prometheus/node: add textfile collector sensors (WIP)

8 files changed, 213 insertions, 20 deletions

diff --git a/files/common/openwrt/sensors.module_lua b/files/common/openwrt/sensors.module_lua
index 6c279188..d5d869d5 100644
--- a/files/common/openwrt/sensors.module_lua
+++ b/files/common/openwrt/sensors.module_lua
@@ -12,6 +12,12 @@ local _internal_ = {}
 _internal_.setup = {}
 _internal_.read = {}
 
+-- https://github.com/luaposix/luaposix/releases/tag/v34.1
+if glob.GLOB_NOCHECK == nil then
+    _internal_.glob = function(pattern, flags) return glob.glob(pattern) end
+else
+    _internal_.glob = glob.glob
+end
 
 
 -- ############# utils ###########
@@ -26,15 +32,23 @@ function _internal_.write_to_file(path, data)
     return ret, err
 end
 
-function _internal_.read_from_file(path)
+function _internal_.read_from_file(path, what)
     local f, err = io.open(path, "r")
     if not f then return nil, err end
 
-    local data, err = f:read("*a")
+    local data, err = f:read(what)
     f:close()
     return data, err
 end
 
+function _internal_.read_line_from_file(path)
+    return _internal_.read_from_file(path, "*l")
+end
+
+function _internal_.read_whole_file(path)
+    return _internal_.read_from_file(path, "*a")
+end
+
 
 
 -- ############# i2c ###########
@@ -67,8 +81,8 @@ function _M.i2c_add_or_replace_device(bus, address, name)
         return true, nil
     end
 
-    local contents, _ = _internal_.read_from_file(_M.i2c_device_path(bus, address) .. "/name")
-    if contents == nil or contents:gsub("%s+", "") ~= name then
+    local contents, _ = _internal_.read_line_from_file(_M.i2c_device_path(bus, address) .. "/name")
+    if contents == nil or contents ~= name then
         local ret, err = _M.i2c_delete_device(bus, address)
         if not ret then return nil, err end
 
@@ -83,7 +97,7 @@ function _M.i2c_add_or_replace_device(bus, address, name)
 end
 
 function _M.i2c_get_bus_number_from_name(name)
-    local bus_paths, glob_result = glob.glob("/sys/bus/i2c/devices/i2c-*", 0)
+    local bus_paths, glob_result = _internal_.glob("/sys/bus/i2c/devices/i2c-*", 0)
     if not bus_paths then
         if glob_result == glob.GLOB_NOMATCH then return {} end
         if glob_result == glob.GLOB_ABORTED then return nil, "glob(): aborted" end
@@ -95,14 +109,14 @@ function _M.i2c_get_bus_number_from_name(name)
         local num = string.match(bus_path, '/i2c%-(%d+)$')
         if not num then return nil, "unable to parse bus number from path: " .. bus_path end
 
-        local contents, err = _internal_.read_from_file(bus_path .. "/name")
-        if contents ~= nil and contents:gsub("%s+", "") == name then return num end
+        local contents, err = _internal_.read_line_from_file(bus_path .. "/name")
+        if contents ~= nil and contents == name then return num end
     end
     return nil, "unable to find i2c bus with name: " .. name
 end
 
 function _M.i2c_get_mux_channels(parent, address)
-    local channel_paths, glob_result = glob.glob(_M.i2c_device_path(parent, address) .. "/channel-*", 0)
+    local channel_paths, glob_result = _internal_.glob(_M.i2c_device_path(parent, address) .. "/channel-*", 0)
     if not channel_paths then
         if glob_result == glob.GLOB_NOMATCH then return {} end
         if glob_result == glob.GLOB_ABORTED then return nil, "glob(): aborted" end
@@ -281,7 +295,7 @@ function _M.gpio_setup_pin(number)
 end
 
 function _M.gpio_read_pin(number)
-    local value, err = _internal_.read_from_file(_M.gpio_pin_path(number) .. "/value")
+    local value, err = _internal_.read_line_from_file(_M.gpio_pin_path(number) .. "/value")
     if not value then return nil, err end
     return value:match( "^%s*(.-)%s*$" )
 end
@@ -318,7 +332,7 @@ end
 -- ############# iio ###########
 
 function _M.iio_device_path_from_i2c_device(bus, address)
-    local device_paths, glob_result = glob.glob(_M.i2c_device_path(bus, address) .. "/iio:device*", 0)
+    local device_paths, glob_result = _internal_.glob(_M.i2c_device_path(bus, address) .. "/iio:device*", 0)
     if not device_paths then
         if glob_result == glob.GLOB_NOMATCH then return nil, "couldn't find iio device handle for i2c device " .. _M.i2c_device_name(bus, address) end
         if glob_result == glob.GLOB_ABORTED then return nil, "glob(): aborted" end
@@ -348,12 +362,12 @@ function _M.iio_read_bmp280(bus, address)
     if not iio_device then return end
 
     local values = {}
-    local tmp, err = _internal_.read_from_file(iio_device .. '/in_pressure_input')
+    local tmp, err = _internal_.read_line_from_file(iio_device .. '/in_pressure_input')
     if tmp ~= nil then
         local val = base.tonumber(tmp)
         if val ~= nil then values['pressure'] = val*1000 end
     end
-    tmp, err = _internal_.read_from_file(iio_device .. '/in_temp_input')
+    tmp, err = _internal_.read_line_from_file(iio_device .. '/in_temp_input')
     if tmp ~= nil then
         local val = base.tonumber(tmp)
         if val ~= nil then values['temperature'] = val/1000 end
@@ -384,17 +398,17 @@ function _M.iio_read_bme280(bus, address)
     if not iio_device then return end
 
     local values = {}
-    local tmp, err = _internal_.read_from_file(iio_device .. '/in_humidityrelative_input')
+    local tmp, err = _internal_.read_line_from_file(iio_device .. '/in_humidityrelative_input')
     if tmp ~= nil then
         local val = base.tonumber(tmp)
         if val ~= nil then values['humidity'] = val/1000 end
     end
-    local tmp, err = _internal_.read_from_file(iio_device .. '/in_pressure_input')
+    local tmp, err = _internal_.read_line_from_file(iio_device .. '/in_pressure_input')
     if tmp ~= nil then
         local val = base.tonumber(tmp)
         if val ~= nil then values['pressure'] = val*1000 end
     end
-    tmp, err = _internal_.read_from_file(iio_device .. '/in_temp_input')
+    tmp, err = _internal_.read_line_from_file(iio_device .. '/in_temp_input')
     if tmp ~= nil then
         local val = base.tonumber(tmp)
         if val ~= nil then values['temperature'] = val/1000 end
@@ -415,13 +429,13 @@ function _M.iio_read_am2315(bus, address)
     if not iio_device then return end
 
     local values = {}
-    local tmp, err = _internal_.read_from_file(iio_device .. '/in_humidityrelative_raw')
+    local tmp, err = _internal_.read_line_from_file(iio_device .. '/in_humidityrelative_raw')
     if tmp ~= nil then
         local val = base.tonumber(tmp)
         if val ~= nil then values['humidity'] = val/10  end
     end
     time.nanosleep({tv_sec = 0, tv_nsec = 100000000})
-    tmp, err = _internal_.read_from_file(iio_device .. '/in_temp_raw')
+    tmp, err = _internal_.read_line_from_file(iio_device .. '/in_temp_raw')
     if tmp ~= nil then
         local val = base.tonumber(tmp)
         if val ~= nil then values['temperature'] = val/10 end
@@ -436,7 +450,7 @@ _internal_.read['am2315'] = _M.iio_read_am2315
 -- ############# hwmon ###########
 
 function _M.hwmon_device_path_from_w1_device(address)
-    local device_paths, glob_result = glob.glob(_M.w1_device_path(address) .. "/hwmon/hwmon*", 0)
+    local device_paths, glob_result = _internal_.glob(_M.w1_device_path(address) .. "/hwmon/hwmon*", 0)
     if not device_paths then
         if glob_result == glob.GLOB_NOMATCH then return nil, "couldn't find hwmon device handle for 1-wire device " .. address end
         if glob_result == glob.GLOB_ABORTED then return nil, "glob(): aborted" end
@@ -462,7 +476,7 @@ function _M.hwmon_read_w1_therm(address)
     if not hwmon_device then return end
 
     local values = {}
-    local tmp, err = _internal_.read_from_file(hwmon_device .. '/temp1_input')
+    local tmp, err = _internal_.read_line_from_file(hwmon_device .. '/temp1_input')
     if tmp ~= nil then
         local val = base.tonumber(tmp)
         if val ~= nil then values['temperature'] = val/1000 end
@@ -481,7 +495,7 @@ _internal_.read['w1-42'] = function(address) return _M.hwmon_read_w1_therm(addre
 -- ############# high-level interface ###########
 
 function _M.read_config(path)
-    sensors_json, err = _internal_.read_from_file(path)
+    sensors_json, err = _internal_.read_whole_file(path)
     if not sensors_json then return nil, err end
 
     return cjson.decode(sensors_json)
diff --git a/inventory/host_vars/ch-phoebe.yml b/inventory/host_vars/ch-phoebe.yml
index 8627f34a..e155ee2c 100644
--- a/inventory/host_vars/ch-phoebe.yml
+++ b/inventory/host_vars/ch-phoebe.yml
@@ -35,6 +35,19 @@ spreadspace_apt_repo_components:
 prometheus_exporter_node_textfile_collector_scripts:
   - deleted-libraries
   - smartmon
+  - sensors
+
+prometheus_exporter_node_textfile_collector__sensors:
+  i2c:
+  - name: "AST i2c bit bus" ## "i2c-tiny-usb"
+    devices:
+    - address: 0x18
+      type: ds2482
+  w1:
+  - name: rack-intake
+    address: 28-????????????
+  - name: ceiling
+    address: 28-????????????
 
 prometheus_exporters_extra:
   - ssl
diff --git a/roles/monitoring/prometheus/exporter/node/defaults/main.yml b/roles/monitoring/prometheus/exporter/node/defaults/main.yml
index 0bcea14f..9e8bcf8b 100644
--- a/roles/monitoring/prometheus/exporter/node/defaults/main.yml
+++ b/roles/monitoring/prometheus/exporter/node/defaults/main.yml
@@ -16,3 +16,39 @@ prometheus_exporter_node_textfile_collector_scripts:
   - deleted-libraries
 #  - smartmon
 #  - chrony
+#  - sensors
+
+# prometheus_exporter_node_textfile_collector__sensors:
+#   i2c:
+#   - number: 1
+#     devices:
+#     - name: blub
+#       address: 0x5c
+#       type: am2315
+#     - address: 0x18
+#       type: ds2482
+#   - name: "10000900.i2c"
+#     devices:
+#     - address: 0x70
+#       type: pca9548
+#       channels:
+#       - number: 0
+#         devices:
+#         - name: foo
+#           address: 0x76
+#           type: bme280
+#       - number: 1
+#         devices:
+#         - name: bar
+#           address: 0x77
+#           type: bmp280
+#   w1:
+#   - name: temp1
+#     address: 28-987654321098
+#   - name: temp2
+#     address: 28-012345678901
+#   gpio:
+#   - name: everything-is-fine-and-dandy
+#     number: 0
+#   - name: something-went-horribly-wrong
+#     number: 1
diff --git a/roles/monitoring/prometheus/exporter/node/handlers/main.yml b/roles/monitoring/prometheus/exporter/node/handlers/main.yml
index 56056ea6..8f5cb37c 100644
--- a/roles/monitoring/prometheus/exporter/node/handlers/main.yml
+++ b/roles/monitoring/prometheus/exporter/node/handlers/main.yml
@@ -8,3 +8,8 @@
   service:
     name: nginx
     state: reloaded
+
+- name: remove sensors state file
+  file:
+    path: /run/prometheus-node-exporter_sensors/state
+    state: absent
diff --git a/roles/monitoring/prometheus/exporter/node/tasks/textfile_collector_sensors.yml b/roles/monitoring/prometheus/exporter/node/tasks/textfile_collector_sensors.yml
new file mode 100644
index 00000000..966c6d7f
--- /dev/null
+++ b/roles/monitoring/prometheus/exporter/node/tasks/textfile_collector_sensors.yml
@@ -0,0 +1,32 @@
+---
+- name: install lua and libs needed by sensors script
+  apt:
+    name:
+    - lua5.1
+    - lua-cjson
+    - lua-posix
+    state: present
+
+- name: create lua 5.1 module path in /usr/local
+  file:
+    path: /usr/local/share/lua/5.1
+    state: directory
+
+- name: install sensors module
+  copy:
+    src: "{{ global_files_dir }}/common/openwrt/sensors.module_lua"
+    dest: /usr/local/share/lua/5.1/sensors.lua
+
+- name: create configure directory
+  file:
+    path: /etc/prometheus/exporter/node
+    state: directory
+
+- name: generate senors config
+  copy:
+    content: "{{ prometheus_exporter_node_textfile_collector__sensors | to_nice_json }}\n"
+    dest: /etc/prometheus/exporter/node/sensors.json
+  notify: remove sensors state file
+
+- name: install the sensors textfile collector script
+  include_tasks: textfile_collector_generic.yml
diff --git a/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.j2 b/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.j2
new file mode 100644
index 00000000..822aa3b5
--- /dev/null
+++ b/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.j2
@@ -0,0 +1,51 @@
+#!/usr/bin/lua5.1
+
+local sensors = require "sensors"
+local units = {
+    temperature = "celsius",
+    humidity = "percent",
+    pressure = "pascals",
+    gpio = "status",
+}
+
+local function metric(name, mtype, labels, value)
+  print("# TYPE " .. name .. " " .. mtype)
+
+  local label_string = ""
+  if labels then
+    for label,value in pairs(labels) do
+      label_string =  label_string .. label .. '="' .. value .. '",'
+    end
+    label_string = "{" .. string.sub(label_string, 1, -2) .. "}"
+  end
+  print(string.format("%s%s %s", name, label_string, value))
+end
+
+local function scrape(config, num_sensors)
+    local readings, err = sensors.read(config)
+    if not readings then return end
+
+    metric("sensors_count_total", "gauge", nil, num_sensors)
+    for name, values in pairs(readings) do
+        labels = { name = name, kind = values._kind_ }
+        for t, v in pairs(values) do
+            local unit = units[t]
+            if unit ~= nil then
+                metric("sensors_" .. t .. "_" .. unit, "gauge", labels, v)
+            end
+        end
+    end
+end
+
+function main()
+    local config, _ = sensors.read_config('/etc/prometheus/exporter/node/sensors.json')
+
+    -- TODO: only do this if sensor state file does not exist - else read num_sensors from state file
+    local num_sensors, err = sensors.setup(config)
+    if num_sensors == nil then error(err) end
+    -- TODO: write num_sensors to state file
+
+    scrape(config, num_sensors, units)
+end
+
+main()
diff --git a/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.service.j2 b/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.service.j2
new file mode 100644
index 00000000..7a438317
--- /dev/null
+++ b/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.service.j2
@@ -0,0 +1,33 @@
+[Unit]
+Description=Promethues node exporter textfile collector sensors
+
+[Service]
+Type=oneshot
+Environment=TMPDIR=/var/lib/prometheus-node-exporter/textfile-collector
+Environment=LC_NUMERIC=C
+ExecStart=bash -c "/usr/local/share/prometheus-node-exporter/sensors | sponge /var/lib/prometheus-node-exporter/textfile-collector/sensors.prom"
+TimeoutStartSec=30s
+
+# systemd hardening-options
+AmbientCapabilities=
+CapabilityBoundingSet=
+LockPersonality=true
+MemoryDenyWriteExecute=true
+NoNewPrivileges=true
+PrivateTmp=true
+ProtectControlGroups=true
+ProtectHome=true
+ProtectKernelModules=true
+ProtectKernelTunables=true
+ProtectSystem=strict
+ReadWritePaths=/var/lib/prometheus-node-exporter/textfile-collector
+RuntimeDirectory=prometheus-node-exporter_sensors
+RuntimeDirectoryPreserve=yes
+RemoveIPC=true
+RestrictNamespaces=true
+RestrictRealtime=true
+RestrictAddressFamilies=AF_UNIX
+SystemCallArchitectures=native
+
+[Install]
+WantedBy=multi-user.target
diff --git a/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.timer.j2 b/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.timer.j2
new file mode 100644
index 00000000..164b4681
--- /dev/null
+++ b/roles/monitoring/prometheus/exporter/node/templates/textfile-collector-scripts/sensors.timer.j2
@@ -0,0 +1,9 @@
+[Unit]
+Description=Promethues node exporter textfile collector sensors
+
+[Timer]
+OnBootSec=50s
+OnUnitActiveSec=5min
+
+[Install]
+WantedBy=timers.target