[IND-429]: Create precautionary monitors for tracking block processin…

…g speed (#47)

indexer/indexer_monitors.tf

@@ -1,13 +1,14 @@
 module "indexer_monitors" {
   count = var.enable_monitoring ? 1 : 0
 
-  source           = "../modules/indexer_monitors"
-  env_tag          = "v4-${var.environment}"
-  environment      = var.environment
-  slack_channel    = var.monitoring_slack_channel
-  pagerduty_tag    = var.monitoring_pagerduty_tag
-  ecs_cluster_name = var.full_node_name
-  msk_cluster_name = aws_msk_cluster.main.cluster_name
-  team             = var.monitoring_team
-  url              = var.indexer_url
+  source                        = "../modules/indexer_monitors"
+  env_tag                       = "v4-${var.environment}"
+  environment                   = var.environment
+  slack_channel                 = var.monitoring_slack_channel
+  pagerduty_tag                 = var.monitoring_pagerduty_tag
+  ecs_cluster_name              = var.full_node_name
+  msk_cluster_name              = aws_msk_cluster.main.cluster_name
+  team                          = var.monitoring_team
+  url                           = var.indexer_url
+  enable_precautionary_monitors = var.enable_precautionary_monitors
 }

indexer/variables.tf

@@ -332,6 +332,12 @@ variable "monitoring_team" {
   default     = "v4-indexer"
 }
 
+variable "enable_precautionary_monitors" {
+  type        = bool
+  description = "Whether to enable precautionary monitors"
+  default     = true
+}
+
 variable "indexer_url" {
   type        = string
   description = "indexer URL to monitor, should not include https:// or www. Should be something like `indexer.dydx.exchange`"

modules/indexer_monitors/monitors.tf

@@ -1,4 +1,6 @@
 resource "datadog_monitor_json" "socks_kafka_offset" {
+  count = var.enable_precautionary_monitors ? 1 : 0
+
   monitor = <<EOF
 {
 	"id": 117804982,
@@ -33,6 +35,8 @@ EOF
 }
 
 resource "datadog_monitor_json" "orderbook_crossed" {
+  count = var.enable_precautionary_monitors ? 1 : 0
+
   monitor = <<EOF
 {
 	"id": 120397508,

modules/indexer_monitors/precautionary_monitors.tf

@@ -0,0 +1,62 @@
+resource "datadog_monitor_json" "average_block_processing_rate" {
+  monitor = <<EOF
+{
+	"id": 117804982,
+	"name": "[${var.environment}] Average Indexer block processing is slow",
+	"type": "query alert",
+	"query": "avg(last_5m):avg:ender.processed_block.timing.avg{env:testnet, service:indexer, success:true} / avg:dydxprotocol.blocktime_block_time_ms{env:testnet} > 0.5",
+	"message": "This is not an actionable alert. When this alert fires, that means that the Indexer is processing blocks slow and more time should be invested in improving Ender latency. Please notify Trading if this alert fires.\n\n${local.monitor_suffix_literal}",
+	"tags": [
+		"team:${var.team}",
+		"env:${var.env_tag}"
+	],
+	"options": {
+		"thresholds": {
+			"critical": 0.5
+		},
+		"notify_audit": false,
+		"require_full_window": false,
+		"notify_no_data": true,
+		"renotify_interval": 0,
+		"include_tags": false,
+		"no_data_timeframe": 60,
+		"new_host_delay": 300,
+		"silenced": {}
+	},
+	"priority": null,
+	"restricted_roles": null
+}
+EOF
+}
+
+resource "datadog_monitor_json" "p95_block_processing_rate" {
+  monitor = <<EOF
+{
+	"id": 117804982,
+	"name": "[${var.environment}] p95 Indexer block processing is slow",
+	"type": "query alert",
+	"query": "avg(last_5m):avg:ender.processed_block.timing.95percentile{env:testnet, service:indexer, success:true} / avg:dydxprotocol.blocktime_block_time_ms{env:testnet} > 0.75",
+	"message": "This is not an actionable alert. When this alert fires, that means that the Indexer is processing blocks slow and more time should be invested in improving Ender latency. Please notify Trading if this alert fires.\n\n${local.monitor_suffix_literal}",
+	"tags": [
+		"team:${var.team}",
+		"env:${var.env_tag}"
+	],
+	"options": {
+		"thresholds": {
+			"critical": 0.75
+		},
+		"notify_audit": false,
+		"require_full_window": false,
+		"notify_no_data": true,
+		"renotify_interval": 0,
+		"include_tags": false,
+		"no_data_timeframe": 60,
+		"new_host_delay": 300,
+		"silenced": {}
+	},
+	"priority": null,
+	"restricted_roles": null
+}
+EOF
+}
+

modules/indexer_monitors/variables.tf

@@ -37,3 +37,8 @@ variable "url" {
   type        = string
   description = "Indexer URL to monitor, should not include https:// or www. Should be something like `indexer.dydx.exchange`"
 }
+
+variable "enable_precautionary_monitors" {
+  type        = bool
+  description = "Whether to enable precautionary monitors"
+}