Added systemd unit monitor

host/systemd/README.md

@@ -0,0 +1,68 @@
+<!-- BEGIN_TF_DOCS -->
+## Requirements
+
+| Name | Version |
+|------|---------|
+| <a name="requirement_terraform"></a> [terraform](#requirement\_terraform) | ~> 1.5 |
+| <a name="requirement_datadog"></a> [datadog](#requirement\_datadog) | >= 3.37 |
+| <a name="requirement_null"></a> [null](#requirement\_null) | >= 3.1.0 |
+
+## Providers
+
+| Name | Version |
+|------|---------|
+| <a name="provider_datadog"></a> [datadog](#provider\_datadog) | >= 3.37 |
+
+## Modules
+
+No modules.
+
+## Resources
+
+| Name | Type |
+|------|------|
+| [datadog_monitor.systemd_unit](https://registry.terraform.io/providers/datadog/datadog/latest/docs/resources/monitor) | resource |
+
+## Inputs
+
+| Name | Description | Type | Default | Required |
+|------|-------------|------|---------|:--------:|
+| <a name="input_additional_tags"></a> [additional\_tags](#input\_additional\_tags) | Additional tags to apply to all monitors | `list(string)` | `[]` | no |
+| <a name="input_alert_critical_priority"></a> [alert\_critical\_priority](#input\_alert\_critical\_priority) | Priority for alerts within critical threshold (P1-P5, uses monitor defaults if not specified) | `string` | `null` | no |
+| <a name="input_alert_message"></a> [alert\_message](#input\_alert\_message) | Message to prepend to alert notifications | `string` | `"Alert"` | no |
+| <a name="input_alert_nodata_priority"></a> [alert\_nodata\_priority](#input\_alert\_nodata\_priority) | Priority for alerts within warning threshold (P1-P5, uses monitor defaults if not specified) | `string` | `null` | no |
+| <a name="input_base_tags"></a> [base\_tags](#input\_base\_tags) | Base tags to apply to all monitors | `list(string)` | `[]` | no |
+| <a name="input_cost_center"></a> [cost\_center](#input\_cost\_center) | Cost Center of the monitored resource (leave blank to omit tag) | `string` | `null` | no |
+| <a name="input_dashboard_link"></a> [dashboard\_link](#input\_dashboard\_link) | Dashboard link to include in message | `string` | `null` | no |
+| <a name="input_env"></a> [env](#input\_env) | Environment the monitored resource is in (leave blank to omit tag) | `string` | `null` | no |
+| <a name="input_evaluation_delay"></a> [evaluation\_delay](#input\_evaluation\_delay) | Monitor evaluation delay (see [https://docs.datadoghq.com/monitors/configuration/?tab=thresholdalert#set-alert-conditions](Datadog Docs)) | `number` | `900` | no |
+| <a name="input_monitor_exclude_tags"></a> [monitor\_exclude\_tags](#input\_monitor\_exclude\_tags) | Tags to be excluded in the monitoring query. Specify in key:value format | `list(string)` | `[]` | no |
+| <a name="input_monitor_include_tags"></a> [monitor\_include\_tags](#input\_monitor\_include\_tags) | Tags to be included in the monitoring query. Specify in key:value format | `list(string)` | `[]` | no |
+| <a name="input_new_group_delay"></a> [new\_group\_delay](#input\_new\_group\_delay) | Delay in seconds before generating alerts for a new resource | `number` | `300` | no |
+| <a name="input_notify_alert_override"></a> [notify\_alert\_override](#input\_notify\_alert\_override) | List of notifications for alerts in critical threshold (uses `notify_default` otherwise) | `list(string)` | `[]` | no |
+| <a name="input_notify_crit_override"></a> [notify\_crit\_override](#input\_notify\_crit\_override) | List of notifications for 24x7 alerts in critical threshold (uses `notify_default` otherwise) | `list(string)` | `[]` | no |
+| <a name="input_notify_default"></a> [notify\_default](#input\_notify\_default) | List of alert notifications (can be overridden based on alert type) | `list(string)` | n/a | yes |
+| <a name="input_notify_no_data"></a> [notify\_no\_data](#input\_notify\_no\_data) | Alert if no matching data is found | `bool` | `false` | no |
+| <a name="input_notify_nodata_override"></a> [notify\_nodata\_override](#input\_notify\_nodata\_override) | List of notifications for no data (uses `notify_default` otherwise) | `list(string)` | `[]` | no |
+| <a name="input_notify_nonprod_override"></a> [notify\_nonprod\_override](#input\_notify\_nonprod\_override) | List of notifications for non-prod alerts in critical threshold (uses `notify_default` otherwise) | `list(string)` | `[]` | no |
+| <a name="input_notify_prod_override"></a> [notify\_prod\_override](#input\_notify\_prod\_override) | List of notifications for 12x5 prod alerts in critical threshold (uses `notify_default` otherwise) | `list(string)` | `[]` | no |
+| <a name="input_notify_recovery_override"></a> [notify\_recovery\_override](#input\_notify\_recovery\_override) | List of notifications for alert recovery (uses `notify_default` otherwise) | `list(string)` | `[]` | no |
+| <a name="input_notify_warn_override"></a> [notify\_warn\_override](#input\_notify\_warn\_override) | List of notifications for alerts in warning threshold (uses `notify_default` otherwise) | `list(string)` | `[]` | no |
+| <a name="input_renotify_interval"></a> [renotify\_interval](#input\_renotify\_interval) | Interval in minutes to re-send notifications about an alert | `number` | `60` | no |
+| <a name="input_runbook_link"></a> [runbook\_link](#input\_runbook\_link) | Runbook link to include in message | `string` | `null` | no |
+| <a name="input_service"></a> [service](#input\_service) | Service associated with the monitored resource (leave blank to omit tag) | `string` | `null` | no |
+| <a name="input_systemd_unit_alert_enabled"></a> [systemd\_unit\_alert\_enabled](#input\_systemd\_unit\_alert\_enabled) | Enable or disable the Systemd service alert monitor | `bool` | `true` | no |
+| <a name="input_systemd_unit_alert_threshold_critical"></a> [systemd\_unit\_alert\_threshold\_critical](#input\_systemd\_unit\_alert\_threshold\_critical) | Critical threshold for the Systemd service alert (count of services not running/failed) | `number` | `2` | no |
+| <a name="input_systemd_unit_alert_threshold_warning"></a> [systemd\_unit\_alert\_threshold\_warning](#input\_systemd\_unit\_alert\_threshold\_warning) | Warning threshold for the Systemd service alert (count of services not running/failed) | `number` | `1` | no |
+| <a name="input_systemd_unit_alert_use_message"></a> [systemd\_unit\_alert\_use\_message](#input\_systemd\_unit\_alert\_use\_message) | Whether to use the base message for the Systemd service alert | `bool` | `true` | no |
+| <a name="input_systemd_units_filter"></a> [systemd\_units\_filter](#input\_systemd\_units\_filter) | List of specific systemd units (services) to monitor. If empty, monitors all. | `list(string)` | `[]` | no |
+| <a name="input_team"></a> [team](#input\_team) | Team supporting the monitored resource (leave blank to omit tag) | `string` | `null` | no |
+| <a name="input_timeout_h"></a> [timeout\_h](#input\_timeout\_h) | Auto-resolve alert in specified hours if condition no longer matches | `number` | `0` | no |
+| <a name="input_title_prefix"></a> [title\_prefix](#input\_title\_prefix) | Prefix all alerts with specified value in brackets | `string` | `null` | no |
+| <a name="input_title_suffix"></a> [title\_suffix](#input\_title\_suffix) | Suffix all alerts with specified value in parenthesis | `string` | `null` | no |
+| <a name="input_warn_priority"></a> [warn\_priority](#input\_warn\_priority) | Priority for alerts with no data (P1-P5, uses monitor defaults if not specified) | `string` | `null` | no |
+
+## Outputs
+
+No outputs.
+<!-- END_TF_DOCS -->

host/systemd/common.tf

@@ -0,0 +1 @@
+../../common/common.tf

host/systemd/main.tf

@@ -0,0 +1,34 @@
+locals {
+  monitor_alert_default_priority  = null
+  monitor_warn_default_priority   = null
+  monitor_nodata_default_priority = null
+
+  title_prefix = var.title_prefix == null ? "" : "[${var.title_prefix}]"
+  title_suffix = var.title_suffix == null ? "" : " (${var.title_suffix})"
+}
+
+resource "datadog_monitor" "systemd_unit" {
+  count = var.systemd_unit_alert_enabled ? 1 : 0
+
+  name    = join("", [local.title_prefix, "Systemd Unit Status - {{host.name}}", local.title_suffix])
+  type    = "service check"
+  message = var.systemd_unit_alert_use_message ? local.query_alert_base_message : ""
+  tags    = concat(local.common_tags, var.base_tags, var.additional_tags)
+
+  evaluation_delay    = var.evaluation_delay
+  notify_no_data      = false
+  notify_audit        = false
+  renotify_interval   = 60
+  timeout_h           = var.timeout_h
+  include_tags        = false
+  require_full_window = false
+
+  query = <<EOT
+"systemd.unit.state"${local.service_filter}.by("host","unit").last(3).count_by_status()
+EOT
+
+  monitor_thresholds {
+    critical = var.systemd_unit_alert_threshold_critical
+    warning  = var.systemd_unit_alert_threshold_warning
+  }
+}

host/systemd/variables.tf

@@ -0,0 +1,41 @@
+variable "systemd_unit_alert_enabled" {
+  description = "Enable or disable the Systemd service alert monitor"
+  type        = bool
+  default     = true
+}
+
+variable "systemd_unit_alert_use_message" {
+  description = "Whether to use the base message for the Systemd service alert"
+  type        = bool
+  default     = true
+}
+
+variable "systemd_unit_alert_threshold_critical" {
+  description = "Critical threshold for the Systemd service alert (count of services not running/failed)"
+  type        = number
+  default     = 2
+}
+
+variable "systemd_unit_alert_threshold_warning" {
+  description = "Warning threshold for the Systemd service alert (count of services not running/failed)"
+  type        = number
+  default     = 1
+}
+
+variable "systemd_units_filter" {
+  description = "List of specific systemd units (services) to monitor. If empty, monitors all."
+  type        = list(string)
+  default     = []
+}
+
+variable "base_tags" {
+  description = "Base tags to apply to all monitors"
+  type        = list(string)
+  default     = []
+}
+
+variable "additional_tags" {
+  description = "Additional tags to apply to all monitors"
+  type        = list(string)
+  default     = []
+}

host/systemd/versions.tf

@@ -0,0 +1 @@
+../../common/versions.tf

host/windows/main.tf

@@ -18,7 +18,7 @@ resource "datadog_monitor" "windows_service" {
 
   evaluation_delay    = var.evaluation_delay
   notify_no_data      = false
-  renotify_interval   = 0
+  renotify_interval   = 60
   notify_audit        = false
   timeout_h           = var.timeout_h
   include_tags        = false