1098 lines
26 KiB
HCL
1098 lines
26 KiB
HCL
#
|
|
# Datadog global variables
|
|
#
|
|
variable "environment" {
|
|
description = "Architecture environment"
|
|
type = "string"
|
|
}
|
|
|
|
variable "filter_tags_use_defaults" {
|
|
description = "Use default filter tags convention"
|
|
default = "true"
|
|
}
|
|
|
|
variable "filter_tags_custom" {
|
|
description = "Tags used for custom filtering when filter_tags_use_defaults is false"
|
|
default = "*"
|
|
}
|
|
|
|
variable "message" {
|
|
description = "Message sent when a monitor is triggered"
|
|
}
|
|
|
|
variable "evaluation_delay" {
|
|
description = "Delay in seconds for the metric evaluation"
|
|
default = 15
|
|
}
|
|
|
|
variable "new_host_delay" {
|
|
description = "Delay in seconds before begin to monitor new host"
|
|
default = 300
|
|
}
|
|
|
|
#
|
|
# Cluster Status Not Green
|
|
#
|
|
variable "cluster_status_not_green_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "cluster_status_not_green_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "cluster_status_not_green_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_5m"
|
|
}
|
|
|
|
variable "cluster_status_not_green_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 1
|
|
}
|
|
|
|
variable "cluster_status_not_green_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 0
|
|
}
|
|
|
|
variable "cluster_status_not_green_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "cluster_status_not_green_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Cluster Initializing Shards
|
|
#
|
|
variable "cluster_initializing_shards_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "cluster_initializing_shards_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "cluster_initializing_shards_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_5m"
|
|
}
|
|
|
|
variable "cluster_initializing_shards_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 1
|
|
}
|
|
|
|
variable "cluster_initializing_shards_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 2
|
|
}
|
|
|
|
variable "cluster_initializing_shards_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "cluster_initializing_shards_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Cluster Relocating Shards
|
|
#
|
|
variable "cluster_relocating_shards_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "cluster_relocating_shards_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "cluster_relocating_shards_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_5m"
|
|
}
|
|
|
|
variable "cluster_relocating_shards_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 1
|
|
}
|
|
|
|
variable "cluster_relocating_shards_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 2
|
|
}
|
|
|
|
variable "cluster_relocating_shards_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "cluster_relocating_shards_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Cluster Unassigned Shards
|
|
#
|
|
variable "cluster_unassigned_shards_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "cluster_unassigned_shards_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "cluster_unassigned_shards_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_5m"
|
|
}
|
|
|
|
variable "cluster_unassigned_shards_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 1
|
|
}
|
|
|
|
variable "cluster_unassigned_shards_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 2
|
|
}
|
|
|
|
variable "cluster_unassigned_shards_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "cluster_unassigned_shards_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Free Space in nodes
|
|
#
|
|
variable "node_free_space_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "node_free_space_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "sum"
|
|
}
|
|
|
|
variable "node_free_space_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_5m"
|
|
}
|
|
|
|
variable "node_free_space_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 20
|
|
}
|
|
|
|
variable "node_free_space_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 10
|
|
}
|
|
|
|
variable "node_free_space_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "node_free_space_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# JVM Heap Memory Usage
|
|
#
|
|
variable "jvm_heap_memory_usage_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "jvm_heap_memory_usage_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "jvm_heap_memory_usage_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_5m"
|
|
}
|
|
|
|
variable "jvm_heap_memory_usage_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 80
|
|
}
|
|
|
|
variable "jvm_heap_memory_usage_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 90
|
|
}
|
|
|
|
variable "jvm_heap_memory_usage_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "jvm_heap_memory_usage_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# JVM Memory Young Usage
|
|
#
|
|
variable "jvm_memory_young_usage_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "jvm_memory_young_usage_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "jvm_memory_young_usage_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "jvm_memory_young_usage_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 80
|
|
}
|
|
|
|
variable "jvm_memory_young_usage_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 90
|
|
}
|
|
|
|
variable "jvm_memory_young_usage_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "jvm_memory_young_usage_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# JVM Memory Old Usage
|
|
#
|
|
variable "jvm_memory_old_usage_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "jvm_memory_old_usage_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "jvm_memory_old_usage_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "jvm_memory_old_usage_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 80
|
|
}
|
|
|
|
variable "jvm_memory_old_usage_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 90
|
|
}
|
|
|
|
variable "jvm_memory_old_usage_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "jvm_memory_old_usage_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# JVM Garbace Collector Old Collection Latency
|
|
#
|
|
variable "jvm_gc_old_collection_latency_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "jvm_gc_old_collection_latency_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "jvm_gc_old_collection_latency_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "jvm_gc_old_collection_latency_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 160
|
|
}
|
|
|
|
variable "jvm_gc_old_collection_latency_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 200
|
|
}
|
|
|
|
variable "jvm_gc_old_collection_latency_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "jvm_gc_old_collection_latency_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# JVM Garbace Collector Young Collection Latency
|
|
#
|
|
variable "jvm_gc_young_collection_latency_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "jvm_gc_young_collection_latency_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "jvm_gc_young_collection_latency_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "jvm_gc_young_collection_latency_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 20
|
|
}
|
|
|
|
variable "jvm_gc_young_collection_latency_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 25
|
|
}
|
|
|
|
variable "jvm_gc_young_collection_latency_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "jvm_gc_young_collection_latency_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Indexing Latency
|
|
#
|
|
variable "indexing_latency_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "indexing_latency_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "indexing_latency_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "indexing_latency_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 10
|
|
}
|
|
|
|
variable "indexing_latency_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 15
|
|
}
|
|
|
|
variable "indexing_latency_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "indexing_latency_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Flush Latency
|
|
#
|
|
variable "flush_latency_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "flush_latency_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "flush_latency_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "flush_latency_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 50
|
|
}
|
|
|
|
variable "flush_latency_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 100
|
|
}
|
|
|
|
variable "flush_latency_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "flush_latency_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Open HTTP Connections Anomaly
|
|
#
|
|
variable "http_connections_anomaly_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "http_connections_anomaly_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "http_connections_anomaly_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_4h"
|
|
}
|
|
|
|
variable "http_connections_anomaly_detection_algorithm" {
|
|
description = "Anomaly Detection Algorithm used"
|
|
type = "string"
|
|
default = "agile"
|
|
}
|
|
|
|
variable "http_connections_anomaly_deviations" {
|
|
description = "Deviations to detect the anomaly"
|
|
type = "string"
|
|
default = 2
|
|
}
|
|
|
|
variable "http_connections_anomaly_direction" {
|
|
description = "Direction of the anomaly. It can be both, below or above."
|
|
type = "string"
|
|
default = "above"
|
|
}
|
|
|
|
variable "http_connections_anomaly_alert_window" {
|
|
description = "Alert window."
|
|
type = "string"
|
|
default = "last_15m"
|
|
}
|
|
|
|
variable "http_connections_anomaly_interval" {
|
|
description = "Interval."
|
|
type = "string"
|
|
default = 60
|
|
}
|
|
|
|
variable "http_connections_anomaly_count_default_zero" {
|
|
description = "Count default zero."
|
|
type = "string"
|
|
default = "true"
|
|
}
|
|
|
|
variable "http_connections_anomaly_seasonality" {
|
|
description = "Seasonality of the algorithm"
|
|
type = "string"
|
|
default = "hourly"
|
|
}
|
|
|
|
variable "http_connections_anomaly_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 0.75
|
|
}
|
|
|
|
variable "http_connections_anomaly_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 1
|
|
}
|
|
|
|
variable "http_connections_anomaly_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "http_connections_anomaly_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Query Latency
|
|
#
|
|
variable "search_query_latency_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "search_query_latency_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "search_query_latency_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "search_query_latency_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 0.5
|
|
}
|
|
|
|
variable "search_query_latency_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 1
|
|
}
|
|
|
|
variable "search_query_latency_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "search_query_latency_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Fetch Latency
|
|
#
|
|
variable "fetch_latency_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "fetch_latency_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "fetch_latency_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "fetch_latency_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 2
|
|
}
|
|
|
|
variable "fetch_latency_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 4
|
|
}
|
|
|
|
variable "fetch_latency_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "fetch_latency_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Search Query Change
|
|
#
|
|
variable "search_query_change_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "search_query_change_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "search_query_change_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "search_query_change_timeshift" {
|
|
description = "Timeshift for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "search_query_change_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 75
|
|
}
|
|
|
|
variable "search_query_change_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 100
|
|
}
|
|
|
|
variable "search_query_change_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "search_query_change_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Fetch Change
|
|
#
|
|
variable "fetch_change_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "fetch_change_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "fetch_change_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "fetch_change_timeshift" {
|
|
description = "Timeshift for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "fetch_change_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 75
|
|
}
|
|
|
|
variable "fetch_change_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 100
|
|
}
|
|
|
|
variable "fetch_change_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "fetch_change_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Field Data Evictions
|
|
#
|
|
variable "field_data_evictions_change_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "field_data_evictions_change_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "field_data_evictions_change_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_15m"
|
|
}
|
|
|
|
variable "field_data_evictions_change_timeshift" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_15m"
|
|
}
|
|
|
|
variable "field_data_evictions_change_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 60
|
|
}
|
|
|
|
variable "field_data_evictions_change_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 120
|
|
}
|
|
|
|
variable "field_data_evictions_change_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "field_data_evictions_change_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Query Cache Evictions
|
|
#
|
|
variable "query_cache_evictions_change_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "query_cache_evictions_change_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "query_cache_evictions_change_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_15m"
|
|
}
|
|
|
|
variable "query_cache_evictions_change_timeshift" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_15m"
|
|
}
|
|
|
|
variable "query_cache_evictions_change_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 60
|
|
}
|
|
|
|
variable "query_cache_evictions_change_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 120
|
|
}
|
|
|
|
variable "query_cache_evictions_change_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "query_cache_evictions_change_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Request Cache Evictions
|
|
#
|
|
variable "request_cache_evictions_change_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "request_cache_evictions_change_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "request_cache_evictions_change_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_15m"
|
|
}
|
|
|
|
variable "request_cache_evictions_change_timeshift" {
|
|
description = "Timeshift for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_15m"
|
|
}
|
|
|
|
variable "request_cache_evictions_change_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 60
|
|
}
|
|
|
|
variable "request_cache_evictions_change_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 120
|
|
}
|
|
|
|
variable "request_cache_evictions_change_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "request_cache_evictions_change_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Task Time in Queue
|
|
#
|
|
variable "task_time_in_queue_change_message" {
|
|
description = "Custom message for the Cluster Status monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "task_time_in_queue_change_time_aggregator" {
|
|
description = "Time aggregator for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "avg"
|
|
}
|
|
|
|
variable "task_time_in_queue_change_timeframe" {
|
|
description = "Timeframe for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "task_time_in_queue_change_timeshift" {
|
|
description = "Timeshift for the Cluster Status monitor"
|
|
type = "string"
|
|
default = "last_10m"
|
|
}
|
|
|
|
variable "task_time_in_queue_change_threshold_warning" {
|
|
description = "Cluster Status warning threshold"
|
|
type = "string"
|
|
default = 100
|
|
}
|
|
|
|
variable "task_time_in_queue_change_threshold_critical" {
|
|
description = "Cluster Status critical threshold"
|
|
type = "string"
|
|
default = 200
|
|
}
|
|
|
|
variable "task_time_in_queue_change_silenced" {
|
|
description = "Groups to mute for Cluster Status monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "task_time_in_queue_change_extra_tags" {
|
|
description = "Extra tags for Cluster Status monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|
|
|
|
#
|
|
# Service Check
|
|
#
|
|
variable "not_responding_silenced" {
|
|
description = "Groups to mute for Elasticsearch does not respond monitor"
|
|
type = "map"
|
|
default = {}
|
|
}
|
|
|
|
variable "not_responding_message" {
|
|
description = "Custom message for Elasticsearch does not respond monitor"
|
|
type = "string"
|
|
default = ""
|
|
}
|
|
|
|
variable "not_responding_by" {
|
|
description = "Group by for the service check"
|
|
type = "string"
|
|
default = "\"port\",\"server\""
|
|
}
|
|
|
|
variable "not_responding_last" {
|
|
description = "Parameter 'last' for the service check"
|
|
type = "string"
|
|
default = 1
|
|
}
|
|
|
|
variable "not_responding_threshold_critical" {
|
|
description = "Not responding limit (critical threshold)"
|
|
default = 5
|
|
}
|
|
|
|
variable "not_responding_threshold_warning" {
|
|
description = "Not responding limit (warning threshold)"
|
|
default = 0
|
|
}
|
|
|
|
variable "not_responding_extra_tags" {
|
|
description = "Extra tags for Elasticsearch does not respond monitor"
|
|
type = "list"
|
|
default = []
|
|
}
|