Staging: cleanup garage job

2 files changed, 60 insertions, 116 deletions

diff --git a/cluster/staging/app/garage/config/garage.toml b/cluster/staging/app/garage/config/garage.toml
index 3686338..f14a602 100644
--- a/cluster/staging/app/garage/config/garage.toml
+++ b/cluster/staging/app/garage/config/garage.toml
@@ -10,9 +10,6 @@ rpc_bind_addr = "[{{ env "meta.public_ipv6" }}]:3991"
 rpc_public_addr = "[{{ env "meta.public_ipv6" }}]:3991"
 rpc_secret = "{{ key "secrets/garage-staging/rpc_secret" | trimSpace }}"
 
-#consul_host = "localhost:8500"
-#consul_service_name = "garage-staging-rpc-self-advertised"
-
 bootstrap_peers = []
 
 [consul_discovery]
diff --git a/cluster/staging/app/garage/deploy/garage-nix2.hcl b/cluster/staging/app/garage/deploy/garage-nix2.hcl
index a60b506..5baf797 100644
--- a/cluster/staging/app/garage/deploy/garage-nix2.hcl
+++ b/cluster/staging/app/garage/deploy/garage-nix2.hcl
@@ -1,13 +1,13 @@
 job "garage-staging" {
   type = "system"
-  #datacenters = [ "neptune", "pluton" ]
-  datacenters = [ "neptune" ]
+  priority = 90
 
-  priority = 80
+  datacenters = [ "neptune" ]
 
-  constraint {
-    attribute = "${attr.cpu.arch}"
-    value     = "amd64"
+  update {
+    max_parallel = 1
+    stagger = "30s"
+    min_healthy_time = "10s"
   }
 
   group "garage-staging" {
@@ -19,43 +19,6 @@ job "garage-staging" {
       port "admin" { static = 3909 }
     }
 
-    update {
-      max_parallel = 1
-      min_healthy_time = "30s"
-      healthy_deadline = "5m"
-    }
-
-    # task "repair" {
-    #   lifecycle {
-    #     hook = "prestart"
-    #     sidecar = false
-    #   }
-
-    #   driver = "docker"
-
-    #   config {
-    #     image = "dxflrs/amd64_garage:v0.7.99-k2v"
-    #     command = "/garage"
-    #     args = [ "offline-repair", "--yes", "object_counters" ]
-    #     network_mode = "host"
-    #     volumes = [
-    #       "/mnt/storage/garage-staging/data:/data",
-    #       "/mnt/ssd/garage-staging/meta:/meta",
-    #       "secrets/garage.toml:/etc/garage.toml",
-    #     ]
-    #   }
-
-    #   template {
-    #     data = file("../config/garage.toml")
-    #     destination = "secrets/garage.toml"
-    #   }
-
-    #   resources {
-    #     memory = 2000
-    #     cpu = 1000
-    #   }
-    # }
-
     task "server" {
       driver = "nix2"
 
@@ -71,6 +34,12 @@ job "garage-staging" {
           "/mnt/ssd/garage-staging/meta" = "/meta",
         }
       }
+
+      env = {
+        RUST_LOG = "garage=debug",
+      }
+
+      # files currently owned by root, we don't want to chown everything
       user = "root"
 
       template {
@@ -93,14 +62,6 @@ job "garage-staging" {
         destination = "etc/garage/consul-client.key"
       }
 
-      template {
-        data = <<EOH
-RUST_LOG=garage=debug
-EOH
-        destination = "secrets/env"
-        env = true
-      }
-
       resources {
         memory = 2000
         memory_max = 3000
@@ -111,25 +72,30 @@ EOH
       kill_timeout = "20s"
 
       service {
+        name = "garage-staging-rpc"
+        tags = ["garage-staging-rpc"]
+        port = "rpc"
+        # No check on RPC, it wouldn't try connecting to the correct address
+        # (Garage listens only on IPv6 for RPC, see config file)
+      }
+
+      service {
         name = "garage-staging-s3-api"
         tags = [
           "garage-staging-api",
-          "tricot garage-staging.home.adnab.me",
           "tricot garage.staging.deuxfleurs.org",
           "tricot-add-header Access-Control-Allow-Origin *",
         ]
         port = "s3"
-        #address_mode = "host"
-        #check {
-        #  type = "tcp"
-        #  interval = "60s"
-        #  timeout = "5s"
-        #  check_restart {
-        #    limit = 3
-        #    grace = "90s"
-        #    ignore_warnings = false
-        #  }
-        #}
+        check {
+          type = "tcp"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+          }
+        }
       }
 
       service {
@@ -140,38 +106,15 @@ EOH
           "tricot-add-header Access-Control-Allow-Origin *",
         ]
         port = "k2v"
-        #address_mode = "driver"
-        # check {
-        #   type = "tcp"
-        #   port = 3993
-        #   address_mode = "driver"
-        #   interval = "60s"
-        #   timeout = "5s"
-        #   check_restart {
-        #     limit = 3
-        #     grace = "90s"
-        #     ignore_warnings = false
-        #   }
-        # }
-      }
-
-      service {
-        name = "garage-staging-rpc"
-        tags = ["garage-staging-rpc"]
-        port = "rpc"
-        #address_mode = "driver"
-        #check {
-        #  type = "tcp"
-        #  port = 3991
-        #  address_mode = "driver"
-        #  interval = "60s"
-        #  timeout = "5s"
-        #  check_restart {
-        #    limit = 3
-        #    grace = "90s"
-        #    ignore_warnings = false
-        #  }
-        #}
+        check {
+          type = "tcp"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+          }
+        }
       }
 
       service {
@@ -184,19 +127,15 @@ EOH
             "tricot-add-header Access-Control-Allow-Origin *",
         ]
         port = "web"
-        #address_mode = "driver"
-        #check {
-        #  type = "tcp"
-        #  port = 3992
-        #  address_mode = "driver"
-        #  interval = "60s"
-        #  timeout = "5s"
-        #  check_restart {
-        #    limit = 3
-        #    grace = "90s"
-        #    ignore_warnings = false
-        #  }
-        #}
+        check {
+          type = "tcp"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+          }
+        }
       }
 
       service {
@@ -205,13 +144,21 @@ EOH
           "garage-staging-admin",
         ]
         port = "admin"
-        #address_mode = "driver"
+        check {
+          type = "tcp"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+          }
+        }
       }
 
-      restart { 
-        interval = "1m"  
-        attempts = 10  
-        delay    = "15s"  
+      restart {
+        interval = "5m"
+        attempts = 10
+        delay    = "15s"
         mode     = "delay"
       }
     }