CLOUD AWS ELASTICACHE MEMCACHED DataDog monitors
How to use this module
module "datadog-monitors-cloud-aws-elasticache-memcached" {
source = "git::ssh://git@git.fr.clara.net/claranet/cloudnative/projects/datadog/terraform/monitors.git//cloud/aws/elasticache/memcached?ref={revision}"
environment = "${var.environment}"
message = "${module.datadog-message-alerting.alerting-message}"
}
Purpose
Creates DataDog monitors with the following checks:
- Elasticache memcached cache hit ratio
- Elasticache memcached CPU
Inputs
| Name | Description | Type | Default | Required |
|---|---|---|---|---|
| cpu_high_enabled | Flag to enable Elasticache memcached cpu high monitor | string | "true" |
no |
| cpu_high_extra_tags | Extra tags for Elasticache memcached cpu high monitor | list | [] |
no |
| cpu_high_message | Custom message for Elasticache memcached cpu high monitor | string | "" |
no |
| cpu_high_silenced | Groups to mute for Elasticache memcached cpu high monitor | map | {} |
no |
| cpu_high_threshold_critical | Elasticache memcached cpu high critical threshold in percentage | string | "90" |
no |
| cpu_high_threshold_warning | Elasticache memcached cpu high warning threshold in percentage | string | "75" |
no |
| cpu_high_time_aggregator | Monitor aggregator for Elasticache memcached cpu high [available values: min, max or avg] | string | "min" |
no |
| cpu_high_timeframe | Monitor timeframe for Elasticache memcached cpu high [available values: last_#m (1, 5, 10, 15, or 30), last_#h (1, 2, or 4), or last_1d] |
string | "last_15m" |
no |
| environment | Infrastructure Environment | string | n/a | yes |
| evaluation_delay | Delay in seconds for the metric evaluation | string | "900" |
no |
| filter_tags_custom | Tags used for custom filtering when filter_tags_use_defaults is false | string | "*" |
no |
| filter_tags_custom_excluded | Tags excluded for custom filtering when filter_tags_use_defaults is false | string | "" |
no |
| filter_tags_use_defaults | Use default filter tags convention | string | "true" |
no |
| get_hits_enabled | Flag to enable Elasticache memcached get hits monitor | string | "true" |
no |
| get_hits_extra_tags | Extra tags for Elasticache memcached get hits monitor | list | [] |
no |
| get_hits_message | Custom message for Elasticache memcached get hits monitor | string | "" |
no |
| get_hits_silenced | Groups to mute for Elasticache memcached get hits monitor | map | {} |
no |
| get_hits_threshold_critical | Elasticache memcached get hits critical threshold in percentage | string | "60" |
no |
| get_hits_threshold_warning | Elasticache memcached get hits warning threshold in percentage | string | "80" |
no |
| get_hits_time_aggregator | Monitor aggregator for Elasticache memcached get hits [available values: min, max or avg] | string | "max" |
no |
| get_hits_timeframe | Monitor timeframe for Elasticache memcached get hits [available values: last_#m (1, 5, 10, 15, or 30), last_#h (1, 2, or 4), or last_1d] |
string | "last_15m" |
no |
| message | Message sent when an alert is triggered | string | n/a | yes |
| new_host_delay | Delay in seconds before monitor new resource | string | "300" |
no |
Outputs
| Name | Description |
|---|---|
| memcached_cpu_high_id | id for monitor memcached_cpu_high |
| memcached_get_hits_id | id for monitor memcached_get_hits |
Related documentation
DataDog documentation: