From 0e5b90e70c2609bee95d06a3c825bc9cd44ec63e Mon Sep 17 00:00:00 2001
From: Daniel <116079873+heydbut@users.noreply.github.com>
Date: Thu, 27 Apr 2023 01:03:14 +0330
Subject: [PATCH] Add cockroachdb dashboard (#64)

---
 .../main.tf                                   | 245 ++++++++++++++++++
 .../outputs.tf                                |   4 +
 .../variables.tf                              |   4 +
 3 files changed, 253 insertions(+)
 create mode 100644 collector-dashboards/otel-collector-cockroachdb-dashboard/main.tf
 create mode 100644 collector-dashboards/otel-collector-cockroachdb-dashboard/outputs.tf
 create mode 100644 collector-dashboards/otel-collector-cockroachdb-dashboard/variables.tf

diff --git a/collector-dashboards/otel-collector-cockroachdb-dashboard/main.tf b/collector-dashboards/otel-collector-cockroachdb-dashboard/main.tf
new file mode 100644
index 0000000..3de1ccc
--- /dev/null
+++ b/collector-dashboards/otel-collector-cockroachdb-dashboard/main.tf
@@ -0,0 +1,245 @@
+terraform {
+  required_providers {
+    lightstep = {
+      source  = "lightstep/lightstep"
+      version = "~> 1.70.10"
+    }
+  }
+  required_version = ">= v1.0.11"
+}
+
+resource "lightstep_dashboard" "otel_collector_cockroachdb_dashboard" {
+  project_name          = var.lightstep_project
+  dashboard_name        = "CockroachDB - Overview"
+  dashboard_description = "This dashboard provides a high-level overview of key CockroachDB metrics so you can ensure that your cluster is available, serving queries, and has sufficient resources to maintain high levels of performance."
+
+  chart {
+    name = "KV Pairs"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "big_number"
+      hidden       = false
+      query_string = "metric syscount | reduce mean | group_by [], mean"
+    }
+
+  }
+
+  chart {
+    name = "Memory Usage Per SQL Statement"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "line"
+      hidden       = false
+      query_string = "metric sql_stats_mem_max | delta | group_by [], sum | point percentile(value, 50.0), percentile(value, 95.0), percentile(value, 99.0), percentile(value, 99.9)"
+    }
+
+  }
+
+  chart {
+    name = "Capacity Used"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "big_number"
+      hidden       = false
+      query_string = "metric capacity_used | reduce mean | group_by [], mean"
+    }
+
+  }
+
+  chart {
+    name = "Total Capacity"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "big_number"
+      hidden       = false
+      query_string = "metric capacity | reduce mean | group_by [], mean"
+    }
+
+  }
+
+  chart {
+    name = "SQL UPDATE Count"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "bar"
+      hidden       = false
+      query_string = "metric sql_update_count | rate | group_by [], sum"
+    }
+
+  }
+
+  chart {
+    name = "SQL INSERT Count"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "bar"
+      hidden       = false
+      query_string = "metric sql_insert_count | rate | group_by [], sum"
+    }
+
+  }
+
+  chart {
+    name = "SQL DELETE Count"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "bar"
+      hidden       = false
+      query_string = "metric sql_delete_count | rate | group_by [], sum"
+    }
+
+  }
+
+  chart {
+    name = "SQL SELECT Count"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "bar"
+      hidden       = false
+      query_string = "metric sql_select_count | rate | group_by [], sum"
+    }
+
+  }
+
+  chart {
+    name = "Batch Request Latency"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "line"
+      hidden       = false
+      query_string = "metric exec_latency | delta | group_by [], sum | point percentile(value, 50.0), percentile(value, 95.0), percentile(value, 99.0), percentile(value, 99.9)"
+    }
+
+  }
+
+  chart {
+    name = "SQL Connections"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "big_number"
+      hidden       = false
+      query_string = "metric sql_conns | reduce mean | group_by [], mean"
+    }
+
+  }
+
+  chart {
+    name = "SQL Service Latency"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "line"
+      hidden       = false
+      query_string = "metric sql_service_latency | delta | group_by [], sum | point percentile(value, 50.0), percentile(value, 95.0), percentile(value, 99.0), percentile(value, 99.9)"
+    }
+
+  }
+
+  chart {
+    name = "Capacity Available (%)"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "((a/b)*100)"
+      display      = "big_number"
+      hidden       = false
+      query_string = <<EOT
+with
+  a = metric capacity_available | latest | group_by [], mean;
+  b = metric capacity | reduce mean | group_by [], mean;
+join (((a / b)*100)), a=0, b=0
+EOT
+    }
+
+  }
+
+  chart {
+    name = "Current User CPU (%)"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "line"
+      hidden       = false
+      query_string = "metric sys_cpu_user_percent | latest | group_by [], mean"
+    }
+
+  }
+
+  chart {
+    name = "System CPU (%)"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "line"
+      hidden       = false
+      query_string = "metric sys_cpu_sys_percent | latest | group_by [], mean"
+    }
+
+  }
+
+  chart {
+    name = "Live Nodes"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "big_number"
+      hidden       = false
+      query_string = "metric liveness_livenodes | reduce mean | group_by [], mean"
+    }
+
+  }
+
+  chart {
+    name = "Memory Usage Per SQL Session"
+    rank = "0"
+    type = "timeseries"
+
+    query {
+      query_name   = "a"
+      display      = "line"
+      hidden       = false
+      query_string = "metric sql_mem_sql_session_max | delta | group_by [], sum | point percentile(value, 50.0), percentile(value, 95.0), percentile(value, 99.0), percentile(value, 99.9)"
+    }
+
+  }
+
+}
\ No newline at end of file
diff --git a/collector-dashboards/otel-collector-cockroachdb-dashboard/outputs.tf b/collector-dashboards/otel-collector-cockroachdb-dashboard/outputs.tf
new file mode 100644
index 0000000..a8e8d8a
--- /dev/null
+++ b/collector-dashboards/otel-collector-cockroachdb-dashboard/outputs.tf
@@ -0,0 +1,4 @@
+output "dashboard_url" {
+  value       = "https://app.lightstep.com/${var.lightstep_project}/dashboard/${lightstep_dashboard.otel_collector_cockroachdb_dashboard.id}"
+  description = "OpenTelemetry Collector CockroachDB Metrics Dashboard URL"
+}
diff --git a/collector-dashboards/otel-collector-cockroachdb-dashboard/variables.tf b/collector-dashboards/otel-collector-cockroachdb-dashboard/variables.tf
new file mode 100644
index 0000000..21ee69f
--- /dev/null
+++ b/collector-dashboards/otel-collector-cockroachdb-dashboard/variables.tf
@@ -0,0 +1,4 @@
+variable "lightstep_project" {
+  description = "Name of Lightstep project"
+  type        = string
+}