit (critical threshold) | string | `1000` | no |
| invalid_d2c_telemetry_egress_threshold_warning | D2C Telemetry Invalid limit (warning threshold) | string | `500` | no | | invalid_d2c_telemetry_egress_threshold_warning | D2C Telemetry Invalid limit (warning threshold) | string | `500` | no |
| message | Message sent when an alert is triggered | string | - | yes | | message | Message sent when an alert is triggered | string | - | yes |
| orphaned_d2c_telemetry_egress_threshold_critical | D2C Telemetry Orphaned limit (critical threshold) | string | `1000` | no | | orphaned_d2c_telemetry_egress_threshold_critical | D2C Telemetry Orphaned limit (critical threshold) | string | `1000` | no |
| orphaned_d2c_telemetry_egress_threshold_warning | D2C Telemetry Orphaned limit (warning threshold) | string | `500` | no | | orphaned_d2c_telemetry_egress_threshold_warning | D2C Telemetry Orphaned limit (warning threshold) | string | `500` | no |
| provider | Cloud provider which the monitor and its based metric depend on | string | `azure` | no |
| service | Service monitored by this set of monitors | string | `storage` | no |
| subscription_id | Azure account id used as filter for monitors | string | - | yes |
| use_filter_tags | Filter the data with service tags if true | string | `true` | no |
Related documentation Related documentation
--------------------- ---------------------

26
cloud/azure/iothubs/inputs.tf
View File

@ -4,23 +4,6 @@ variable "environment" {
type = "string" type = "string"
} }
variable "subscription_id" {
description = "Azure account id used as filter for monitors"
type = "string"
}
variable "provider" {
description = "Cloud provider which the monitor and its based metric depend on"
type = "string"
default = "azure"
}
variable "service" {
description = "Service monitored by this set of monitors"
type = "string"
default = "storage"
}
# Global DataDog # Global DataDog
variable "delay" { variable "delay" {
description = "Delay in seconds for the metric evaluation" description = "Delay in seconds for the metric evaluation"
@ -31,11 +14,16 @@ variable "message" {
description = "Message sent when an alert is triggered" description = "Message sent when an alert is triggered"
} }
variable "use_filter_tags" { variable "filter_tags_use_defaults" {
description = "Filter the data with service tags if true" description = "Use default filter tags convention"
default = "true" default = "true"
} }
variable "filter_tags_custom" {
description = "Tags used for custom filtering when filter_tags_use_defaults is false"
default = "*"
}
# Azure IOT hubs specific # Azure IOT hubs specific
variable "failed_jobs_rate_threshold_warning" { variable "failed_jobs_rate_threshold_warning" {
description = "Jobs Failed rate limit (warning threshold)" description = "Jobs Failed rate limit (warning threshold)"

62
cloud/azure/iothubs/monitors-iothubs.tf
View File

@ -2,7 +2,7 @@ data "template_file" "filter" {
template = "$${filter}" template = "$${filter}"
vars { vars {
filter = "${var.use_filter_tags == "true" ? format("dd_monitoring:enabled,dd_azure_storage:enabled,env:%s", var.environment) : "${data.template_file.filter.rendered}"}" filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_azure_iothub:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}"
} }
} }
@ -18,7 +18,7 @@ resource "datadog_monitor" "too_many_jobs_failed" {
) * 100 > ${var.failed_jobs_rate_threshold_critical} ) * 100 > ${var.failed_jobs_rate_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.failed_jobs_rate_threshold_warning}" warning = "${var.failed_jobs_rate_threshold_warning}"
@ -36,7 +36,7 @@ resource "datadog_monitor" "too_many_jobs_failed" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_list_jobs_failed" { resource "datadog_monitor" "too_many_list_jobs_failed" {
@ -51,7 +51,7 @@ resource "datadog_monitor" "too_many_list_jobs_failed" {
) * 100 > ${var.failed_listjobs_rate_threshold_critical} ) * 100 > ${var.failed_listjobs_rate_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.failed_listjobs_rate_threshold_warning}" warning = "${var.failed_listjobs_rate_threshold_warning}"
@ -69,7 +69,7 @@ resource "datadog_monitor" "too_many_list_jobs_failed" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_query_jobs_failed" { resource "datadog_monitor" "too_many_query_jobs_failed" {
@ -84,7 +84,7 @@ resource "datadog_monitor" "too_many_query_jobs_failed" {
) * 100 > ${var.failed_queryjobs_rate_threshold_critical} ) * 100 > ${var.failed_queryjobs_rate_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.failed_queryjobs_rate_threshold_warning}" warning = "${var.failed_queryjobs_rate_threshold_warning}"
@ -102,7 +102,7 @@ resource "datadog_monitor" "too_many_query_jobs_failed" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "status" { resource "datadog_monitor" "status" {
@ -113,7 +113,7 @@ resource "datadog_monitor" "status" {
avg(last_5m):avg:azure.devices_iothubs.status{${data.template_file.filter.rendered}} by {name,resource_group} < 1 avg(last_5m):avg:azure.devices_iothubs.status{${data.template_file.filter.rendered}} by {name,resource_group} < 1
EOF EOF
type = "query alert" type = "metric alert"
notify_no_data = true notify_no_data = true
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
@ -126,7 +126,7 @@ resource "datadog_monitor" "status" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "total_devices" { resource "datadog_monitor" "total_devices" {
@ -137,7 +137,7 @@ resource "datadog_monitor" "total_devices" {
avg(last_5m):avg:azure.devices_iothubs.devices.total_devices{${data.template_file.filter.rendered}} by {name,resource_group} == 0 avg(last_5m):avg:azure.devices_iothubs.devices.total_devices{${data.template_file.filter.rendered}} by {name,resource_group} == 0
EOF EOF
type = "query alert" type = "metric alert"
notify_no_data = true notify_no_data = true
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
@ -150,7 +150,7 @@ resource "datadog_monitor" "total_devices" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_c2d_methods_failed" { resource "datadog_monitor" "too_many_c2d_methods_failed" {
@ -165,7 +165,7 @@ resource "datadog_monitor" "too_many_c2d_methods_failed" {
) * 100 > ${var.failed_c2d_methods_rate_threshold_critical} ) * 100 > ${var.failed_c2d_methods_rate_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.failed_c2d_methods_rate_threshold_warning}" warning = "${var.failed_c2d_methods_rate_threshold_warning}"
@ -183,7 +183,7 @@ resource "datadog_monitor" "too_many_c2d_methods_failed" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_c2d_twin_read_failed" { resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
@ -198,7 +198,7 @@ resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
) * 100 > ${var.failed_c2d_twin_read_rate_threshold_critical} ) * 100 > ${var.failed_c2d_twin_read_rate_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.failed_c2d_twin_read_rate_threshold_warning}" warning = "${var.failed_c2d_twin_read_rate_threshold_warning}"
@ -216,7 +216,7 @@ resource "datadog_monitor" "too_many_c2d_twin_read_failed" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_c2d_twin_update_failed" { resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
@ -231,7 +231,7 @@ resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
) * 100 > ${var.failed_c2d_twin_update_rate_threshold_critical} ) * 100 > ${var.failed_c2d_twin_update_rate_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.failed_c2d_twin_update_rate_threshold_warning}" warning = "${var.failed_c2d_twin_update_rate_threshold_warning}"
@ -249,7 +249,7 @@ resource "datadog_monitor" "too_many_c2d_twin_update_failed" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_d2c_twin_read_failed" { resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
@ -264,7 +264,7 @@ resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
) * 100 > ${var.failed_d2c_twin_read_rate_threshold_critical} ) * 100 > ${var.failed_d2c_twin_read_rate_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.failed_d2c_twin_read_rate_threshold_warning}" warning = "${var.failed_d2c_twin_read_rate_threshold_warning}"
@ -282,7 +282,7 @@ resource "datadog_monitor" "too_many_d2c_twin_read_failed" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_d2c_twin_update_failed" { resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
@ -297,7 +297,7 @@ resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
) * 100 > ${var.failed_d2c_twin_update_rate_threshold_critical} ) * 100 > ${var.failed_d2c_twin_update_rate_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.failed_d2c_twin_update_rate_threshold_warning}" warning = "${var.failed_d2c_twin_update_rate_threshold_warning}"
@ -315,7 +315,7 @@ resource "datadog_monitor" "too_many_d2c_twin_update_failed" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" { resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
@ -328,7 +328,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
) > ${var.dropped_d2c_telemetry_egress_threshold_critical} ) > ${var.dropped_d2c_telemetry_egress_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.dropped_d2c_telemetry_egress_threshold_warning}" warning = "${var.dropped_d2c_telemetry_egress_threshold_warning}"
@ -346,7 +346,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_dropped" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" { resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
@ -359,7 +359,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
) > ${var.orphaned_d2c_telemetry_egress_threshold_critical} ) > ${var.orphaned_d2c_telemetry_egress_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.orphaned_d2c_telemetry_egress_threshold_warning}" warning = "${var.orphaned_d2c_telemetry_egress_threshold_warning}"
@ -377,7 +377,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_orphaned" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" { resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
@ -390,7 +390,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
) > ${var.invalid_d2c_telemetry_egress_threshold_critical} ) > ${var.invalid_d2c_telemetry_egress_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.invalid_d2c_telemetry_egress_threshold_warning}" warning = "${var.invalid_d2c_telemetry_egress_threshold_warning}"
@ -408,7 +408,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_invalid" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" { resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" {
@ -421,7 +421,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" {
) > ${var.fallback_d2c_telemetry_egress_threshold_critical} ) > ${var.fallback_d2c_telemetry_egress_threshold_critical}
EOF EOF
type = "query alert" type = "metric alert"
thresholds { thresholds {
warning = "${var.fallback_d2c_telemetry_egress_threshold_warning}" warning = "${var.fallback_d2c_telemetry_egress_threshold_warning}"
@ -439,7 +439,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_egress_fallback" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }
resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" { resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" {
@ -453,7 +453,7 @@ resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" {
) > 0 ) > 0
EOF EOF
type = "query alert" type = "metric alert"
notify_no_data = false notify_no_data = false
evaluation_delay = "${var.delay}" evaluation_delay = "${var.delay}"
@ -466,5 +466,5 @@ resource "datadog_monitor" "too_many_d2c_telemetry_ingress_nosent" {
new_host_delay = "${var.delay}" new_host_delay = "${var.delay}"
no_data_timeframe = 20 no_data_timeframe = 20
tags = ["env:${var.environment}", "resource:${var.service}", "team:${var.provider}"] tags = ["env:${var.environment}", "resource:iothub", "team:azure", "provider:azure"]
} }