Merge branch 'MON-547-MySQL_replication_monitor' into 'master'

MON-547 Add MySQL replication monitors

See merge request claranet/pt-monitoring/projects/datadog/terraform/monitors!151

database/mysql/README.md

@@ -22,6 +22,8 @@ Creates DataDog monitors with the following checks:
 - Mysql Innodb buffer pool efficiency
 - Mysql Innodb buffer pool utilization
 - Mysql queries changed abnormally
+- Mysql replication lag (disabled by default)
+- Mysql replication status changed abnormally (disabled by default)
 - Mysql server does not respond
 - Mysql Slow queries
 - Mysql threads changed abnormally
@@ -82,6 +84,18 @@ Creates DataDog monitors with the following checks:
 | mysql\_questions\_threshold\_critical | Maximum critical acceptable number of queries | string | `"1"` | no |
 | mysql\_questions\_time\_aggregator | Monitor time aggregator for MySQL queries monitor [available values: min, max or avg] | string | `"avg"` | no |
 | mysql\_questions\_timeframe | Monitor timeframe for MySQL queries monitor [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `"last_4h"` | no |
+| mysql\_replication\_lag\_enabled | Flag to enable mysql replication lag monitor | string | `"false"` | no |
+| mysql\_replication\_lag\_extra\_tags | Extra tags for MySQL replication lag monitor | list(string) | `[]` | no |
+| mysql\_replication\_lag\_message | Custom message for MySQL replication lag monitor | string | `""` | no |
+| mysql\_replication\_lag\_threshold\_critical | Maximum critical acceptable seconds of replication lag | string | `"200"` | no |
+| mysql\_replication\_lag\_threshold\_warning | Maximum warning acceptable seconds of replication lag | string | `"100"` | no |
+| mysql\_replication\_lag\_time\_aggregator | Monitor time aggregator for MySQL replication lag monitor [available values: min, max or avg] | string | `"min"` | no |
+| mysql\_replication\_lag\_timeframe | Monitor timeframe for MySQL replication lag monitor [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `"last_15m"` | no |
+| mysql\_replication\_status\_enabled | Flag to enable mysql replication status monitor | string | `"false"` | no |
+| mysql\_replication\_status\_extra\_tags | Extra tags for MySQL replication status monitor | list(string) | `[]` | no |
+| mysql\_replication\_status\_message | Custom message for MySQL replication status monitor | string | `""` | no |
+| mysql\_replication\_status\_time\_aggregator | Monitor time aggregator for MySQL replication status monitor [available values: min, max or avg] | string | `"min"` | no |
+| mysql\_replication\_status\_timeframe | Monitor timeframe for MySQL replication status monitor [available values: `last_#m` (1, 5, 10, 15, or 30), `last_#h` (1, 2, or 4), or `last_1d`] | string | `"last_5m"` | no |
 | mysql\_slow\_enabled | Flag to enable MySQL slow queries monitor | string | `"true"` | no |
 | mysql\_slow\_extra\_tags | Extra tags for MySQL slow queries monitor | list(string) | `[]` | no |
 | mysql\_slow\_message | Custom message for MySQL slow queries monitor | string | `""` | no |
@@ -116,6 +130,8 @@ Creates DataDog monitors with the following checks:
 | mysql\_pool\_efficiency\_id | id for monitor mysql_pool_efficiency |
 | mysql\_pool\_utilization\_id | id for monitor mysql_pool_utilization |
 | mysql\_questions\_anomaly\_id | id for monitor mysql_questions_anomaly |
+| mysql\_replication\_lag\_id | id for monitor mysql_replication_lag |
+| mysql\_replication\_status\_id | id for monitor mysql_replication_status |
 | mysql\_slow\_id | id for monitor mysql_slow |
 | mysql\_threads\_anomaly\_id | id for monitor mysql_threads_anomaly |
 

database/mysql/inputs.tf

@@ -460,3 +460,80 @@ variable "mysql_questions_timeframe" {
   default     = "last_4h"
 }
 
+#################################
+###   MySQL replication lag  ###
+#################################
+
+variable "mysql_replication_lag_enabled" {
+  description = "Flag to enable mysql replication lag monitor"
+  type        = string
+  default     = "false"
+}
+
+variable "mysql_replication_lag_extra_tags" {
+  description = "Extra tags for MySQL replication lag monitor"
+  type        = list(string)
+  default     = []
+}
+
+variable "mysql_replication_lag_message" {
+  description = "Custom message for MySQL replication lag monitor"
+  type        = string
+  default     = ""
+}
+
+variable "mysql_replication_lag_threshold_warning" {
+  default     = 100
+  description = "Maximum warning acceptable seconds of replication lag"
+}
+
+variable "mysql_replication_lag_threshold_critical" {
+  default     = 200
+  description = "Maximum critical acceptable seconds of replication lag"
+}
+
+variable "mysql_replication_lag_time_aggregator" {
+  description = "Monitor time aggregator for MySQL replication lag monitor [available values: min, max or avg]"
+  type        = string
+  default     = "min"
+}
+
+variable "mysql_replication_lag_timeframe" {
+  description = "Monitor timeframe for MySQL replication lag 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"
+}
+
+###################################
+###   MySQL replication status  ###
+###################################
+
+variable "mysql_replication_status_enabled" {
+  description = "Flag to enable mysql replication status monitor"
+  type        = string
+  default     = "false"
+}
+
+variable "mysql_replication_status_extra_tags" {
+  description = "Extra tags for MySQL replication status monitor"
+  type        = list(string)
+  default     = []
+}
+
+variable "mysql_replication_status_message" {
+  description = "Custom message for MySQL replication status monitor"
+  type        = string
+  default     = ""
+}
+
+variable "mysql_replication_status_time_aggregator" {
+  description = "Monitor time aggregator for MySQL replication status monitor [available values: min, max or avg]"
+  type        = string
+  default     = "min"
+}
+
+variable "mysql_replication_status_timeframe" {
+  description = "Monitor timeframe for MySQL replication status 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/mysql/monitors-mysql.tf

@@ -288,3 +288,62 @@ EOQ
   }
 }
 
+resource "datadog_monitor" "mysql_replication_lag" {
+  count   = var.mysql_replication_lag_enabled == "true" ? 1 : 0
+  name    = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Mysql replication lag {{#is_alert}}{{{comparator}}} {{threshold}}s ({{value}}s){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}s ({{value}}s){{/is_warning}}"
+  message = coalesce(var.mysql_replication_lag_message, var.message)
+  type    = "query alert"
+
+  query = <<EOQ
+    ${var.mysql_replication_lag_time_aggregator}(${var.mysql_replication_lag_timeframe}):avg:mysql.replication.seconds_behind_master${module.filter-tags.query_alert} by {server} > ${var.mysql_replication_lag_threshold_critical}
+EOQ
+
+  thresholds = {
+    warning  = var.mysql_replication_lag_threshold_warning
+    critical = var.mysql_replication_lag_threshold_critical
+  }
+
+  evaluation_delay    = var.evaluation_delay
+  new_host_delay      = var.new_host_delay
+  notify_no_data      = false
+  renotify_interval   = 0
+  require_full_window = false
+  timeout_h           = 0
+  include_tags        = true
+
+  tags = concat(["env:${var.environment}", "type:database", "provider:mysql", "resource:mysql", "team:claranet", "created-by:terraform"], var.mysql_replication_lag_extra_tags)
+
+  lifecycle {
+    ignore_changes = [silenced]
+  }
+}
+
+resource "datadog_monitor" "mysql_replication_status" {
+  count   = var.mysql_replication_status_enabled == "true" ? 1 : 0
+  name    = "${var.prefix_slug == "" ? "" : "[${var.prefix_slug}]"}[${var.environment}] Mysql replication status changed abnormally"
+  message = coalesce(var.mysql_replication_status_message, var.message)
+  type    = "metric alert"
+
+  query = <<EOQ
+    ${var.mysql_replication_status_time_aggregator}(${var.mysql_replication_status_timeframe}):avg:mysql.replication.slave_running${module.filter-tags.query_alert} by {server} < 1
+EOQ
+
+  thresholds = {
+    critical = 1
+  }
+
+  new_host_delay      = var.new_host_delay
+  notify_no_data      = false
+  renotify_interval   = 0
+  notify_audit        = false
+  timeout_h           = 0
+  include_tags        = true
+  locked              = false
+  require_full_window = true
+
+  tags = concat(["env:${var.environment}", "type:database", "provider:mysql", "resource:mysql", "team:claranet", "created-by:terraform"], var.mysql_replication_status_extra_tags)
+
+  lifecycle {
+    ignore_changes = [silenced]
+  }
+}

database/mysql/outputs.tf

@@ -28,6 +28,16 @@ output "mysql_questions_anomaly_id" {
   value       = datadog_monitor.mysql_questions_anomaly.*.id
 }
 
+output "mysql_replication_lag_id" {
+  description = "id for monitor mysql_replication_lag"
+  value       = datadog_monitor.mysql_replication_lag.*.id
+}
+
+output "mysql_replication_status_id" {
+  description = "id for monitor mysql_replication_status"
+  value       = datadog_monitor.mysql_replication_status.*.id
+}
+
 output "mysql_slow_id" {
   description = "id for monitor mysql_slow"
   value       = datadog_monitor.mysql_slow.*.id