xpk/terraform.aws-baseline-infra
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

NEW: external data source for identifying EC2 OS platform

Browse Source
This commit is contained in:
xpk 2022-11-10 14:04:44 +08:00
parent a348848966
commit d5f2af07c5
Signed by: xpk
GPG Key ID: CD4FF6793F09AB86
3 changed files with 85 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
modules/ManagementGovernance/Monitoring.EC2/get-os-platform.sh Executable file
View File

@ -0,0 +1,11 @@
#!/bin/bash
eval "$(jq -r '@sh "id=\(.input)"')"
EC2OS=$(aws ec2 describe-instances --instance-ids $id | jq -r '.Reservations[].Instances[].PlatformDetails')
if [ $EC2OS == "Windows" ]; then
echo '{"os": "Windows"}'
else
echo '{"os": "Linux"}'
fi

71
modules/ManagementGovernance/Monitoring.EC2/main.tf
View File

@ -73,7 +73,17 @@ data "aws_instance" "ec2-instance" {
instance_id = var.ec2-instance-id
}
# get instance OS
data "external" "ec2-os" {
program = ["bash", "${path.module}/get-os-platform.sh"]
query = {
input = var.ec2-instance-id
}
}
# Linux specific checks
resource "aws_cloudwatch_metric_alarm" "ec2-mem_free" {
count = data.external.ec2-os.result.os == "Linux" ? 1 : 0
alarm_name = "${var.cw-alarm-prefix}:EC2:mem_free:${var.ec2-instance-id}"
comparison_operator = "LessThanThreshold"
evaluation_periods = "2"
@ -99,6 +109,7 @@ resource "aws_cloudwatch_metric_alarm" "ec2-mem_free" {
}
resource "aws_cloudwatch_metric_alarm" "ec2-swap_free" {
count = data.external.ec2-os.result.os == "Linux" ? 1 : 0
# zero is fine as most ec2 instances are deployed without any swap
alarm_name = "${var.cw-alarm-prefix}:EC2:swap_free:${var.ec2-instance-id}"
comparison_operator = "LessThanThreshold"
@ -146,6 +157,7 @@ data "external" "disk-device" {
}
resource "aws_cloudwatch_metric_alarm" "ec2-disk_free" {
count = data.external.ec2-os.result.os == "Linux" ? 1 : 0
alarm_name = "${var.cw-alarm-prefix}:EC2:disk_free:${var.ec2-instance-id}"
comparison_operator = "LessThanThreshold"
evaluation_periods = "2"
@ -175,6 +187,7 @@ resource "aws_cloudwatch_metric_alarm" "ec2-disk_free" {
resource "aws_cloudwatch_metric_alarm" "ec2-disk_inodes_free" {
count = data.external.ec2-os.result.os == "Linux" ? 1 : 0
alarm_name = "${var.cw-alarm-prefix}:EC2:disk_inodes_free:${var.ec2-instance-id}"
comparison_operator = "LessThanThreshold"
evaluation_periods = "2"
@ -203,6 +216,7 @@ resource "aws_cloudwatch_metric_alarm" "ec2-disk_inodes_free" {
}
resource "aws_cloudwatch_metric_alarm" "ec2-processes_total" {
count = data.external.ec2-os.result.os == "Linux" ? 1 : 0
alarm_name = "${var.cw-alarm-prefix}:EC2:processes_total:${var.ec2-instance-id}"
comparison_operator = "GreaterThanThreshold"
evaluation_periods = "2"
@ -225,4 +239,61 @@ resource "aws_cloudwatch_metric_alarm" "ec2-processes_total" {
lifecycle {
ignore_changes = [tags]
}
}
# Windows specific checks
resource "aws_cloudwatch_metric_alarm" "ec2-MemoryCommittedPct" {
count = data.external.ec2-os.result.os == "Windows" ? 1 : 0
alarm_name = "${var.cw-alarm-prefix}:EC2:MemoryCommittedPct:${var.ec2-instance-id}"
comparison_operator = "GreaterThanThreshold"
evaluation_periods = "2"
metric_name = "Memory % Committed Bytes In Use"
period = "900"
statistic = "Average"
threshold = var.threshold-MemoryCommittedPct
alarm_description = "EC2:MemoryCommittedBytes"
namespace = "CWAgent"
insufficient_data_actions = []
actions_enabled = var.actions-enabled
alarm_actions = [var.sns-targets.alarm-actions-urgent]
ok_actions = [var.sns-targets.alarm-actions-urgent]
dimensions = {
objectname = "Memory"
InstanceId = var.ec2-instance-id
ImageId = data.aws_instance.ec2-instance.ami
InstanceType = data.aws_instance.ec2-instance.instance_type
}
tags = var.default-tags
lifecycle {
ignore_changes = [tags]
}
}
resource "aws_cloudwatch_metric_alarm" "ec2-OsDiskFreePct" {
count = data.external.ec2-os.result.os == "Windows" ? 1 : 0
alarm_name = "${var.cw-alarm-prefix}:EC2:OsDiskFreePct:${var.ec2-instance-id}"
comparison_operator = "LessThanThreshold"
evaluation_periods = "2"
metric_name = "LogicalDisk % Free Space"
period = "300"
statistic = "Average"
threshold = var.threshold-LogicalDiskFreePct
alarm_description = "EC2:OsDiskFreePct"
namespace = "CWAgent"
insufficient_data_actions = []
actions_enabled = var.actions-enabled
alarm_actions = [var.sns-targets.alarm-actions-urgent]
ok_actions = [var.sns-targets.alarm-actions-urgent]
dimensions = {
instance = "C:"
objectname = "LogicalDisk"
InstanceId = var.ec2-instance-id
ImageId = data.aws_instance.ec2-instance.ami
InstanceType = data.aws_instance.ec2-instance.instance_type
}
tags = var.default-tags
lifecycle {
ignore_changes = [tags]
}
}

4
modules/ManagementGovernance/Monitoring.EC2/variables.tf
View File

@ -10,4 +10,6 @@ variable "threshold-mem_free" {}
variable "threshold-swap_free" {}
variable "threshold-disk_free" {}
variable "threshold-disk_inodes_free" {}
variable "threshold-processes_total" {}
variable "threshold-processes_total" {}
variable threshold-MemoryCommittedPct {}
variable threshold-LogicalDiskFreePct {}