MON-105 add service check

This commit is contained in:
Quentin Manfroi 2018-09-18 17:05:14 +02:00
parent b2bcc58b00
commit c2dfc0ed24
5 changed files with 85 additions and 1 deletions

View File

@ -26,7 +26,7 @@ resource "datadog_monitor" "mysql_availability" {
silenced = "${var.mysql_availability_silenced}" silenced = "${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" { resource "datadog_monitor" "mysql_connection" {

View File

@ -17,6 +17,7 @@ module "datadog-monitors-database-postgresql" {
Creates DataDog monitors with the following checks: Creates DataDog monitors with the following checks:
- PostgreSQL Connections - PostgreSQL Connections
- PostgreSQL server does not respond
- PostgreSQL too many locks - PostgreSQL too many locks
## Inputs ## Inputs
@ -30,6 +31,12 @@ Creates DataDog monitors with the following checks:
| message | Message sent when an alert is triggered | string | - | yes | | message | Message sent when an alert is triggered | string | - | yes |
| new_host_delay | Delay in seconds for the metric evaluation | string | `300` | no | | 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 | | 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_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_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_message | Custom message for PostgreSQL connection monitor | string | `` | no |
@ -50,6 +57,7 @@ Creates DataDog monitors with the following checks:
| Name | Description | | Name | Description |
|------|-------------| |------|-------------|
| postgresql_availability_id | id for monitor postgresql_availability |
| postgresql_connection_too_high_id | id for monitor postgresql_connection_too_high | | postgresql_connection_too_high_id | id for monitor postgresql_connection_too_high |
| postgresql_too_many_locks_id | id for monitor postgresql_too_many_locks | | postgresql_too_many_locks_id | id for monitor postgresql_too_many_locks |

View File

@ -29,6 +29,46 @@ variable "filter_tags_custom" {
} }
# PostgreSQL specific # 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 ### ### PostgreSQL connections ###
################################## ##################################

View File

@ -1,3 +1,34 @@
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" { resource "datadog_monitor" "postgresql_connection_too_high" {
count = "${var.postgresql_connection_enabled ? 1 : 0}" count = "${var.postgresql_connection_enabled ? 1 : 0}"
name = "[${var.environment}] PostgreSQL Connections {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}" name = "[${var.environment}] PostgreSQL Connections {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"

View File

@ -1,3 +1,8 @@
output "postgresql_availability_id" {
description = "id for monitor postgresql_availability"
value = "${datadog_monitor.postgresql_availability.*.id}"
}
output "postgresql_connection_too_high_id" { output "postgresql_connection_too_high_id" {
description = "id for monitor postgresql_connection_too_high" description = "id for monitor postgresql_connection_too_high"
value = "${datadog_monitor.postgresql_connection_too_high.*.id}" value = "${datadog_monitor.postgresql_connection_too_high.*.id}"