Merge remote-tracking branch 'prism/master'

10 files changed, 490 insertions, 9 deletions

diff --git a/krebs/2configs/shack/gitlab-runner.nix b/krebs/2configs/shack/gitlab-runner.nix
index 0fd06426a..5f2ca02d9 100644
--- a/krebs/2configs/shack/gitlab-runner.nix
+++ b/krebs/2configs/shack/gitlab-runner.nix
@@ -6,16 +6,17 @@ let
   };
 in
 {
-  systemd.services.gitlab-runner.path = [
-    "/run/wrappers" # /run/wrappers/bin/su
-    "/" # /bin/sh
-  ];
   imports = [
     "${runner-src}/gitlab-runner.nix"
   ];
-  services.gitlab-runner2.enable = true;
-  ## registrationConfigurationFile contains:
-  # CI_SERVER_URL=<CI server URL>
-  # REGISTRATION_TOKEN=<registration secret>
-  services.gitlab-runner2.registrationConfigFile = <secrets/shackspace-gitlab-ci>;
+  services.gitlab-runner2 = {
+    enable = true;
+    ## registrationConfigurationFile contains:
+    # CI_SERVER_URL=<CI server URL>
+    # REGISTRATION_TOKEN=<registration secret>
+    registrationConfigFile = <secrets/shackspace-gitlab-ci>;
+    #gracefulTermination = true;
+  };
+  systemd.services.gitlab-runner2.restartIfChanged = false;
+  systemd.services.docker.restartIfChanged = false;
 }
diff --git a/krebs/2configs/shack/mqtt.nix b/krebs/2configs/shack/mqtt.nix
new file mode 100644
index 000000000..8e5438db2
--- /dev/null
+++ b/krebs/2configs/shack/mqtt.nix
@@ -0,0 +1,14 @@
+{
+  networking.firewall.allowedTCPPorts = [ 1883 ];
+  networking.firewall.allowedUDPPorts = [ 1883 ];
+  services.mosquitto = {
+    enable = true;
+    host = "0.0.0.0";
+    users = {};
+    # TODO: secure that shit
+    aclExtraConf = ''
+      pattern readwrite #
+    '';
+    allowAnonymous = true;
+  };
+}
diff --git a/krebs/2configs/shack/muell_mail.nix b/krebs/2configs/shack/muell_mail.nix
new file mode 100644
index 000000000..39d49918d
--- /dev/null
+++ b/krebs/2configs/shack/muell_mail.nix
@@ -0,0 +1,33 @@
+{ config, lib, pkgs, ... }:
+
+let
+  pkg = pkgs.callPackage (
+    pkgs.fetchgit {
+      url = "https://git.shackspace.de/rz/muell_mail";
+      rev = "861ec25ab22797d8961efb32e72d79e113aa9f0f";
+      sha256 = "sha256:18cw95zbr7isv4cw80cbpd84n5z208fwh5390i6j10jkn398mjq2";
+    }) {};
+    home = "/var/lib/muell_mail";
+    cfg = toString <secrets/shack/muell_mail.js>;
+in {
+  users.users.muell_mail = {
+    inherit home;
+    createHome = true;
+  };
+  systemd.services.muell_mail = {
+    description = "muell_mail";
+    wantedBy = [ "multi-user.target" ];
+    environment.CONFIG = "${home}/muell_mail.js";
+    serviceConfig = {
+      User = "muell_mail";
+      ExecStartPre = pkgs.writeDash "muell_mail-pre" ''
+        install -D -omuell_mail -m700 ${cfg} ${home}/muell_mail.js
+      '';
+      WorkingDirectory = home;
+      PermissionsStartOnly = true;
+      ExecStart = "${pkg}/bin/muell_mail";
+      Restart = "always";
+      PrivateTmp = true;
+    };
+  };
+}
diff --git a/krebs/2configs/shack/muellshack.nix b/krebs/2configs/shack/muellshack.nix
new file mode 100644
index 000000000..1e6843bdf
--- /dev/null
+++ b/krebs/2configs/shack/muellshack.nix
@@ -0,0 +1,43 @@
+{ config, lib, pkgs, ... }:
+
+let
+  pkg = pkgs.callPackage (
+    pkgs.fetchgit {
+      url = "https://git.shackspace.de/rz/muellshack";
+      rev = "d8a5e2d4c0a22804838675ac42b468299dcd9a76";
+      sha256 = "0ff6q64dgdxmpszp94z100fdic175b1vvxn4crg8p0jcabzxsv0m";
+    }) {};
+    home = "/var/lib/muellshack";
+    port = "8081";
+in {
+  users.users.muellshack = {
+    inherit home;
+    createHome = true;
+  };
+  services.nginx.virtualHosts."muell.shack" = {
+    locations."/" = {
+      proxyPass = "http://localhost:${port}/muellshack/";
+    };
+  };
+  services.nginx.virtualHosts."openhab.shack" = {
+    locations."/muellshack/".proxyPass = "http://localhost:${port}/muellshack/";
+  };
+  systemd.services.muellshack = {
+    description = "muellshack";
+    wantedBy = [ "multi-user.target" ];
+    environment.PORT = port;
+    serviceConfig = {
+      User = "muellshack";
+      # do not override the current storage fil
+      ExecStartPre = pkgs.writeDash "call-muell-pre" ''
+        cp -vf ${pkg}/share/static_muelldata.json ${home}
+        cp -vn  ${pkg}/share/storage.json ${home}
+        chmod 700 ${home}/storage.json
+      '';
+      WorkingDirectory = home;
+      ExecStart = "${pkg}/bin/muellshack";
+      Restart = "always";
+      PrivateTmp = true;
+    };
+  };
+}
diff --git a/krebs/2configs/shack/node-light.nix b/krebs/2configs/shack/node-light.nix
new file mode 100644
index 000000000..63772f182
--- /dev/null
+++ b/krebs/2configs/shack/node-light.nix
@@ -0,0 +1,53 @@
+{ config, lib, pkgs, ... }:
+
+let
+  pkg = pkgs.callPackage (
+    pkgs.fetchgit {
+      url = "https://git.shackspace.de/rz/node-light.git";
+      rev = "a32c782650c4cc0adf51250fe249167d7246c59b";
+      sha256 = "0clvcp1m2ay0a9ibh7s21q7d9a6nam3497bysvc6mdygblks22qy";
+    }) {};
+    home = "/var/lib/node-light";
+    port = "8082";
+in {
+  # receive response from light.shack / standby.shack
+  networking.firewall.allowedUDPPorts = [ 2342 ];
+  users.users.node-light = {
+    inherit home;
+    createHome = true;
+  };
+  services.nginx.virtualHosts."lounge.light.shack" = {
+    locations."/" = {
+      proxyPass = "http://localhost:${port}/lounge/";
+    };
+  };
+  services.nginx.virtualHosts."power.light.shack" = {
+    locations."/" = {
+      proxyPass = "http://localhost:${port}/power/";
+    };
+  };
+
+  services.nginx.virtualHosts."openhab.shack" = {
+    serverAliases = [ "lightapi.shack" ];
+    locations."/power/".proxyPass = "http://localhost:${port}/power/";
+    locations."/lounge/".proxyPass = "http://localhost:${port}/lounge/";
+  };
+  systemd.services.node-light= {
+    description = "node-light";
+    wantedBy = [ "multi-user.target" ];
+    environment.PORT = port;
+    serviceConfig = {
+      User = "node-light";
+      # do not override the current storage file
+      ExecStartPre = pkgs.writeDash "call-light-pre" ''
+        cp -vn  ${pkg}/share/storage.json ${home}
+        chmod 700 ${home}/storage.json
+
+      '';
+      WorkingDirectory = home;
+      ExecStart = "${pkg}/bin/node-light";
+      Restart = "always";
+      PrivateTmp = true;
+    };
+  };
+}
diff --git a/krebs/2configs/shack/powerraw.nix b/krebs/2configs/shack/powerraw.nix
new file mode 100644
index 000000000..af3f129c1
--- /dev/null
+++ b/krebs/2configs/shack/powerraw.nix
@@ -0,0 +1,72 @@
+{ config, lib, pkgs, ... }:
+# Replacement for powerraw.shack pollin box
+# Requires usb-serial device on host
+# Requires mqtt available at mqtt.shack
+# Requires hostname powerraw.shack
+let
+  influx-url = "http://influx.shack:8086";
+  pkg = pkgs.python3.pkgs.callPackage (
+    pkgs.fetchgit {
+      url = "https://git.shackspace.de/rz/powermeter.git";
+      rev = "96609f0d632e0732afa768ddd7b3f8841ca37c1b";
+      sha256 = "sha256:0wfpm3ik5r081qv2crmpjwylgg2v8ximq347qh0fzq1rwv0dqbnn";
+    }) {};
+in {
+  # receive response from light.shack / standby.shack
+  networking.firewall.allowedUDPPorts = [ 11111 ];
+  users.users.powermeter.extraGroups = [ "dialout" ];
+
+  systemd.services.powermeter-serial2mqtt = {
+    description = "powerraw Serial -> mqtt";
+    wantedBy = [ "multi-user.target" ];
+    serviceConfig = {
+      User = "powermeter";
+      ExecStart = "${pkg}/bin/powermeter-serial2mqtt";
+      Restart = "always";
+      PrivateTmp = true;
+    };
+  };
+
+  systemd.services.powermeter-mqtt2socket = {
+    description = "powerraw mqtt -> raw socket 11111";
+    wantedBy = [ "multi-user.target" ];
+    serviceConfig = {
+      User = "powermeter";
+      ExecStart = "${pkg}/bin/powermeter-mqtt2socket";
+      Restart = "always";
+      PrivateTmp = true;
+    };
+  };
+
+  services.telegraf = {
+    enable = true;
+    extraConfig = {
+      agent.debug = false;
+      outputs = {
+        influxdb = [{
+          urls = [ influx-url ];
+          database = "telegraf";
+        }];
+      };
+    };
+  };
+
+  services.telegraf.extraConfig.inputs.mqtt_consumer = let
+    genTopic = name: topic: tags: {
+      servers = [ "tcp://mqtt.shack:1883" ];
+      qos = 0;
+      connection_timeout = "30s";
+      topics = [ topic ];
+      inherit tags;
+      persistent_session = false;
+      name_override = name;
+      data_format = "value";
+      data_type = "float";
+    };
+    sensor = "total";
+    types  = [ "Voltage" "Current" "Power" ];
+    phases = [ 1 2 3 ];
+  in
+    [ (genTopic "Power consumed" "/power/${sensor}/consumed"  { inherit sensor; }) ] ++
+    (lib.flatten (map (type: (map (phase: (genTopic "Power" "/power/${sensor}/L${toString phase}/${type}" { inherit sensor phase type; }) ) phases)) types));
+}
diff --git a/krebs/2configs/shack/prometheus/node.nix b/krebs/2configs/shack/prometheus/node.nix
new file mode 100644
index 000000000..5462464d5
--- /dev/null
+++ b/krebs/2configs/shack/prometheus/node.nix
@@ -0,0 +1,27 @@
+{ config, lib, pkgs, ... }:
+{
+  networking.firewall.allowedTCPPorts = [ 9100 ];
+
+  services.prometheus.exporters.node = {
+    enable = true;
+    enabledCollectors = [
+      "conntrack"
+      "diskstats"
+      "entropy"
+      "filefd"
+      "filesystem"
+      "loadavg"
+      "mdadm"
+      "meminfo"
+      "netdev"
+      "netstat"
+      "stat"
+      "time"
+      "vmstat"
+      "systemd"
+      "logind"
+      "interrupts"
+      "ksmd"
+    ];
+  };
+}
diff --git a/krebs/2configs/shack/prometheus/server.nix b/krebs/2configs/shack/prometheus/server.nix
new file mode 100644
index 000000000..c936f2531
--- /dev/null
+++ b/krebs/2configs/shack/prometheus/server.nix
@@ -0,0 +1,195 @@
+{ pkgs, lib, config, ... }:
+# from https://gist.github.com/globin/02496fd10a96a36f092a8e7ea0e6c7dd
+{
+  networking = {
+    firewall.allowedTCPPorts = [
+      3000  # grafana
+      9090  # prometheus
+      9093  # alertmanager
+    ];
+    useDHCP = true;
+  };
+
+  services = {
+    prometheus = {
+      enable = true;
+      extraFlags = [
+        "-storage.local.retention 8760h"
+        "-storage.local.series-file-shrink-ratio 0.3"
+        "-storage.local.memory-chunks 2097152"
+        "-storage.local.max-chunks-to-persist 1048576"
+        "-storage.local.index-cache-size.fingerprint-to-metric 2097152"
+        "-storage.local.index-cache-size.fingerprint-to-timerange 1048576"
+        "-storage.local.index-cache-size.label-name-to-label-values 2097152"
+        "-storage.local.index-cache-size.label-pair-to-fingerprints 41943040"
+      ];
+      alertmanagerURL = [ "http://localhost:9093" ];
+      rules = [
+        ''
+          ALERT node_down
+          IF up == 0
+          FOR 5m
+          LABELS {
+            severity="page"
+          }
+          ANNOTATIONS {
+            summary = "{{$labels.alias}}: Node is down.",
+            description = "{{$labels.alias}} has been down for more than 5 minutes."
+          }
+          ALERT node_systemd_service_failed
+          IF node_systemd_unit_state{state="failed"} == 1
+          FOR 4m
+          LABELS {
+            severity="page"
+          }
+          ANNOTATIONS {
+            summary = "{{$labels.alias}}: Service {{$labels.name}} failed to start.",
+            description = "{{$labels.alias}} failed to (re)start service {{$labels.name}}."
+          }
+          ALERT node_filesystem_full_90percent
+          IF sort(node_filesystem_free{device!="ramfs"} < node_filesystem_size{device!="ramfs"} * 0.1) / 1024^3
+          FOR 5m
+          LABELS {
+            severity="page"
+          }
+          ANNOTATIONS {
+            summary = "{{$labels.alias}}: Filesystem is running out of space soon.",
+            description = "{{$labels.alias}} device {{$labels.device}} on {{$labels.mountpoint}} got less than 10% space left on its filesystem."
+          }
+          ALERT node_filesystem_full_in_4h
+          IF predict_linear(node_filesystem_free{device!="ramfs"}[1h], 4*3600) <= 0
+          FOR 5m
+          LABELS {
+            severity="page"
+          }
+          ANNOTATIONS {
+            summary = "{{$labels.alias}}: Filesystem is running out of space in 4 hours.",
+            description = "{{$labels.alias}} device {{$labels.device}} on {{$labels.mountpoint}} is running out of space of in approx. 4 hours"
+          }
+          ALERT node_filedescriptors_full_in_3h
+          IF predict_linear(node_filefd_allocated[1h], 3*3600) >= node_filefd_maximum
+          FOR 20m
+          LABELS {
+            severity="page"
+          }
+          ANNOTATIONS {
+            summary = "{{$labels.alias}} is running out of available file descriptors in 3 hours.",
+            description = "{{$labels.alias}} is running out of available file descriptors in approx. 3 hours"
+          }
+          ALERT node_load1_90percent
+          IF node_load1 / on(alias) count(node_cpu{mode="system"}) by (alias) >= 0.9
+          FOR 1h
+          LABELS {
+            severity="page"
+          }
+          ANNOTATIONS {
+            summary = "{{$labels.alias}}: Running on high load.",
+            description = "{{$labels.alias}} is running with > 90% total load for at least 1h."
+          }
+          ALERT node_cpu_util_90percent
+          IF 100 - (avg by (alias) (irate(node_cpu{mode="idle"}[5m])) * 100) >= 90
+          FOR 1h
+          LABELS {
+            severity="page"
+          }
+          ANNOTATIONS {
+            summary = "{{$labels.alias}}: High CPU utilization.",
+            description = "{{$labels.alias}} has total CPU utilization over 90% for at least 1h."
+          }
+          ALERT node_ram_using_90percent
+          IF node_memory_MemFree + node_memory_Buffers + node_memory_Cached < node_memory_MemTotal * 0.1
+          FOR 30m
+          LABELS {
+            severity="page"
+          }
+          ANNOTATIONS {
+            summary="{{$labels.alias}}: Using lots of RAM.",
+            description="{{$labels.alias}} is using at least 90% of its RAM for at least 30 minutes now.",
+          }
+        ''
+      ];
+      scrapeConfigs = [
+        {
+          job_name = "node";
+          scrape_interval = "10s";
+          static_configs = [
+            {
+              targets = [
+                "localhost:9100"
+              ];
+              labels = {
+                alias = "wolf.shack";
+              };
+            }
+            {
+              targets = [
+                "localhost:9130"
+              ];
+              labels = {
+                alias = "unifi.shack";
+              };
+            }
+            {
+              targets = [
+                "10.42.22.184:9100" # puyak.shack
+              ];
+              labels = {
+                alias = "puyak.shack";
+              };
+            }
+            {
+              targets = [
+                "phenylbutazon.shack:9100"
+              ];
+              labels = {
+                alias = "phenylbutazon.shack";
+              };
+            }
+            {
+              targets = [
+                "ibuprofen.shack:9100"
+              ];
+              labels = {
+                alias = "ibuprofen.shack";
+              };
+            }
+          ];
+        }
+      ];
+      alertmanager = {
+        enable = true;
+        listenAddress = "0.0.0.0";
+        configuration = {
+          "global" = {
+            "smtp_smarthost" = "smtp.example.com:587";
+            "smtp_from" = "alertmanager@example.com";
+          };
+          "route" = {
+            "group_by" = [ "alertname" "alias" ];
+            "group_wait" = "30s";
+            "group_interval" = "2m";
+            "repeat_interval" = "4h";
+            "receiver" = "team-admins";
+          };
+          "receivers" = [
+            {
+              "name" = "team-admins";
+              "email_configs" = [
+                {
+                  "to" = "devnull@example.com";
+                  "send_resolved" = true;
+                }
+              ];
+              "webhook_configs" = [
+                {
+                  "url" = "https://example.com/prometheus-alerts";
+                  "send_resolved" = true;
+                }
+              ];
+            }
+          ];
+        };
+      };
+    };
+  };
+}
diff --git a/krebs/2configs/shack/prometheus/unifi.nix b/krebs/2configs/shack/prometheus/unifi.nix
new file mode 100644
index 000000000..401ecb024
--- /dev/null
+++ b/krebs/2configs/shack/prometheus/unifi.nix
@@ -0,0 +1,10 @@
+{lib, ... }:
+{
+  services.prometheus.exporters.unifi = {
+    enable = true;
+    unifiAddress = "https://unifi.shack:8443/";
+    unifiInsecure = true;
+    unifiUsername = "prometheus"; # needed manual login after setup to confirm the password
+    unifiPassword = lib.replaceChars ["\n"] [""] (builtins.readFile <secrets/shack/unifi-prometheus-pw>);
+  };
+}
diff --git a/krebs/2configs/shack/s3-power.nix b/krebs/2configs/shack/s3-power.nix
new file mode 100644
index 000000000..40c42260f
--- /dev/null
+++ b/krebs/2configs/shack/s3-power.nix
@@ -0,0 +1,33 @@
+{ config, lib, pkgs, ... }:
+
+let
+  pkg = pkgs.callPackage (
+    pkgs.fetchgit {
+      url = "https://git.shackspace.de/rz/s3-power";
+      rev = "b2b87b56bb40d714dbbecd1285566870b256aec4";
+      sha256 = "sha256:02wikwf3rgkkggwbwqisdvhlwd38w5pw011xhwvhnj114s3rynan";
+    }) {};
+    home = "/var/lib/s3-power";
+    cfg = toString <secrets/shack/s3-power.json>;
+in {
+  users.users.s3_power = {
+    inherit home;
+    createHome = true;
+  };
+  systemd.services.s3-power = {
+    startAt = "daily";
+    description = "s3-power";
+    environment.CONFIG = "${home}/s3-power.json";
+    serviceConfig = {
+      Type = "oneshot";
+      User = "s3_power";
+      ExecStartPre = pkgs.writeDash "s3-power-pre" ''
+        install -D -os3_power -m700 ${cfg} ${home}/s3-power.json
+      '';
+      WorkingDirectory = home;
+      PermissionsStartOnly = true;
+      ExecStart = "${pkg}/bin/s3-power";
+      PrivateTmp = true;
+    };
+  };
+}