MON-271: Basic monitors for Redis

README.md

@@ -100,5 +100,6 @@ The `//` is very important, it's a terraform specific syntax used to separate gi
 	- [apache](https://bitbucket.org/morea/terraform.feature.datadog/src/master/middleware/apache/)
 	- [nginx](https://bitbucket.org/morea/terraform.feature.datadog/src/master/middleware/nginx/)
 	- [php-fpm](https://bitbucket.org/morea/terraform.feature.datadog/src/master/middleware/php-fpm/)
+	- [redis](https://bitbucket.org/morea/terraform.feature.datadog/src/master/middleware/redis/)
 - [system](https://bitbucket.org/morea/terraform.feature.datadog/src/master/system/)
 	- [generic](https://bitbucket.org/morea/terraform.feature.datadog/src/master/system/generic/)

middleware/redis/README.md

@@ -0,0 +1,72 @@
+# MIDDLEWARE REDIS DataDog monitors
+
+## How to use this module
+
+```
+module "datadog-monitors-middleware-redis" {
+  source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//middleware/redis?ref={revision}"
+
+  environment = "${var.environment}"
+  message     = "${module.datadog-message-alerting.alerting-message}"
+}
+
+```
+
+## Purpose
+
+Creates DataDog monitors with the following checks:
+
+- Redis too many evictedkeys
+- Redis too many expired keys
+- Redis too many blocked clients
+- Redis keyspace seems full
+
+## Inputs
+
+| Name | Description | Type | Default | Required |
+|------|-------------|:----:|:-----:|:-----:|
+| blocked_clients_message | Custom message for Redis Blocked clients monitor | string | `` | no |
+| blocked_clients_silenced | Groups to mute for Redis Blocked clients monitor | map | `<map>` | no |
+| blocked_clients_threshold_critical | Blocked clients rate (critical threshold) | string | `90` | no |
+| blocked_clients_threshold_warning | Blocked clients rate (warning threshold) | string | `70` | no |
+| blocked_clients_time_aggregator | Monitor aggregator for Redis Blocked clients [available values: min, max or avg] | string | `min` | no |
+| blocked_clients_timeframe | Monitor timeframe for Redis Blocked clients [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
+| delay | Delay in seconds for the metric evaluation | string | `900` | no |
+| environment | Architecture environment | string | - | yes |
+| evictedkeys_limit_message | Custom message for Redis evicted keys monitor | string | `` | no |
+| evictedkeys_limit_silenced | Groups to mute for Redis evicted keys monitor | map | `<map>` | no |
+| evictedkeys_limit_threshold_critical | Evicted keys limit (critical threshold) | string | `100` | no |
+| evictedkeys_limit_threshold_warning | Evicted keys limit (warning threshold) | string | `0` | no |
+| evictedkeys_limit_time_aggregator | Monitor aggregator for Redis evicted keys [available values: min, max or avg] | string | `avg` | no |
+| evictedkeys_limit_timeframe | Monitor timeframe for Redis evicted keys [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
+| expirations_limit_message | Custom message for Redis keys expirations monitor | string | `` | no |
+| expirations_limit_threshold_critical | Expirations percent (critical threshold) | string | `80` | no |
+| expirations_limit_threshold_warning | Expirations percent (warning threshold) | string | `60` | no |
+| expirations_limit_time_aggregator | Monitor aggregator for Redis keys expirations [available values: min, max or avg] | string | `min` | no |
+| expirations_limit_timeframe | Monitor timeframe for Redis keys expirations [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
+| expirations_silenced | Groups to mute for Redis keys expirations monitor | map | `<map>` | 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 |
+| keyspace_message | Custom message for Redis keyspace monitor | string | `` | no |
+| keyspace_silenced | Groups to mute for Redis keyspace monitor | map | `<map>` | no |
+| keyspace_threshold_critical | Keyspace changement (critical threshold) | string | `90` | no |
+| keyspace_threshold_warning | Keyspace changement (warning threshold) | string | `70` | no |
+| keyspace_time_aggregator | Monitor aggregator for Redis keyspace [available values: min, max or avg] | string | `min` | no |
+| keyspace_timeframe | Monitor timeframe for Redis keyspace [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
+| message | Message sent when a Redis monitor is triggered | string | - | yes |
+| redis_silenced | Groups to mute for Redis monitors | map | `<map>` | no |
+
+## Outputs
+
+| Name | Description |
+|------|-------------|
+| redis_blocked_clients_id | id for monitor redis_blocked_clients |
+| redis_evictedkeys_id | id for monitor redis_evictedkeys |
+| redis_expirations_id | id for monitor redis_expirations |
+| redis_keyspace_id | id for monitor redis_keyspace |
+
+## Related documentation
+
+[Datadog blog: How to monitor Redis](https://www.datadoghq.com/blog/how-to-monitor-redis-performance-metrics/)
+
+[Datadog Redis integration doc](https://docs.datadoghq.com/integrations/redisdb/)

middleware/redis/inputs.tf

@@ -0,0 +1,168 @@
+# Global Terraform
+variable "environment" {
+  description = "Architecture environment"
+  type        = "string"
+}
+
+# Global DataDog
+variable "message" {
+  description = "Message sent when a Redis monitor is triggered"
+}
+
+variable "delay" {
+  description = "Delay in seconds for the metric evaluation"
+  default     = 900
+}
+
+variable "filter_tags_use_defaults" {
+  description = "Use default filter tags convention"
+  default     = "true"
+}
+
+variable "filter_tags_custom" {
+  description = "Tags used for custom filtering when filter_tags_use_defaults is false"
+  default     = "*"
+}
+
+# Redis specific variables
+variable "redis_silenced" {
+  description = "Groups to mute for Redis monitors"
+  type        = "map"
+  default     = {}
+}
+
+variable "evictedkeys_limit_silenced" {
+  description = "Groups to mute for Redis evicted keys monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "evictedkeys_limit_message" {
+  description = "Custom message for Redis evicted keys monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "evictedkeys_limit_time_aggregator" {
+  description = "Monitor aggregator for Redis evicted keys [available values: min, max or avg]"
+  type        = "string"
+  default     = "avg"
+}
+
+variable "evictedkeys_limit_timeframe" {
+  description = "Monitor timeframe for Redis evicted keys [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
+variable "evictedkeys_limit_threshold_warning" {
+  description = "Evicted keys limit (warning threshold)"
+  default     = 0
+}
+
+variable "evictedkeys_limit_threshold_critical" {
+  description = "Evicted keys limit (critical threshold)"
+  default     = 100
+}
+
+variable "expirations_silenced" {
+  description = "Groups to mute for Redis keys expirations monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "expirations_limit_message" {
+  description = "Custom message for Redis keys expirations monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "expirations_limit_time_aggregator" {
+  description = "Monitor aggregator for Redis keys expirations [available values: min, max or avg]"
+  type        = "string"
+  default     = "min"
+}
+
+variable "expirations_limit_timeframe" {
+  description = "Monitor timeframe for Redis keys expirations [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
+variable "expirations_limit_threshold_critical" {
+  description = "Expirations percent (critical threshold)"
+  default     = 80
+}
+
+variable "expirations_limit_threshold_warning" {
+  description = "Expirations percent (warning threshold)"
+  default     = 60
+}
+
+variable "blocked_clients_silenced" {
+  description = "Groups to mute for Redis Blocked clients monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "blocked_clients_message" {
+  description = "Custom message for Redis Blocked clients monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "blocked_clients_time_aggregator" {
+  description = "Monitor aggregator for Redis Blocked clients [available values: min, max or avg]"
+  type        = "string"
+  default     = "min"
+}
+
+variable "blocked_clients_timeframe" {
+  description = "Monitor timeframe for Redis Blocked clients [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
+variable "blocked_clients_threshold_critical" {
+  description = "Blocked clients rate (critical threshold)"
+  default     = 90
+}
+
+variable "blocked_clients_threshold_warning" {
+  description = "Blocked clients rate (warning threshold)"
+  default     = 70
+}
+
+variable "keyspace_silenced" {
+  description = "Groups to mute for Redis keyspace monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "keyspace_message" {
+  description = "Custom message for Redis keyspace monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "keyspace_time_aggregator" {
+  description = "Monitor aggregator for Redis keyspace [available values: min, max or avg]"
+  type        = "string"
+  default     = "min"
+}
+
+variable "keyspace_timeframe" {
+  description = "Monitor timeframe for Redis keyspace [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}
+
+variable "keyspace_threshold_critical" {
+  description = "Keyspace changement (critical threshold)"
+  default     = 90
+}
+
+variable "keyspace_threshold_warning" {
+  description = "Keyspace changement (warning threshold)"
+  default     = 70
+}

middleware/redis/monitors-redis.tf

@@ -0,0 +1,135 @@
+data "template_file" "filter" {
+  template = "$${filter}"
+
+  vars {
+    filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_redis:enabled,db_env:%s", var.environment) : "${var.filter_tags_custom}"}"
+  }
+}
+
+resource "datadog_monitor" "redis_evictedkeys" {
+  name    = "[${var.environment}] Redis too many evictedkeys {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
+  message = "${coalesce(var.evictedkeys_limit_message, var.message)}"
+
+  query = <<EOL
+    ${var.evictedkeys_limit_time_aggregator}(${var.evictedkeys_limit_timeframe}): (
+      avg:redis.keys.evicted{${data.template_file.filter.rendered}} by {name,host}
+     ) > ${var.evictedkeys_limit_threshold_critical}
+EOL
+
+  type = "metric alert"
+
+  thresholds {
+    warning  = "${var.evictedkeys_limit_threshold_warning}"
+    critical = "${var.evictedkeys_limit_threshold_critical}"
+  }
+
+  silenced = "${var.evictedkeys_limit_silenced}"
+
+  notify_no_data      = false
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = false
+  new_host_delay      = "${var.delay}"
+
+  tags = ["env:${var.environment}", "resource:redis"]
+}
+
+resource "datadog_monitor" "redis_expirations" {
+  name    = "[${var.environment}] Redis too many expired keys {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
+  message = "${coalesce(var.expirations_limit_message, var.message)}"
+
+  query = <<EOL
+    ${var.expirations_limit_time_aggregator}(${var.expirations_limit_timeframe}): (
+      avg:redis.expires.percent{${data.template_file.filter.rendered}} by {name,host}
+     ) > ${var.expirations_limit_threshold_critical}
+EOL
+
+  type = "metric alert"
+
+  thresholds {
+    warning  = "${var.expirations_limit_threshold_warning}"
+    critical = "${var.expirations_limit_threshold_critical}"
+  }
+
+  silenced = "${var.expirations_silenced}"
+
+  notify_no_data      = false
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = false
+  new_host_delay      = "${var.delay}"
+
+  tags = ["env:${var.environment}", "resource:redis"]
+}
+
+resource "datadog_monitor" "redis_blocked_clients" {
+  name    = "[${var.environment}] Redis too many blocked clients {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
+  message = "${coalesce(var.blocked_clients_message, var.message)}"
+
+  query = <<EOL
+    ${var.blocked_clients_time_aggregator}(${var.blocked_clients_timeframe}): (
+      avg:redis.clients.blocked{${data.template_file.filter.rendered}} by {name,host}
+     ) > ${var.blocked_clients_threshold_critical}
+EOL
+
+  type = "metric alert"
+
+  thresholds {
+    warning  = "${var.blocked_clients_threshold_warning}"
+    critical = "${var.blocked_clients_threshold_critical}"
+  }
+
+  silenced = "${var.blocked_clients_silenced}"
+
+  notify_no_data      = false
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = false
+  new_host_delay      = "${var.delay}"
+
+  tags = ["env:${var.environment}", "resource:redis"]
+}
+
+resource "datadog_monitor" "redis_keyspace" {
+  name    = "[${var.environment}] Redis keyspace seems full {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
+  message = "${coalesce(var.keyspace_message, var.message)}"
+
+  query = <<EOL
+    ${var.keyspace_time_aggregator}(${var.keyspace_timeframe}): (
+      avg:redis.key.length{${data.template_file.filter.rendered}} by {name,host}
+     ) > ${var.keyspace_threshold_critical}
+EOL
+
+  type = "metric alert"
+
+  thresholds {
+    warning  = "${var.keyspace_threshold_warning}"
+    critical = "${var.keyspace_threshold_critical}"
+  }
+
+  silenced = "${var.keyspace_silenced}"
+
+  notify_no_data      = false
+  evaluation_delay    = "${var.delay}"
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = false
+  new_host_delay      = "${var.delay}"
+
+  tags = ["env:${var.environment}", "resource:redis"]
+}

middleware/redis/outputs.tf

@@ -0,0 +1,19 @@
+output "redis_evictedkeys_id" {
+  description = "id for monitor redis_evictedkeys"
+  value       = "${datadog_monitor.redis_evictedkeys.id}"
+}
+
+output "redis_expirations_id" {
+  description = "id for monitor redis_expirations"
+  value       = "${datadog_monitor.redis_expirations.id}"
+}
+
+output "redis_blocked_clients_id" {
+  description = "id for monitor redis_blocked_clients"
+  value       = "${datadog_monitor.redis_blocked_clients.id}"
+}
+
+output "redis_keyspace_id" {
+  description = "id for monitor redis_keyspace"
+  value       = "${datadog_monitor.redis_keyspace.id}"
+}