148 lines
5.2 KiB
HCL
148 lines
5.2 KiB
HCL
data "template_file" "filter" {
|
|
template = "$${filter}"
|
|
|
|
vars {
|
|
filter = "${var.filter_tags_use_defaults == "true" ? format("dd_monitoring:enabled,dd_aws_mem:enabled,env:%s", var.environment) : "${var.filter_tags_custom}"}"
|
|
}
|
|
}
|
|
|
|
module "datadog-monitors-aws-elasticcache-common" {
|
|
source = "../common"
|
|
|
|
message = "${var.message}"
|
|
environment = "${var.environment}"
|
|
filter_tags = "${data.template_file.filter.rendered}"
|
|
resource = "memcached"
|
|
}
|
|
|
|
resource "datadog_monitor" "memcached_get_hits" {
|
|
name = "[${var.environment}] Elasticache memcached get hits {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
|
|
message = "${coalesce(var.get_hits_message, var.message)}"
|
|
|
|
type = "metric alert"
|
|
|
|
query = <<EOF
|
|
${var.get_hits_aggregator}(${var.get_hits_timeframe}): (
|
|
${var.get_hits_aggregator}:aws.elasticache.get_hits{${data.template_file.filter.rendered}} by {region,cluster} /
|
|
(${var.get_hits_aggregator}:aws.elasticache.get_hits{${data.template_file.filter.rendered}} by {region,cluster} +
|
|
${var.get_hits_aggregator}:aws.elasticache.get_misses{${data.template_file.filter.rendered}} by {region,cluster})
|
|
) < ${var.get_hits_threshold_critical}
|
|
EOF
|
|
|
|
thresholds {
|
|
warning = "${var.get_hits_threshold_warning}"
|
|
critical = "${var.get_hits_threshold_critical}"
|
|
}
|
|
|
|
notify_no_data = true
|
|
evaluation_delay = "${var.delay}"
|
|
renotify_interval = 0
|
|
notify_audit = false
|
|
timeout_h = 0
|
|
include_tags = true
|
|
locked = false
|
|
require_full_window = false
|
|
new_host_delay = "${var.delay}"
|
|
|
|
silenced = "${var.get_hits_silenced}"
|
|
|
|
tags = ["env:${var.environment}", "resource:memcached", "team:aws", "provider:aws"]
|
|
}
|
|
|
|
resource "datadog_monitor" "memcached_cpu_high" {
|
|
name = "[${var.environment}] Elasticache memcached CPU {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
|
|
message = "${coalesce(var.cpu_high_message, var.message)}"
|
|
|
|
type = "metric alert"
|
|
|
|
query = <<EOF
|
|
${var.cpu_high_aggregator}(${var.cpu_high_timeframe}): (
|
|
${var.cpu_high_aggregator}:aws.elasticache.cpuutilization{${data.template_file.filter.rendered}} by {region,cluster,node}
|
|
) > ${var.cpu_high_threshold_critical}
|
|
EOF
|
|
|
|
thresholds {
|
|
warning = "${var.cpu_high_threshold_warning}"
|
|
critical = "${var.cpu_high_threshold_critical}"
|
|
}
|
|
|
|
notify_no_data = true
|
|
evaluation_delay = "${var.delay}"
|
|
renotify_interval = 0
|
|
notify_audit = false
|
|
timeout_h = 0
|
|
include_tags = true
|
|
locked = false
|
|
require_full_window = false
|
|
new_host_delay = "${var.delay}"
|
|
|
|
silenced = "${var.cpu_high_silenced}"
|
|
|
|
tags = ["env:${var.environment}", "resource:memcached", "team:aws", "provider:aws"]
|
|
}
|
|
|
|
resource "datadog_monitor" "memcached_swap" {
|
|
name = "[${var.environment}] Elasticache memcached swap {{#is_alert}}{{{comparator}}} {{threshold}}MB ({{value}}MB){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}MB ({{value}}MB){{/is_warning}}"
|
|
message = "${coalesce(var.swap_message, var.message)}"
|
|
|
|
type = "metric alert"
|
|
|
|
query = <<EOF
|
|
${var.swap_aggregator}(${var.swap_timeframe}): (
|
|
${var.swap_aggregator}:aws.elasticache.swap_usage{${data.template_file.filter.rendered}} by {region,cluster}
|
|
) > ${var.swap_threshold_critical}
|
|
EOF
|
|
|
|
thresholds {
|
|
warning = "${var.swap_threshold_warning}"
|
|
critical = "${var.swap_threshold_critical}"
|
|
}
|
|
|
|
notify_no_data = false
|
|
evaluation_delay = "${var.delay}"
|
|
renotify_interval = 0
|
|
notify_audit = false
|
|
timeout_h = 0
|
|
include_tags = true
|
|
locked = false
|
|
require_full_window = false
|
|
new_host_delay = "${var.delay}"
|
|
|
|
silenced = "${var.swap_silenced}"
|
|
|
|
tags = ["env:${var.environment}", "resource:memcached", "team:aws", "provider:aws"]
|
|
}
|
|
|
|
resource "datadog_monitor" "memcached_free_memory" {
|
|
name = "[${var.environment}] Elasticache memcached free memory {{#is_alert}}{{{comparator}}} {{threshold}}% ({{value}}%){{/is_alert}}{{#is_warning}}{{{comparator}}} {{warn_threshold}}% ({{value}}%){{/is_warning}}"
|
|
message = "${coalesce(var.free_memory_message, var.message)}"
|
|
|
|
type = "metric alert"
|
|
|
|
query = <<EOF
|
|
${var.free_memory_aggregator}(${var.free_memory_timeframe}): (
|
|
${var.free_memory_aggregator}:aws.elasticache.freeable_memory{${data.template_file.filter.rendered}} by {region,cluster,node} /
|
|
${var.memory[var.elasticache_size]} * 100
|
|
) < ${var.free_memory_threshold_critical}
|
|
EOF
|
|
|
|
thresholds {
|
|
warning = "${var.free_memory_threshold_warning}"
|
|
critical = "${var.free_memory_threshold_critical}"
|
|
}
|
|
|
|
notify_no_data = true
|
|
evaluation_delay = "${var.delay}"
|
|
renotify_interval = 0
|
|
notify_audit = false
|
|
timeout_h = 0
|
|
include_tags = true
|
|
locked = false
|
|
require_full_window = false
|
|
new_host_delay = "${var.delay}"
|
|
|
|
silenced = "${var.free_memory_silenced}"
|
|
|
|
tags = ["env:${var.environment}", "resource:memcached", "team:aws", "provider:aws"]
|
|
}
|