MON-91 - Update inputs and add README
This commit is contained in:
parent
d9605c5d98
commit
e94ae0dcb7
33
cloud/aws/vpn/README.md
Normal file
33
cloud/aws/vpn/README.md
Normal file
@ -0,0 +1,33 @@
|
|||||||
|
AWS VPN DataDog monitors
|
||||||
|
===============================
|
||||||
|
|
||||||
|
How to use this module
|
||||||
|
----------------------
|
||||||
|
|
||||||
|
```
|
||||||
|
module "vpn" {
|
||||||
|
source = "git::ssh://git@bitbucket.org/morea/terraform.feature.datadog.git//cloud/aws/vpn?ref=MON-91-added-aws-vpn-state-check"
|
||||||
|
|
||||||
|
environment = "${var.environment}"
|
||||||
|
message = "${module.datadog-message-alerting.alerting-message}"
|
||||||
|
vpn_tunnel_address = "${var.vpn_to_monitor}"
|
||||||
|
}
|
||||||
|
```
|
||||||
|
|
||||||
|
Purpose
|
||||||
|
-------
|
||||||
|
Creates a DataDog monitors with the following checks :
|
||||||
|
* VPN status
|
||||||
|
|
||||||
|
Inputs
|
||||||
|
-------
|
||||||
|
|
||||||
|
| Name | Description | Type | Default | Required |
|
||||||
|
|------|-------------|:----:|:-----:|:-----:|
|
||||||
|
| environment | Architecture Environment | string | - | yes |
|
||||||
|
| evaluation_delay | Delay in seconds for the metric evaluation | string | `600` | no |
|
||||||
|
| message | Message sent when an alert is triggered | string | - | yes |
|
||||||
|
| vpn_status_critical | Both tunnels are down | string | `0` | no |
|
||||||
|
| vpn_status_ok | Both tunnels are up, everything is ok | string | `2` | no |
|
||||||
|
| vpn_status_warning | Only one tunnel is up | string | `1` | no |
|
||||||
|
| vpn_tunnel_address | List the VPN tunnels you want to monitor | list | `<list>` | no |
|
||||||
@ -1,19 +1,37 @@
|
|||||||
variable env {}
|
# Global Terraform
|
||||||
|
variable "environment" {
|
||||||
|
description = "Architecture Environment"
|
||||||
|
type = "string"
|
||||||
|
}
|
||||||
|
|
||||||
|
# Global DataDog
|
||||||
|
variable "evaluation_delay" {
|
||||||
|
description = "Delay in seconds for the metric evaluation"
|
||||||
|
default = 600
|
||||||
|
}
|
||||||
|
|
||||||
|
variable "message" {
|
||||||
|
description = "Message sent when an alert is triggered"
|
||||||
|
}
|
||||||
|
|
||||||
|
# AWS VPN Service specific
|
||||||
variable "vpn_tunnel_address" {
|
variable "vpn_tunnel_address" {
|
||||||
description = "List the VPN tunnels you want to monitor"
|
description = "List the VPN tunnels you want to monitor"
|
||||||
type = "list"
|
type = "list"
|
||||||
default = []
|
default = []
|
||||||
}
|
}
|
||||||
|
|
||||||
variable "vpn_config" {
|
variable "vpn_status_ok" {
|
||||||
description = "Ok means that both tunnels are up, warning means only one tunnel is up and critical means that both tunnels are down "
|
description = "Both tunnels are up, everything is ok"
|
||||||
type = "map"
|
default = 2
|
||||||
|
}
|
||||||
default = {
|
|
||||||
delay = 900
|
variable "vpn_status_warning" {
|
||||||
warning = 1
|
description = "Only one tunnel is up"
|
||||||
critical = 0
|
default = 1
|
||||||
ok = 2
|
}
|
||||||
}
|
|
||||||
|
variable "vpn_status_critical" {
|
||||||
|
description = "Both tunnels are down"
|
||||||
|
default = 0
|
||||||
}
|
}
|
||||||
|
|||||||
@ -1,37 +1,30 @@
|
|||||||
module "message" {
|
|
||||||
source = "../../common/alerting-message"
|
|
||||||
oncall_24x7 = "${var.hno_escalation_group}"
|
|
||||||
oncall_office_hours = "${var.ho_escalation_group}"
|
|
||||||
}
|
|
||||||
|
|
||||||
resource "datadog_monitor" "VPN_status" {
|
resource "datadog_monitor" "VPN_status" {
|
||||||
name = "[${var.env}] VPN Down {{ tunnelipaddress }}"
|
name = "[${var.environment}] VPN Down {{ tunnelipaddress }}"
|
||||||
message = "${module.message.alerting-message}"
|
message = "${var.message}"
|
||||||
|
|
||||||
query = <<EOF
|
query = <<EOF
|
||||||
avg(last_5m): (
|
avg(last_5m): (
|
||||||
avg:aws.vpn.tunnel_state{tunnelipaddress:${var.vpn_tunnel_address[0]}} by {region}
|
avg:aws.vpn.tunnel_state{tunnelipaddress:${var.vpn_tunnel_address[0]}} by {region,name}
|
||||||
+ avg:aws.vpn.tunnel_state{tunnelipaddress:${var.vpn_tunnel_address[1]}} by {region}
|
+ avg:aws.vpn.tunnel_state{tunnelipaddress:${var.vpn_tunnel_address[1]}} by {region,name}
|
||||||
) == ${var.vpn_config["critical"]}
|
) == ${var.vpn_status_critical}
|
||||||
EOF
|
EOF
|
||||||
|
|
||||||
type = "query alert"
|
type = "metric alert"
|
||||||
|
|
||||||
thresholds {
|
thresholds {
|
||||||
ok = "${var.vpn_config["ok"]}"
|
ok = "${var.vpn_status_ok}"
|
||||||
critical = "${var.vpn_config["critical"]}"
|
warning = "${var.vpn_status_warning}"
|
||||||
warning = "${var.vpn_config["warning"]}"
|
critical = "${var.vpn_status_critical}"
|
||||||
}
|
}
|
||||||
|
|
||||||
notify_no_data = true
|
notify_no_data = true
|
||||||
renotify_interval = 0
|
renotify_interval = 0
|
||||||
evaluation_delay = "${var.vpn_config["delay"]}"
|
evaluation_delay = "${var.evaluation_delay}"
|
||||||
|
new_host_delay = "${var.evaluation_delay}"
|
||||||
notify_audit = false
|
notify_audit = false
|
||||||
timeout_h = 0
|
timeout_h = 0
|
||||||
include_tags = true
|
include_tags = true
|
||||||
locked = false
|
require_full_window = false
|
||||||
require_full_window = true
|
|
||||||
new_host_delay = "${var.vpn_config["delay"]}}"
|
|
||||||
|
|
||||||
tags = ["env: ${var.env}", "provider: aws", "resource: vpn"]
|
tags = ["env: ${var.environment}", "resource:vpn", "team:aws", "provider:aws"]
|
||||||
}
|
}
|
||||||
|
|||||||
@ -60,8 +60,6 @@ variable "rds_cpu_threshold" {
|
|||||||
}
|
}
|
||||||
|
|
||||||
variable "rds_mem_threshold" {
|
variable "rds_mem_threshold" {
|
||||||
type = "map"
|
|
||||||
|
|
||||||
default = {
|
default = {
|
||||||
warning = 20
|
warning = 20
|
||||||
critical = 10
|
critical = 10
|
||||||
@ -83,8 +81,6 @@ variable "elb_config" {
|
|||||||
}
|
}
|
||||||
|
|
||||||
variable "elb_5xx_threshold" {
|
variable "elb_5xx_threshold" {
|
||||||
type = "map"
|
|
||||||
|
|
||||||
default = {
|
default = {
|
||||||
warning = 5
|
warning = 5
|
||||||
critical = 10
|
critical = 10
|
||||||
@ -92,8 +88,6 @@ variable "elb_5xx_threshold" {
|
|||||||
}
|
}
|
||||||
|
|
||||||
variable "elb_4xx_threshold" {
|
variable "elb_4xx_threshold" {
|
||||||
type = "map"
|
|
||||||
|
|
||||||
default = {
|
default = {
|
||||||
warning = 5
|
warning = 5
|
||||||
critical = 10
|
critical = 10
|
||||||
@ -102,7 +96,6 @@ variable "elb_4xx_threshold" {
|
|||||||
|
|
||||||
variable "elb_backend_latency" {
|
variable "elb_backend_latency" {
|
||||||
description = "Average time elapsed after the request leaves the load balancer until a response is received. In seconds"
|
description = "Average time elapsed after the request leaves the load balancer until a response is received. In seconds"
|
||||||
|
|
||||||
default = {
|
default = {
|
||||||
warning = 1
|
warning = 1
|
||||||
critical = 5
|
critical = 5
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user