Merged in MON-105-monitors-for-postgresql (pull request #98)

MON-105 monitors for postgresql

Approved-by: Boris Rousseau <boris.rousseau@morea.fr>
Approved-by: Quentin Manfroi <quentin.manfroi@yahoo.fr>
Approved-by: Guillaume Kérivel <guillaume.kerivel@fr.clara.net>

README.md

@@ -116,6 +116,7 @@ The `//` is very important, it's a terraform specific syntax used to separate gi
 	- [elasticsearch](https://bitbucket.org/morea/terraform.feature.datadog/src/master/database/elasticsearch/)
 	- [mongodb](https://bitbucket.org/morea/terraform.feature.datadog/src/master/database/mongodb/)
 	- [mysql](https://bitbucket.org/morea/terraform.feature.datadog/src/master/database/mysql/)
+	- [postgresql](https://bitbucket.org/morea/terraform.feature.datadog/src/master/database/postgresql/)
 	- [redis](https://bitbucket.org/morea/terraform.feature.datadog/src/master/database/redis/)
 - [middleware](https://bitbucket.org/morea/terraform.feature.datadog/src/master/middleware/)
 	- [apache](https://bitbucket.org/morea/terraform.feature.datadog/src/master/middleware/apache/)

database/mysql/monitors-mysql.tf

@@ -26,7 +26,7 @@ resource "datadog_monitor" "mysql_availability" {
 
   silenced = "${var.mysql_availability_silenced}"
 
-  tags = ["env:${var.environment}", "type:database", "provider:mysql", "resource:mysql", "team:claranet", "created-by:terraform", "${var.mysql_connection_extra_tags}"]
+  tags = ["env:${var.environment}", "type:database", "provider:mysql", "resource:mysql", "team:claranet", "created-by:terraform", "${var.mysql_availability_extra_tags}"]
 }
 
 resource "datadog_monitor" "mysql_connection" {

database/postgresql/README.md

@@ -0,0 +1,67 @@
+# DATABASE POSTGRESQL DataDog monitors
+
+## How to use this module
+
+```
+module "datadog-monitors-database-postgresql" {
+  source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//database/postgresql?ref={revision}"
+
+  environment = "${var.environment}"
+  message     = "${module.datadog-message-alerting.alerting-message}"
+}
+
+```
+
+## Purpose
+
+Creates DataDog monitors with the following checks:
+
+- PostgreSQL Connections
+- PostgreSQL server does not respond
+- PostgreSQL too many locks
+
+## Inputs
+
+| Name | Description | Type | Default | Required |
+|------|-------------|:----:|:-----:|:-----:|
+| environment | Environment | string | - | yes |
+| evaluation_delay | Delay in seconds for the metric evaluation | string | `15` | 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 |
+| message | Message sent when an alert is triggered | string | - | yes |
+| new_host_delay | Delay in seconds for the metric evaluation | string | `300` | no |
+| posgresql_lock_silenced | Groups to mute for PostgreSQL lock monitor | map | `<map>` | no |
+| postgresql_availability_enabled | Flag to enable PostgreSQL availability monitor | string | `true` | no |
+| postgresql_availability_extra_tags | Extra tags for PostgreSQL availability monitor | list | `<list>` | no |
+| postgresql_availability_message | Custom message for PostgreSQL availability monitor | string | `` | no |
+| postgresql_availability_no_data_timeframe | PostgreSQL availability monitor no data timeframe | string | `10` | no |
+| postgresql_availability_silenced | Groups to mute for PostgreSQL availability monitor | map | `<map>` | no |
+| postgresql_availability_threshold_warning | PostgreSQL availability monitor (warning threshold) | string | `3` | no |
+| postgresql_connection_enabled | Flag to enable PostgreSQL connection monitor | string | `true` | no |
+| postgresql_connection_extra_tags | Extra tags for PostgreSQL connection connects monitor | list | `<list>` | no |
+| postgresql_connection_message | Custom message for PostgreSQL connection monitor | string | `` | no |
+| postgresql_connection_silenced | Groups to mute for PostgreSQL connection monitor | map | `<map>` | no |
+| postgresql_connection_threshold_critical | Maximum critical acceptable percent of connections | string | `80` | no |
+| postgresql_connection_threshold_warning | Maximum warning acceptable percent of connections | string | `70` | no |
+| postgresql_connection_time_aggregator | Monitor time aggregator for PostgreSQL connection monitor [available values: min, max or avg] | string | `avg` | no |
+| postgresql_connection_timeframe | Monitor timeframe for PostgreSQL connection monitor [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_15m` | no |
+| postgresql_lock_enabled | Flag to enable PostgreSQL lock monitor | string | `true` | no |
+| postgresql_lock_extra_tags | Extra tags for PostgreSQL lock connects monitor | list | `<list>` | no |
+| postgresql_lock_message | Custom message for PostgreSQL lock monitor | string | `` | no |
+| postgresql_lock_threshold_critical | Maximum critical acceptable number of locks | string | `99` | no |
+| postgresql_lock_threshold_warning | Maximum warning acceptable number of locks | string | `70` | no |
+| postgresql_lock_time_aggregator | Monitor time aggregator for PostgreSQL lock monitor [available values: min, max or avg] | string | `min` | no |
+| postgresql_lock_timeframe | Monitor timeframe for PostgreSQL lock monitor [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `last_5m` | no |
+
+## Outputs
+
+| Name | Description |
+|------|-------------|
+| postgresql_availability_id | id for monitor postgresql_availability |
+| postgresql_connection_too_high_id | id for monitor postgresql_connection_too_high |
+| postgresql_too_many_locks_id | id for monitor postgresql_too_many_locks |
+
+Related documentation
+---------------------
+
+DataDog documentation: [https://docs.datadoghq.com/integrations/postgres/](https://docs.datadoghq.com/integrations/postgres/)

database/postgresql/inputs.tf

@@ -0,0 +1,170 @@
+variable "environment" {
+  description = "Environment"
+  type        = "string"
+}
+
+# Global DataDog
+variable "evaluation_delay" {
+  description = "Delay in seconds for the metric evaluation"
+  default     = 15
+}
+
+variable "new_host_delay" {
+  description = "Delay in seconds for the metric evaluation"
+  default     = 300
+}
+
+variable "message" {
+  description = "Message sent when an alert is triggered"
+}
+
+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     = "*"
+}
+
+# PostgreSQL specific
+##################################
+###   PostgreSQL availability  ###
+##################################
+
+variable "postgresql_availability_silenced" {
+  description = "Groups to mute for PostgreSQL availability monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "postgresql_availability_enabled" {
+  description = "Flag to enable PostgreSQL availability monitor"
+  type        = "string"
+  default     = "true"
+}
+
+variable "postgresql_availability_extra_tags" {
+  description = "Extra tags for PostgreSQL availability monitor"
+  type        = "list"
+  default     = []
+}
+
+variable "postgresql_availability_message" {
+  description = "Custom message for PostgreSQL availability monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "postgresql_availability_threshold_warning" {
+  description = "PostgreSQL availability monitor (warning threshold)"
+  type        = "string"
+  default     = 3
+}
+
+variable "postgresql_availability_no_data_timeframe" {
+  description = "PostgreSQL availability monitor no data timeframe"
+  type        = "string"
+  default     = 10
+}
+
+##################################
+###   PostgreSQL connections   ###
+##################################
+
+variable "postgresql_connection_threshold_critical" {
+  default     = 80
+  description = "Maximum critical acceptable percent of connections"
+}
+
+variable "postgresql_connection_threshold_warning" {
+  default     = 70
+  description = "Maximum warning acceptable percent of connections"
+}
+
+variable "postgresql_connection_silenced" {
+  description = "Groups to mute for PostgreSQL connection monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "postgresql_connection_enabled" {
+  description = "Flag to enable PostgreSQL connection monitor"
+  type        = "string"
+  default     = "true"
+}
+
+variable "postgresql_connection_extra_tags" {
+  description = "Extra tags for PostgreSQL connection connects monitor"
+  type        = "list"
+  default     = []
+}
+
+variable "postgresql_connection_message" {
+  description = "Custom message for PostgreSQL connection monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "postgresql_connection_time_aggregator" {
+  description = "Monitor time aggregator for PostgreSQL connection monitor [available values: min, max or avg]"
+  type        = "string"
+  default     = "avg"
+}
+
+variable "postgresql_connection_timeframe" {
+  description = "Monitor timeframe for PostgreSQL connection monitor [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_15m"
+}
+
+############################
+###   PostgreSQL locks   ###
+############################
+
+variable "postgresql_lock_threshold_critical" {
+  default     = 99
+  description = "Maximum critical acceptable number of locks"
+}
+
+variable "postgresql_lock_threshold_warning" {
+  default     = 70
+  description = "Maximum warning acceptable number of locks"
+}
+
+variable "posgresql_lock_silenced" {
+  description = "Groups to mute for PostgreSQL lock monitor"
+  type        = "map"
+  default     = {}
+}
+
+variable "postgresql_lock_enabled" {
+  description = "Flag to enable PostgreSQL lock monitor"
+  type        = "string"
+  default     = "true"
+}
+
+variable "postgresql_lock_extra_tags" {
+  description = "Extra tags for PostgreSQL lock connects monitor"
+  type        = "list"
+  default     = []
+}
+
+variable "postgresql_lock_message" {
+  description = "Custom message for PostgreSQL lock monitor"
+  type        = "string"
+  default     = ""
+}
+
+variable "postgresql_lock_time_aggregator" {
+  description = "Monitor time aggregator for PostgreSQL lock monitor [available values: min, max or avg]"
+  type        = "string"
+  default     = "min"
+}
+
+variable "postgresql_lock_timeframe" {
+  description = "Monitor timeframe for PostgreSQL lock monitor [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`]"
+  type        = "string"
+  default     = "last_5m"
+}

database/postgresql/modules.tf

@@ -0,0 +1,8 @@
+module "filter-tags" {
+  source = "../../common/filter-tags"
+
+  environment              = "${var.environment}"
+  resource                 = "postgres"
+  filter_tags_use_defaults = "${var.filter_tags_use_defaults}"
+  filter_tags_custom       = "${var.filter_tags_custom}"
+}

database/postgresql/monitors-postgresql.tf

@@ -0,0 +1,92 @@
+resource "datadog_monitor" "postgresql_availability" {
+  count   = "${var.postgresql_availability_enabled ? 1 : 0}"
+  name    = "[${var.environment}] PostgreSQL server does not respond"
+  message = "${coalesce(var.postgresql_availability_message, var.message)}"
+
+  type = "service check"
+
+  query = <<EOF
+    "postgres.can_connect".over${module.filter-tags.service_check}.by("port","server").last(6).count_by_status()
+  EOF
+
+  thresholds = {
+    warning  = "${var.postgresql_availability_threshold_warning}"
+    critical = 5
+  }
+
+  notify_no_data      = true
+  no_data_timeframe   = "${var.postgresql_availability_no_data_timeframe}"
+  new_host_delay      = "${var.new_host_delay}"
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+
+  silenced = "${var.postgresql_availability_silenced}"
+
+  tags = ["env:${var.environment}", "type:database", "provider:postgres", "resource:postgresql", "team:claranet", "created-by:terraform", "${var.postgresql_availability_extra_tags}"]
+}
+
+resource "datadog_monitor" "postgresql_connection_too_high" {
+  count   = "${var.postgresql_connection_enabled ? 1 : 0}"
+  name    = "[${var.environment}] PostgreSQL Connections {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
+  message = "${coalesce(var.postgresql_connection_message, var.message)}"
+  type    = "metric alert"
+
+  query = <<EOF
+    ${var.postgresql_connection_time_aggregator}(${var.postgresql_connection_timeframe}):
+      avg:postgresql.percent_usage_connections${module.filter-tags.query_alert} by {server}
+    * 100 > ${var.postgresql_connection_threshold_critical}
+  EOF
+
+  evaluation_delay = "${var.evaluation_delay}"
+  new_host_delay   = "${var.new_host_delay}"
+
+  thresholds {
+    warning  = "${var.postgresql_connection_threshold_warning}"
+    critical = "${var.postgresql_connection_threshold_critical}"
+  }
+
+  notify_no_data      = false
+  renotify_interval   = 0
+  require_full_window = true
+  timeout_h           = 0
+  include_tags        = true
+
+  silenced = "${var.postgresql_connection_silenced}"
+
+  tags = ["env:${var.environment}", "type:database", "provider:postgres", "resource:postgresql", "team:claranet", "created-by:terraform", "${var.postgresql_connection_extra_tags}"]
+}
+
+resource "datadog_monitor" "postgresql_too_many_locks" {
+  count   = "${var.postgresql_lock_enabled ? 1 : 0}"
+  name    = "[${var.environment}] PostgreSQL too many locks {{#is_alert}}{{{comparator}}} {{threshold}} ({{value}}){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}} ({{value}}){{/is_warning}}"
+  message = "${coalesce(var.postgresql_lock_message, var.message)}"
+  type    = "metric alert"
+
+  query = <<EOF
+    ${var.postgresql_lock_time_aggregator}(${var.postgresql_lock_timeframe}):
+      default(avg:postgresql.locks${module.filter-tags.query_alert} by {server}, 0)
+    > ${var.postgresql_lock_threshold_critical}
+  EOF
+
+  evaluation_delay = "${var.evaluation_delay}"
+  new_host_delay   = "${var.new_host_delay}"
+
+  thresholds {
+    warning  = "${var.postgresql_lock_threshold_warning}"
+    critical = "${var.postgresql_lock_threshold_critical}"
+  }
+
+  notify_no_data      = false
+  renotify_interval   = 0
+  require_full_window = true
+  timeout_h           = 0
+  include_tags        = true
+
+  silenced = "${var.posgresql_lock_silenced}"
+
+  tags = ["env:${var.environment}", "type:database", "provider:postgres", "resource:postgresql", "team:claranet", "created-by:terraform", "${var.postgresql_lock_extra_tags}"]
+}

database/postgresql/outputs.tf

@@ -0,0 +1,14 @@
+output "postgresql_availability_id" {
+  description = "id for monitor postgresql_availability"
+  value       = "${datadog_monitor.postgresql_availability.*.id}"
+}
+
+output "postgresql_connection_too_high_id" {
+  description = "id for monitor postgresql_connection_too_high"
+  value       = "${datadog_monitor.postgresql_connection_too_high.*.id}"
+}
+
+output "postgresql_too_many_locks_id" {
+  description = "id for monitor postgresql_too_many_locks"
+  value       = "${datadog_monitor.postgresql_too_many_locks.*.id}"
+}