Merged in MON-281-monitor-for-disk-forecast-system (pull request #121)

MON-281 Disk space forecast monitor in system generic

Approved-by: Rafael Romero Carmona <rafael.romero.carmona@fr.clara.net>
Approved-by: Quentin Manfroi <quentin.manfroi@yahoo.fr>
Approved-by: Alexandre Gaillet <alexandre.gaillet@fr.clara.net>

scripts/02_update_readme.sh

@@ -55,7 +55,7 @@ EOF
     IFS=$(echo -en "\n\b")
     for match in $(grep -E ^[[:space:]]+name[[:space:]]+= $(basename ${path})); do
         name=$(get_name "${match}")
-        echo "- $name" >> README.md
+        echo "- ${name/could reach/forecast}" >> README.md
     done
     IFS=$SAVEIFS
     terraform-docs md ./ >> README.md

system/generic/README.md

@@ -19,6 +19,7 @@ Creates DataDog monitors with the following checks:
 - CPU usage
 - CPU load 5
 - Free disk space
+- Disk Space forecast
 - Free disk inodes
 - Free memory
 
@@ -52,6 +53,19 @@ Creates DataDog monitors with the following checks:
 | free_disk_inodes_time_aggregator | Monitor aggregator for Free disk inodes [available values: min, max or avg] | string | `min` | no |
 | free_disk_inodes_timeframe | Monitor timeframe for Free disk inodes [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
 | free_disk_space_extra_tags | Extra tags for Free diskspace monitor | list | `<list>` | no |
+| free_disk_space_forecast_algorithm | Algorithm for the Free diskspace Forecast monitor [available values: `linear` or `seasonal`] | string | `linear` | no |
+| free_disk_space_forecast_deviations | Deviations for the Free diskspace Forecast monitor [available values: `1`, `2`, `3`, `4` or `5`] | string | `1` | no |
+| free_disk_space_forecast_extra_tags | Extra tags for Free diskspace forecast monitor | list | `<list>` | no |
+| free_disk_space_forecast_interval | Interval for the Free diskspace Forecast monitor [available values: `30m`, `60m` or `120m`] | string | `60m` | no |
+| free_disk_space_forecast_linear_history | History for the Free diskspace Forecast monitor [available values: `12h`, `#d` (1, 2, or 3), `#w` (1, or 2) or `#mo` (1, 2 or 3)] | string | `1w` | no |
+| free_disk_space_forecast_linear_model | Model for the Free diskspace Forecast monitor [available values: `default`, `simple` or `reactive`] | string | `default` | no |
+| free_disk_space_forecast_message | Custom message for Free diskspace forecast monitor | string | `` | no |
+| free_disk_space_forecast_seasonal_seasonality | Seasonality for the Free diskspace Forecast monitor | string | `weekly` | no |
+| free_disk_space_forecast_silenced | Groups to mute for Free diskspace forecast monitor | map | `<map>` | no |
+| free_disk_space_forecast_threshold_critical | Free disk space forecast critical threshold | string | `80` | no |
+| free_disk_space_forecast_threshold_critical_recovery | Free disk space forecast recovery threshold | string | `72` | no |
+| free_disk_space_forecast_time_aggregator | Monitor aggregator for Free diskspace forecast [available values: min, max or avg] | string | `max` | no |
+| free_disk_space_forecast_timeframe | Monitor timeframe for Free diskspace forecast [available values: `next_12h`, `next_#d` (1, 2, or 3), `next_#w` (1 or 2) or `next_#mo` (1, 2 or 3)] | string | `next_1w` | no |
 | free_disk_space_message | Custom message for Free diskspace monitor | string | `` | no |
 | free_disk_space_silenced | Groups to mute for Free diskspace monitor | map | `<map>` | no |
 | free_disk_space_threshold_critical | Free disk space critical threshold | string | `10` | no |
@@ -73,6 +87,7 @@ Creates DataDog monitors with the following checks:
 | Name | Description |
 |------|-------------|
 | datadog_cpu_too_high_id | id for monitor datadog_cpu_too_high |
+| datadog_free_disk_space_forecast_id | id for monitor datadog_free_disk_space_forecast |
 | datadog_free_disk_space_inodes_too_low_id | id for monitor datadog_free_disk_space_inodes_too_low |
 | datadog_free_disk_space_too_low_id | id for monitor datadog_free_disk_space_too_low |
 | datadog_free_memory_id | id for monitor datadog_free_memory |

system/generic/inputs.tf

@@ -151,6 +151,82 @@ variable "free_disk_space_threshold_critical" {
   default     = 10
 }
 
+variable "free_disk_space_forecast_silenced" {
+  description = "Groups to mute for Free diskspace forecast monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "free_disk_space_forecast_extra_tags" {
+  description = "Extra tags for Free diskspace forecast monitor"
+  type        = "list"
+  default     = []
+}
+
+variable "free_disk_space_forecast_message" {
+  description = "Custom message for Free diskspace forecast monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "free_disk_space_forecast_time_aggregator" {
+  description = "Monitor aggregator for Free diskspace forecast [available values: min, max or avg]"
+  type        = "string"
+  default     = "max"
+}
+
+variable "free_disk_space_forecast_timeframe" {
+  description = "Monitor timeframe for Free diskspace forecast [available values: `next_12h`, `next_#d` (1, 2, or 3), `next_#w` (1 or 2) or `next_#mo` (1, 2 or 3)]"
+  type        = "string"
+  default     = "next_1w"
+}
+
+variable "free_disk_space_forecast_algorithm" {
+  description = "Algorithm for the Free diskspace Forecast monitor [available values: `linear` or `seasonal`]"
+  type        = "string"
+  default     = "linear"
+}
+
+variable "free_disk_space_forecast_deviations" {
+  description = "Deviations for the Free diskspace Forecast monitor [available values: `1`, `2`, `3`, `4` or `5`]"
+  type        = "string"
+  default     = 1
+}
+
+variable "free_disk_space_forecast_interval" {
+  description = "Interval for the Free diskspace Forecast monitor [available values: `30m`, `60m` or `120m`]"
+  type        = "string"
+  default     = "60m"
+}
+
+variable "free_disk_space_forecast_linear_history" {
+  description = "History for the Free diskspace Forecast monitor [available values: `12h`, `#d` (1, 2, or 3), `#w` (1, or 2) or `#mo` (1, 2 or 3)]"
+  type        = "string"
+  default     = "1w"
+}
+
+variable "free_disk_space_forecast_linear_model" {
+  description = "Model for the Free diskspace Forecast monitor [available values: `default`, `simple` or `reactive`]"
+  type        = "string"
+  default     = "default"
+}
+
+variable "free_disk_space_forecast_seasonal_seasonality" {
+  description = "Seasonality for the Free diskspace Forecast monitor"
+  type        = "string"
+  default     = "weekly"
+}
+
+variable "free_disk_space_forecast_threshold_critical_recovery" {
+  description = "Free disk space forecast recovery threshold"
+  default     = 72
+}
+
+variable "free_disk_space_forecast_threshold_critical" {
+  description = "Free disk space forecast critical threshold"
+  default     = 80
+}
+
 variable "free_disk_inodes_silenced" {
   description = "Groups to mute for Free disk inodes monitor"
   type        = "map"

system/generic/monitors-system.tf

@@ -93,6 +93,45 @@ resource "datadog_monitor" "datadog_free_disk_space_too_low" {
   tags = ["env:${var.environment}", "type:system", "provider:disk", "resource:generic", "team:claranet", "created-by:terraform", "${var.free_disk_space_extra_tags}"]
 }
 
+resource "datadog_monitor" "datadog_free_disk_space_forecast" {
+  name    = "[${var.environment}] Disk Space could reach {{#is_alert}}{{threshold}}%{{/is_alert}} in a near future"
+  message = "${coalesce(var.free_disk_space_forecast_message, var.message)}"
+
+  query = <<EOF
+    ${var.free_disk_space_forecast_time_aggregator}(${var.free_disk_space_forecast_timeframe}):
+      forecast(avg:system.disk.in_use${module.filter-tags-disk.query_alert} by {host,device} * 100,
+              '${var.free_disk_space_forecast_algorithm}',
+               ${var.free_disk_space_forecast_deviations},
+               interval='${var.free_disk_space_forecast_interval}',
+               ${var.free_disk_space_forecast_algorithm == "linear" ? format("history='%s',model='%s'", var.free_disk_space_forecast_linear_history, var.free_disk_space_forecast_linear_model): ""}
+               ${var.free_disk_space_forecast_algorithm == "seasonal" ? format("seasonality='%s'", var.free_disk_space_forecast_seasonal_seasonality): ""}
+              )
+    >= ${var.free_disk_space_forecast_threshold_critical}
+  EOF
+
+  type = "query alert"
+
+  thresholds {
+    critical_recovery = "${var.free_disk_space_forecast_threshold_critical_recovery}"
+    critical          = "${var.free_disk_space_forecast_threshold_critical}"
+  }
+
+  notify_audit        = false
+  locked              = false
+  timeout_h           = 0
+  include_tags        = true
+  require_full_window = true
+  notify_no_data      = false
+  renotify_interval   = 0
+
+  evaluation_delay = "${var.evaluation_delay}"
+  new_host_delay   = "${var.new_host_delay}"
+
+  silenced = "${var.free_disk_space_forecast_silenced}"
+
+  tags = ["env:${var.environment}", "type:system", "provider:disk", "resource:generic", "team:claranet", "created-by:terraform", "${var.free_disk_space_forecast_extra_tags}"]
+}
+
 resource "datadog_monitor" "datadog_free_disk_space_inodes_too_low" {
   name    = "[${var.environment}] Free disk inodes {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
   message = "${coalesce(var.free_disk_inodes_message, var.message)}"

system/generic/outputs.tf

@@ -13,6 +13,11 @@ output "datadog_free_disk_space_too_low_id" {
   value       = "${datadog_monitor.datadog_free_disk_space_too_low.id}"
 }
 
+output "datadog_free_disk_space_forecast_id" {
+  description = "id for monitor datadog_free_disk_space_forecast"
+  value       = "${datadog_monitor.datadog_free_disk_space_forecast.id}"
+}
+
 output "datadog_free_disk_space_inodes_too_low_id" {
   description = "id for monitor datadog_free_disk_space_inodes_too_low"
   value       = "${datadog_monitor.datadog_free_disk_space_inodes_too_low.id}"