82 lines
2.5 KiB
HCL
82 lines
2.5 KiB
HCL
resource "datadog_monitor" "redis_connection" {
|
|
name = "Redis connection is down (Datadog check)"
|
|
message = "{{#is_alert}}\n${var.alert_HNO} \n{{/is_alert}} \n{{#is_recovery}}\n${var.alert_HNO}\n{{/is_recovery}}"
|
|
query = "\"redis.can_connect\".over(\"app:redis\").by(\"*\").last(1).pct_by_status()"
|
|
|
|
thresholds = {
|
|
critical = 50
|
|
warning = 5
|
|
}
|
|
|
|
type = "service check"
|
|
notify_no_data = false
|
|
renotify_interval = 60
|
|
notify_audit = false
|
|
timeout_h = 0
|
|
include_tags = true
|
|
locked = false
|
|
require_full_window = true
|
|
new_host_delay = 300
|
|
notify_no_data = false
|
|
renotify_interval = 0
|
|
no_data_timeframe = 20
|
|
}
|
|
|
|
resource "datadog_monitor" "redis_eviction" {
|
|
name = "Redis eviction > 0"
|
|
message = "{{#is_alert}}\n${var.alert_HNO} \n{{/is_alert}} \n{{#is_recovery}}\n${var.alert_HNO}\n{{/is_recovery}}"
|
|
query = "min(last_5m):avg:redis.keys.evicted{*} > 0"
|
|
type = "metric alert"
|
|
|
|
notify_no_data = false
|
|
renotify_interval = 60
|
|
notify_audit = false
|
|
timeout_h = 0
|
|
include_tags = true
|
|
locked = false
|
|
require_full_window = true
|
|
new_host_delay = 300
|
|
notify_no_data = false
|
|
renotify_interval = 0
|
|
no_data_timeframe = 20
|
|
}
|
|
|
|
resource "datadog_monitor" "datadog_blocked_client" {
|
|
name = "Redis blocked clients > 0"
|
|
message = "{{#is_alert}}\n${var.alert_HNO} \n{{/is_alert}} \n{{#is_recovery}}\n${var.alert_HNO}\n{{/is_recovery}}"
|
|
query = "min(last_5m):avg:redis.clients.blocked{*} > 0"
|
|
type = "metric alert"
|
|
|
|
notify_no_data = false
|
|
renotify_interval = 60
|
|
notify_audit = false
|
|
timeout_h = 0
|
|
include_tags = true
|
|
locked = false
|
|
require_full_window = true
|
|
new_host_delay = 300
|
|
notify_no_data = false
|
|
renotify_interval = 0
|
|
no_data_timeframe = 20
|
|
}
|
|
|
|
resource "datadog_monitor" "redis_swap" {
|
|
name = "Redis begin to swap"
|
|
message = "{{#is_alert}}\n${var.alert_HNO} \n{{/is_alert}} \n{{#is_recovery}}\n${var.alert_HNO}\n{{/is_recovery}}"
|
|
query = "avg(last_5m):avg:redis.mem.fragmentation_ratio{*} <= 0.8"
|
|
type = "metric alert"
|
|
|
|
notify_no_data = false
|
|
renotify_interval = 60
|
|
notify_audit = false
|
|
timeout_h = 0
|
|
include_tags = true
|
|
locked = false
|
|
require_full_window = true
|
|
new_host_delay = 300
|
|
notify_no_data = false
|
|
renotify_interval = 0
|
|
no_data_timeframe = 20
|
|
}
|
|
|