Remove garage files at bad location, add basic telemetry

14 files changed, 311 insertions, 156 deletions

diff --git a/cluster/prod/app/garage/config/garage.toml b/cluster/prod/app/garage/config/garage.toml
index a721886..224c755 100644
--- a/cluster/prod/app/garage/config/garage.toml
+++ b/cluster/prod/app/garage/config/garage.toml
@@ -21,4 +21,6 @@ bind_addr = "[::]:3902"
 root_domain = ".web.deuxfleurs.fr"
 
 [admin]
-api_bind_addr = "[::1]:3903"
+api_bind_addr = "[::]:3903"
+metrics_token = "{{ key "secrets/garage/metrics_token" | trimSpace }}"
+admin_token = "{{ key "secrets/garage/admin_token" | trimSpace }}"
diff --git a/cluster/prod/app/garage/deploy/garage.hcl b/cluster/prod/app/garage/deploy/garage.hcl
index bbaaec1..5a9c6b5 100644
--- a/cluster/prod/app/garage/deploy/garage.hcl
+++ b/cluster/prod/app/garage/deploy/garage.hcl
@@ -18,6 +18,7 @@ job "garage" {
       port "s3" { static = 3900 }
       port "rpc" { static = 3901 }
       port "web" { static = 3902 }
+      port "admin" { static = 3903 }
     }
 
 	update {
@@ -125,6 +126,24 @@ job "garage" {
         }
       }
 
+      service {
+        port = 3903
+        address_mode = "driver"
+        name = "garage-admin"
+        check {
+          type = "tcp"
+          port = 3903
+          address_mode = "driver"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+            ignore_warnings = false
+          }
+        }
+      }
+
       restart { 
         interval = "30m"  
         attempts = 10  
diff --git a/cluster/prod/garage/secrets/garage/rpc_secret b/cluster/prod/app/garage/secrets/garage/admin_token
index d831d53..d831d53 100644
--- a/cluster/prod/garage/secrets/garage/rpc_secret
+++ b/cluster/prod/app/garage/secrets/garage/admin_token
diff --git a/cluster/prod/app/garage/secrets/garage/metrics_token b/cluster/prod/app/garage/secrets/garage/metrics_token
new file mode 100644
index 0000000..d831d53
--- /dev/null
+++ b/cluster/prod/app/garage/secrets/garage/metrics_token
@@ -0,0 +1 @@
+CMD_ONCE openssl rand -hex 32
diff --git a/cluster/prod/app/telemetry/config/grafana-datasource-prometheus.yaml b/cluster/prod/app/telemetry/config/grafana-datasource-prometheus.yaml
new file mode 100644
index 0000000..9be89f9
--- /dev/null
+++ b/cluster/prod/app/telemetry/config/grafana-datasource-prometheus.yaml
@@ -0,0 +1,7 @@
+apiVersion: 1
+
+datasources:
+  - name: DS_PROMETHEUS
+    type: prometheus
+    access: proxy
+    url: http://prometheus.service.prod.consul:9090
diff --git a/cluster/prod/app/telemetry/config/grafana-litestream.yml b/cluster/prod/app/telemetry/config/grafana-litestream.yml
new file mode 100644
index 0000000..9d4d48a
--- /dev/null
+++ b/cluster/prod/app/telemetry/config/grafana-litestream.yml
@@ -0,0 +1,10 @@
+dbs:
+  - path: /ephemeral/grafana.db
+    replicas:
+      - url: s3://grafana-db/grafana.db
+        region: garage
+        endpoint: http://{{ env "attr.unique.network.ip-address" }}:3900
+        access-key-id: {{ key "secrets/telemetry/grafana/s3_access_key" | trimSpace }}
+        secret-access-key:  {{ key "secrets/telemetry/grafana/s3_secret_key" | trimSpace }}
+        force-path-style: true
+        sync-interval: 60s
diff --git a/cluster/prod/app/telemetry/config/prometheus.yml b/cluster/prod/app/telemetry/config/prometheus.yml
new file mode 100644
index 0000000..d30ee13
--- /dev/null
+++ b/cluster/prod/app/telemetry/config/prometheus.yml
@@ -0,0 +1,30 @@
+global:
+  scrape_interval:     15s # By default, scrape targets every 15 seconds.
+
+scrape_configs:
+  - job_name: 'prometheus'
+    static_configs:
+      - targets: ['localhost:9090']
+
+  - job_name: 'node-exporter'
+    consul_sd_configs:
+    - server: 'https://localhost:8501'
+      services:
+        - 'node-exporter'
+      tls_config:
+        ca_file: /etc/prometheus/consul.crt
+        cert_file: /etc/prometheus/consul-client.crt
+        key_file: /etc/prometheus/consul-client.key
+
+  - job_name: 'garage'
+    authorization:
+      type: Bearer
+      credentials: {{ key "secrets/garage/metrics_token" }}
+    consul_sd_configs:
+    - server: 'https://localhost:8501'
+      services:
+        - 'garage-admin'
+      tls_config:
+        ca_file: /etc/prometheus/consul.crt
+        cert_file: /etc/prometheus/consul-client.crt
+        key_file: /etc/prometheus/consul-client.key
diff --git a/cluster/prod/app/telemetry/deploy/telemetry-system.hcl b/cluster/prod/app/telemetry/deploy/telemetry-system.hcl
new file mode 100644
index 0000000..e4bde1a
--- /dev/null
+++ b/cluster/prod/app/telemetry/deploy/telemetry-system.hcl
@@ -0,0 +1,49 @@
+job "telemetry-system" {
+	datacenters = ["neptune", "orion"]
+	type = "system"
+	priority = "100"
+
+	group "collector" {
+    network {
+      port "node_exporter" { static = 9100 }
+    }
+
+		task "node_exporter" {
+			driver = "docker"
+
+			config {
+				image = "quay.io/prometheus/node-exporter:v1.1.2"
+				network_mode = "host"
+				volumes = [
+					"/:/host:ro,rslave"
+				]
+				args = [ "--path.rootfs=/host" ]
+			}
+
+			resources {
+				cpu = 50
+				memory = 40
+			}
+
+      service {
+        tags = [ "telemetry" ]
+        port = 9100
+        address_mode = "driver"
+        name = "node-exporter"
+        check {
+          type = "http"
+          path = "/"
+          port = 9100
+          address_mode = "driver"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+            ignore_warnings = false
+          }
+        }
+      }
+		}
+	}
+}
diff --git a/cluster/prod/app/telemetry/deploy/telemetry.hcl b/cluster/prod/app/telemetry/deploy/telemetry.hcl
new file mode 100644
index 0000000..e1f1000
--- /dev/null
+++ b/cluster/prod/app/telemetry/deploy/telemetry.hcl
@@ -0,0 +1,189 @@
+job "telemetry" {
+  datacenters = ["neptune"]
+  type = "service"
+
+  group "prometheus" {
+    count = 1
+
+    network {
+      port "prometheus" {
+        static = 9090
+      }
+    }
+
+    task "prometheus" {
+      driver = "docker"
+      config {
+        image = "prom/prometheus:v2.38.0"
+        network_mode = "host"
+        ports = [ "prometheus" ]
+        volumes = [
+          "secrets:/etc/prometheus"
+        ]
+      }
+
+      template {
+        data = file("../config/prometheus.yml")
+        destination = "secrets/prometheus.yml"
+      }
+
+      template {
+        data = "{{ key \"secrets/consul/consul.crt\" }}"
+        destination = "secrets/consul.crt"
+      }
+
+      template {
+        data = "{{ key \"secrets/consul/consul-client.crt\" }}"
+        destination = "secrets/consul-client.crt"
+      }
+
+      template {
+        data = "{{ key \"secrets/consul/consul-client.key\" }}"
+        destination = "secrets/consul-client.key"
+      }
+
+      resources {
+        memory = 500
+        cpu = 500
+      }
+
+      service {
+        port = 9090
+        address_mode = "driver"
+        name = "prometheus"
+        check {
+          type = "http"
+          path = "/"
+          port = 9090
+          address_mode = "driver"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+            ignore_warnings = false
+          }
+        }
+      }
+    }
+  }
+
+  group "grafana" {
+    count = 1
+
+    network {
+      port "grafana" {
+        static = 3719
+      }
+    }
+
+    task "restore-db" {
+      lifecycle {
+        hook = "prestart"
+        sidecar = false
+      }
+
+      driver = "docker"
+      config {
+        image = "litestream/litestream:0.3.7"
+        args = [
+          "restore", "-config", "/etc/litestream.yml", "/ephemeral/grafana.db"
+        ]
+        volumes = [
+          "../alloc/data:/ephemeral",
+          "secrets/litestream.yml:/etc/litestream.yml"
+        ]
+      }
+      user = "472"
+
+      template {
+        data = file("../config/grafana-litestream.yml")
+        destination = "secrets/litestream.yml"
+      }
+
+      resources {
+        memory = 200
+        cpu = 1000
+      }
+    }
+
+    task "grafana" {
+      driver = "docker"
+      config {
+        image = "grafana/grafana:8.4.3"
+        network_mode = "host"
+        ports = [ "grafana" ]
+        volumes = [
+          "../alloc/data:/var/lib/grafana",
+          "secrets/prometheus.yaml:/etc/grafana/provisioning/datasources/prometheus.yaml"
+        ]
+      }
+
+      template {
+        data = file("../config/grafana-datasource-prometheus.yaml")
+        destination = "secrets/prometheus.yaml"
+      }
+
+      template {
+        data = <<EOH
+GF_INSTALL_PLUGINS=grafana-clock-panel,grafana-simple-json-datasource,grafana-piechart-panel,grafana-worldmap-panel,grafana-polystat-panel
+GF_SERVER_HTTP_PORT=3719
+EOH
+        destination = "secrets/env"
+        env = true
+      }
+
+      resources {
+        memory = 500
+        cpu = 100
+      }
+
+      service {
+        tags = [
+            "grafana",
+            "tricot grafana-new.deuxfleurs.fr",
+        ]
+        port = 3719
+        address_mode = "driver"
+        name = "grafana"
+        check {
+          type = "tcp"
+          port = 3719
+          address_mode = "driver"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+            ignore_warnings = false
+          }
+        }
+      }
+    }
+
+    task "replicate-db" {
+      driver = "docker"
+      config {
+        image = "litestream/litestream:0.3.7"
+        args = [
+          "replicate", "-config", "/etc/litestream.yml"
+        ]
+        volumes = [
+          "../alloc/data:/ephemeral",
+          "secrets/litestream.yml:/etc/litestream.yml"
+        ]
+      }
+      user = "472"
+
+      template {
+        data = file("../config/grafana-litestream.yml")
+        destination = "secrets/litestream.yml"
+      }
+
+      resources {
+        memory = 200
+        cpu = 100
+      }
+    }
+  }
+}
diff --git a/cluster/prod/app/telemetry/secrets/telemetry/grafana/admin_password b/cluster/prod/app/telemetry/secrets/telemetry/grafana/admin_password
new file mode 100644
index 0000000..2f36e97
--- /dev/null
+++ b/cluster/prod/app/telemetry/secrets/telemetry/grafana/admin_password
@@ -0,0 +1 @@
+CMD openssl rand -base64 12
diff --git a/cluster/prod/app/telemetry/secrets/telemetry/grafana/s3_access_key b/cluster/prod/app/telemetry/secrets/telemetry/grafana/s3_access_key
new file mode 100644
index 0000000..c7e41a4
--- /dev/null
+++ b/cluster/prod/app/telemetry/secrets/telemetry/grafana/s3_access_key
@@ -0,0 +1 @@
+USER S3 access key for grafana db
diff --git a/cluster/prod/app/telemetry/secrets/telemetry/grafana/s3_secret_key b/cluster/prod/app/telemetry/secrets/telemetry/grafana/s3_secret_key
new file mode 100644
index 0000000..051f41a
--- /dev/null
+++ b/cluster/prod/app/telemetry/secrets/telemetry/grafana/s3_secret_key
@@ -0,0 +1 @@
+USER S3 secret key for grafana db
diff --git a/cluster/prod/garage/config/garage.toml b/cluster/prod/garage/config/garage.toml
deleted file mode 100644
index a721886..0000000
--- a/cluster/prod/garage/config/garage.toml
+++ /dev/null
@@ -1,24 +0,0 @@
-block_size = 1048576
-
-metadata_dir = "/meta"
-data_dir = "/data"
-
-replication_mode = "3"
-
-rpc_bind_addr = "[::]:3901"
-rpc_secret = "{{ key "secrets/garage/rpc_secret" | trimSpace }}"
-
-sled_cache_capacity = 536870912
-sled_sync_interval_ms = 10000
-
-[s3_api]
-s3_region = "garage"
-api_bind_addr = "[::]:3900"
-root_domain = ".garage.deuxfleurs.fr"
-
-[s3_web]
-bind_addr = "[::]:3902"
-root_domain = ".web.deuxfleurs.fr"
-
-[admin]
-api_bind_addr = "[::1]:3903"
diff --git a/cluster/prod/garage/deploy/garage.hcl b/cluster/prod/garage/deploy/garage.hcl
deleted file mode 100644
index 665515a..0000000
--- a/cluster/prod/garage/deploy/garage.hcl
+++ /dev/null
@@ -1,131 +0,0 @@
-job "garage" {
-  datacenters = ["dc1", "saturne", "neptune"]
-  type = "system"
-  priority = 80
-
-  constraint {
-    attribute = "${attr.cpu.arch}"
-    value     = "amd64"
-  }
-
-  group "garage" {
-    network {
-      port "s3" { static = 3900 }
-      port "rpc" { static = 3901 }
-      port "web" { static = 3902 }
-    }
-
-	update {
-		max_parallel = 1
-		min_healthy_time = "30s"
-		healthy_deadline = "5m"
-	}
-
-    task "server" {
-      driver = "docker"
-      config {
-        advertise_ipv6_address = true
-        image = "dxflrs/amd64_garage:v0.7.1"
-        command = "/garage"
-        args = [ "server" ]
-        network_mode = "host"
-        volumes = [
-          "/mnt/storage/garage/data:/data",
-          "/mnt/ssd/garage/meta:/meta",
-          "secrets/garage.toml:/etc/garage.toml",
-        ]
-        logging {
-          type = "journald"
-        }
-      }
-
-      template {
-        data = file("../config/garage.toml")
-        destination = "secrets/garage.toml"
-      }
-
-      resources {
-        memory = 1500
-        cpu = 1000
-      }
-
-      kill_signal = "SIGINT"
-      kill_timeout = "20s"
-
-      service {
-        tags = [
-          "garage_api",
-          "tricot garage.deuxfleurs.fr",
-          "tricot *.garage.deuxfleurs.fr",
-        ]
-        port = 3900
-        address_mode = "driver"
-        name = "garage-api"
-        check {
-          type = "tcp"
-          port = 3900
-          address_mode = "driver"
-          interval = "60s"
-          timeout = "5s"
-          check_restart {
-            limit = 3
-            grace = "90s"
-            ignore_warnings = false
-          }
-        }
-      }
-
-      service {
-        tags = ["garage-rpc"]
-        port = 3901
-        address_mode = "driver"
-        name = "garage-rpc"
-        check {
-          type = "tcp"
-          port = 3901
-          address_mode = "driver"
-          interval = "60s"
-          timeout = "5s"
-          check_restart {
-            limit = 3
-            grace = "90s"
-            ignore_warnings = false
-          }
-        }
-      }
-
-      service {
-        tags = [
-            "garage-web",
-            "tricot * 1",
-            "tricot-add-header Content-Security-Policy default-src 'self' 'unsafe-inline'; script-src 'self' 'unsafe-inline' https://code.jquery.com/; frame-ancestors 'self'",
-            "tricot-add-header Strict-Transport-Security max-age=63072000; includeSubDomains; preload",
-            "tricot-add-header X-Frame-Options SAMEORIGIN",
-            "tricot-add-header X-XSS-Protection 1; mode=block",
-        ]
-        port = 3902
-        address_mode = "driver"
-        name = "garage-web"
-        check {
-          type = "tcp"
-          port = 3902
-          address_mode = "driver"
-          interval = "60s"
-          timeout = "5s"
-          check_restart {
-            limit = 3
-            grace = "90s"
-            ignore_warnings = false
-          }
-        }
-      }
-
-      restart { 
-        interval = "30m"  
-        attempts = 10  
-        delay    = "15s"  
-        mode     = "delay"
-      }
-    }
-  }
-}