Clean up Grafana and Loki bootstraps

core.tf

@@ -32,6 +32,7 @@ module "loki" {
     service_port = 3100
     ingress      = true
     sticky_disk  = true
+    healthcheck  = "/ready"
     templates = jsonencode([
       {
         data = file("./loki-config.yml")
@@ -41,6 +42,34 @@ module "loki" {
   }
 }
 
+resource "consul_config_entry" "loki_intent" {
+  name = "loki"
+  kind = "service-intentions"
+
+  config_json = jsonencode({
+    Sources = [
+      {
+        Action     = "allow"
+        Name       = "grafana"
+        Precedence = 9
+        Type       = "consul"
+      },
+      {
+        Action     = "allow"
+        Name       = "promtail"
+        Precedence = 9
+        Type       = "consul"
+      },
+      {
+        Action     = "allow"
+        Name       = "syslogng-promtail"
+        Precedence = 9
+        Type       = "consul"
+      },
+    ]
+  })
+}
+
 resource "nomad_job" "syslog-ng" {
   hcl2 {
     enabled = true
@@ -48,3 +77,19 @@ resource "nomad_job" "syslog-ng" {
 
   jobspec = file("${path.module}/syslogng.nomad")
 }
+
+resource "consul_config_entry" "syslogng_promtail_intent" {
+  name = "syslogng-promtail"
+  kind = "service-intentions"
+
+  config_json = jsonencode({
+    Sources = [
+      {
+        Action     = "allow"
+        Name       = "syslogng"
+        Precedence = 9
+        Type       = "consul"
+      },
+    ]
+  })
+}

metrics/.terraform.lock.hcl

@@ -5,6 +5,7 @@ provider "registry.terraform.io/hashicorp/consul" {
   version = "2.15.0"
   hashes = [
     "h1:o+Su3YqeOkHgf86GEArIVDZfaZQphYFjAOwpi/b0bzs=",
+    "h1:tAb2gwW+oZ8/t2j7lExdqpNrxmaWsHbyA2crFWClPb0=",
     "zh:0bd2a9873099d89bd52e9eee623dd20ccb275d1e2f750da229a53a4d5b23450c",
     "zh:1c9f87d4d97b2c61d006c0bef159d61d2a661a103025f8276ebbeb000129f931",
     "zh:25b73a34115255c464be10a53f2510c4a1db958a71be31974d30654d5472e624",
@@ -22,6 +23,7 @@ provider "registry.terraform.io/hashicorp/consul" {
 provider "registry.terraform.io/hashicorp/nomad" {
   version = "1.4.16"
   hashes = [
+    "h1:PQxNPNmMVOErxryTWIJwr22k95DTSODmgRylqjc2TjI=",
     "h1:tyfjD/maKzb0RxxD9KWgLnkJu9lnYziYsQgGw85Giz8=",
     "zh:0d4fbb7030d9caac3b123e60afa44f50c83cc2a983e1866aec7f30414abe7b0e",
     "zh:0db080228e07c72d6d8ca8c45249d6f97cd0189fce82a77abbdcd49a52e57572",

metrics/exporters.nomad

@@ -2,32 +2,19 @@ job "metrics" {
   datacenters = ["dc1"]
   type = "system"
 
-  group "exporters" {
+  group "cadvisor" {
 
     network {
       mode = "bridge"
 
       port "cadvisor" {
-        host_network = "nomad-bridge"
         to = 8080
       }
 
-      port "node_exporter" {
-        host_network = "nomad-bridge"
-        to = 9100
-      }
-
-      port "promtail" {
-        host_network = "nomad-bridge"
-        to = 9080
-      }
-
       port "expose" {
-        host_network = "nomad-bridge"
       }
 
       port "cadvisor_envoy_metrics" {
-        host_network = "nomad-bridge"
         to = 9102
       }
     }
@@ -148,6 +135,18 @@ job "metrics" {
       }
     }
 
+  }
+
+  group "node_exporter" {
+
+    network {
+      mode = "bridge"
+
+      port "node_exporter" {
+        to = 9100
+      }
+    }
+
     service {
       name = "nodeexporter"
       port = "node_exporter"
@@ -162,6 +161,15 @@ job "metrics" {
         sidecar_service {
           proxy {
             local_service_port = 9100
+
+            expose {
+              path {
+                path = "/metrics"
+                protocol = "http"
+                local_path_port = 9100
+                listener_port = "node_exporter"
+              }
+            }
           }
         }
 
@@ -205,6 +213,18 @@ job "metrics" {
       }
     }
 
+  }
+
+  group "promtail" {
+
+    network {
+      mode = "bridge"
+
+      port "promtail" {
+        to = 9080
+      }
+    }
+
     service {
       name = "promtail"
       port = "promtail"
@@ -219,6 +239,11 @@ job "metrics" {
         sidecar_service {
           proxy {
             local_service_port = 9080
+
+            upstreams {
+              destination_name = "loki"
+              local_bind_port = 1000
+            }
           }
         }
 
@@ -255,12 +280,6 @@ job "metrics" {
         }
 
         # Bind mount host machine-id and log directories
-        mount {
-          type = "bind"
-          source = "/etc/machine-id"
-          target = "/etc/machine-id"
-          readonly = true
-        }
 
         mount {
           type = "bind"
@@ -283,12 +302,12 @@ job "metrics" {
           readonly = true
         }
 
-        mount {
-          type = "bind"
-          source = "/var/log/audit"
-          target = "/var/log/audit"
-          readonly = true
-        }
+        # mount {
+        #   type = "bind"
+        #   source = "/var/log/audit"
+        #   target = "/var/log/audit"
+        #   readonly = true
+        # }
       }
 
       template {
@@ -299,7 +318,8 @@ server:
   http_listen_port: 9080
 
 clients:
-  - url: http://${NOMAD_UPSTREAM_ADDR_loki}/loki/api/v1/push
+  # loki upstream: {{ env "NOMAD_UPSTREAM_ADDR_loki" }}
+  - url: http://{{ env "NOMAD_UPSTREAM_ADDR_loki" }}/loki/api/v1/push
 
 scrape_configs:
 

metrics/grafana.nomad

@@ -33,6 +33,11 @@ job "grafana" {
               destination_name = "prometheus"
               local_bind_port = 9090
             }
+
+            upstreams {
+              destination_name = "loki"
+              local_bind_port = 3100
+            }
           }
         }
 
@@ -72,10 +77,36 @@ job "grafana" {
       }
 
       env = {
-        "GF_SECURITY_ADMIN_PASSWORD" = "password",
         "GF_INSTALL_PLUGINS" = "grafana-clock-panel,grafana-piechart-panel,grafana-polystat-panel",
       }
 
+      vault {
+        policies = [
+          "access-tables",
+          "nomad-task",
+        ]
+      }
+
+      template {
+        data = <<EOF
+{{ with secret "kv/data/grafana" }}
+GF_SECURITY_ADMIN_PASSWORD={{ .Data.data.admin_pw }}
+GF_SMTP_USER={{ .Data.data.smtp_user }}
+GF_SMTP_PASSWORD={{ .Data.data.smtp_password }}
+GF_EXTERNAL_IMAGE_STORAGE_S3_ACCESS_KEY={{ .Data.data.minio_access_key }}
+GF_EXTERNAL_IMAGE_STORAGE_S3_SECRET_KEY={{ .Data.data.minio_secret_key }}
+GRAFANA_ALERT_EMAIL_ADDRESSES={{ .Data.data.alert_email_addresses }}
+{{ end }}
+{{ with secret "kv/data/slack" }}
+SLACK_BOT_URL={{ .Data.data.bot_url }}
+SLACK_BOT_TOKEN={{ .Data.data.bot_token }}
+SLACK_HOOK_URL={{ .Data.data.hook_url }}
+{{ end }}
+        EOF
+        env = true
+        destination = "secrets/conf.env"
+      }
+
       %{ for config_file in fileset(join("/", [module_path, "grafana"]), "**") ~}
       template {
         data = <<EOF
@@ -85,7 +116,7 @@ ${file(join("/", [module_path, "grafana", config_file]))}
         change_signal = "SIGHUP"
         destination = "local/config/${config_file}"
         # Change template delimeter for dashboard files that use json and have double curly braces and square braces
-        %{ if length(regexall(".*/dashboard/.*", config_file)) > 0 ~}
+        %{ if length(regexall("dashboard", config_file)) > 0 ~}
         left_delimiter = "<<<<"
         right_delimiter = ">>>>"
         %{ endif ~}

metrics/grafana/grafana.ini

@@ -281,11 +281,18 @@ log_queries =
 
 #################################### Auth Proxy ##########################
 [auth.proxy]
+{{ with service "traefik" -}}
 enabled = true
 header_name = X-WEBAUTH-USER
 header_property = username
 auto_sign_up = true
-whitelist = 192.168.2.20
+{{- $last := len . | subtract 1 -}}
+{{- $services := . -}}
+whitelist = {{ range $i := loop $last -}}
+  {{- with index $services $i }}{{ .Address }},{{ end -}}
+{{- end -}}
+  {{- with index . $last }}{{ .Address }}{{ end -}}
+{{- end }}
 
 #################################### Basic Auth ##########################
 [auth.basic]

metrics/grafana/provisioning/datasources/loki.yml

@@ -3,7 +3,7 @@ apiVersion: 1
 
 datasources:
   - name: Loki
-    url: http://loki:3100
+    url: http://{{ env "NOMAD_UPSTREAM_ADDR_loki" }}
     type: loki
     access: proxy
     isDefault: false

metrics/grafana/provisioning/datasources/prometheus.yml

@@ -3,7 +3,7 @@ apiVersion: 1
 
 datasources:
   - name: Prometheus
-    url: http://prom:9090
+    url: http://{{ env "NOMAD_UPSTREAM_ADDR_prometheus" }}
     type: prometheus
     access: proxy
     isDefault: true

metrics/grafana/provisioning/notifiers/main.yml

@@ -1,15 +1,6 @@
 ---
-notifiers:
-  # - name: Telegram
-  #   type: telegram
-  #   uid: telegram-1
-  #   org_id: 1
-  #   is_default: false
-  #   settings:
-  #     chatid: ${TELEGRAM_CHATID}
-  #     bottoken: ${TELEGRAM_BOTTOKEN}
-  #     uploadImage: true
 {{ with secret "kv/data/grafana" -}}
+notifiers:
   - name: Personal email
     type: email
     uid: email-1

metrics/metrics.tf

@@ -3,13 +3,13 @@ variable "consul_address" {
   description = "address of consul server for dynamic scraping"
 }
 
-# resource "nomad_job" "exporters" {
-#   hcl2 {
-#     enabled = true
-#   }
-#
-#   jobspec = file("${path.module}/exporters.nomad")
-# }
+resource "nomad_job" "exporters" {
+  hcl2 {
+    enabled = true
+  }
+
+  jobspec = file("${path.module}/exporters.nomad")
+}
 
 data "consul_nodes" "all-nodes" {
   query_options {

metrics/prometheus.nomad

@@ -97,8 +97,8 @@ scrape_configs:
       format:
         - "prometheus"
     consul_sd_configs:
-      # - server: "${var.consul_address}"
-      - server: "${CONSUL_HTTP_ADDR}"
+      - server: "${var.consul_address}"
+      # - server: "{{ env "CONSUL_HTTP_ADDR" }}"
         services:
           - "nomad"
         tags:
@@ -110,8 +110,8 @@ scrape_configs:
       format:
         - "prometheus"
     consul_sd_configs:
-      # - server: "${var.consul_address}"
-      - server: "${CONSUL_HTTP_ADDR}"
+      - server: "${var.consul_address}"
+      # - server: "{{ env "CONSUL_HTTP_ADDR" }}"
         services:
           - "nomad-client"
 
@@ -121,8 +121,8 @@ scrape_configs:
       format:
         - "prometheus"
     consul_sd_configs:
-      # - server: "${var.consul_address}"
-      - server: "${CONSUL_HTTP_ADDR}"
+      - server: "${var.consul_address}"
+      # - server: "{{ env "CONSUL_HTTP_ADDR" }}"
         services:
           - "consul"
     relabel_configs:
@@ -133,8 +133,8 @@ scrape_configs:
   - job_name: "exporters"
     metrics_path: "/metrics"
     consul_sd_configs:
-      # - server: "${var.consul_address}"
-      - server: "${CONSUL_HTTP_ADDR}"
+      - server: "${var.consul_address}"
+      # - server: "{{ env "CONSUL_HTTP_ADDR" }}"
     relabel_configs:
       - source_labels: [__meta_consul_service]
         action: drop
@@ -156,8 +156,8 @@ scrape_configs:
   - job_name: "envoy"
     metrics_path: "/metrics"
     consul_sd_configs:
-      # - server: "${var.consul_address}"
-      - server: "${CONSUL_HTTP_ADDR}"
+      - server: "${var.consul_address}"
+      # - server: "{{ env "CONSUL_HTTP_ADDR" }}"
     relabel_configs:
       - source_labels: [__meta_consul_service]
         action: keep

syslogng.nomad

@@ -23,8 +23,9 @@ job "syslogng" {
 
       connect {
         sidecar_service {
-          local_service_port = 1514
           proxy {
+            local_service_port = 1514
+
             upstreams {
               destination_name = "loki"
               local_bind_port = 1000
@@ -64,13 +65,13 @@ server:
   http_listen_port: 9080
 
 clients:
-  - url: http://${NOMAD_UPSTREAM_ADDR_loki}/loki/api/v1/push
+  - url: http://{{ env "NOMAD_UPSTREAM_ADDR_loki" }}/loki/api/v1/push
 
 scrape_configs:
   # TCP syslog receiver
   - job_name: syslog
     syslog:
-      listen_address: 0.0.0.0:${NOMAD_PORT_main}
+      listen_address: 0.0.0.0:{{ env "NOMAD_PORT_main" }}
       labels:
         job: syslog
     relabel_configs:
@@ -125,21 +126,13 @@ EOF
           }
         }
       }
-
-      check {
-        type = "tcp"
-        port = "main"
-        interval = "10s"
-        timeout = "10s"
-      }
-
     }
 
     task "syslogng" {
       driver = "docker"
 
       config {
-        image = "balbit/syslog-ng:latest"
+        image = "balabit/syslog-ng:3.37.1"
         ports = ["main"]
         args = ["--no-caps"]
 
@@ -165,7 +158,7 @@ source s_internal {
 
 destination d_loki {
     # Forward to Connect proxy to Promtail
-    syslog("${NOMAD_UPSTREAM_IP_syslogngpromtail}" transport("tcp") port(${NOMAD_UPSTREAM_PORT_syslogngpromtail}));
+    syslog("{{ env "NOMAD_UPSTREAM_IP_syslogng-promtail" }}" transport("tcp") port({{ env "NOMAD_UPSTREAM_PORT_syslogng-promtail" }}));
 };
 
 log { source(s_internal); destination(d_loki); };