Implement Prometheus exporter

lib/ring/sqa/analyzer.rb

@@ -17,6 +17,7 @@ def run
         records = records.all
         @graphite.add @db.id_range(first_id, @db_id_seen).all if @graphite
         @influxdb.add @db.id_range(first_id, @db_id_seen).all if @influxdb
+        @prometheus.add @db.id_range(first_id, @db_id_seen).all if @prometheus
         @buffer.push records.map { |record| record.peer }
         @buffer.exceed_median? ? @alarm.set(@buffer) : @alarm.clear(@buffer)
         delay = INTERVAL-(Time.now-start)
@@ -43,6 +44,7 @@ def initialize database, nodes
       @db_id_seen = 0
       @graphite   = graphite if CFG.graphite?
       @influxdb   = influxdb if CFG.influxdb?
+      @prometheus = prometheus if CFG.prometheus?
     end
 
     def graphite
@@ -55,6 +57,10 @@ def influxdb
       InfluxDBWriter.new @nodes
     end
 
+    def prometheus
+      require_relative 'prometheus'
+      Prometheus.new @nodes
+    end
 
   end
 

lib/ring/sqa/log.rb

@@ -4,13 +4,19 @@ class SQA
   if CFG.debug?
     require 'logger'
     Log = Logger.new STDERR
+    Log.level = Logger::DEBUG
+    AccessLog = Log
   else
     begin
       require 'syslog/logger'
       Log = Syslog::Logger.new 'ring-sqad%i' % ( CFG.afi == "ipv6" ? 6 : 4 )
+      Log.level = Logger::INFO
+      AccessLog = Logger.new STDERR
     rescue LoadError
       require 'logger'
       Log = Logger.new STDERR
+      Log.level = Logger::INFO
+      AccessLog = Log
     end
   end
 

lib/ring/sqa/prometheus.rb

@@ -0,0 +1,115 @@
+require 'webrick'
+require 'prometheus/client'
+require 'prometheus/client/formats/text'
+require "webrick"
+require "webrick/accesslog"
+
+module Ring
+class SQA
+
+  class Prometheus
+    STATE_METRIC_NAME = :nlnog_ring_sqa_node_state
+    LATENCY_METRIC_NAME = :nlnog_ring_sqa_node_latency_microseconds
+    NODE_LAST_CHECKED_METRIC_NAME = :nlnog_ring_sqa_node_last_checked_timestamp_seconds
+
+    def add records
+      host = @hostname.split(".").first
+      node =  @nodes.all
+
+      records.each do |record|
+        nodename = nodecc = node[record.peer][:name].split(".").first
+        nodecc = node[record.peer][:cc].downcase
+        nodestatus = record.result
+        if nodestatus == "ok"
+          nodeup = 1
+          nodelatency = record.latency
+        else
+          nodeup = 0
+          nodelatency = -1
+        end
+
+        labels = {
+          address_family: CFG.afi,
+          country_code: nodecc,
+          ring_source: host,
+          ring_target: nodename,
+          target_status: nodestatus
+        }
+
+        @state_metric.set(nodeup, labels: labels)
+        @latency_metric.set(nodelatency, labels: labels)
+        @refreshed_metric.set(record.time, labels: labels)
+      end
+    end
+
+    private
+
+    def initialize nodes
+      bind = CFG.prometheus.bind? || "127.0.0.1"
+      port = CFG.prometheus.port? || 8129
+      @hostname = Ring::SQA::CFG.host.name
+      @nodes = nodes
+
+      common_labels = [
+        :address_family,
+        :country_code,
+        :ring_source,
+        :ring_target,
+        :target_status,
+      ]
+
+      @registry = ::Prometheus::Client.registry
+      @state_metric = ::Prometheus::Client::Gauge.new(STATE_METRIC_NAME, docstring: "State of NLNOG Ring Node: 0 is down, 1 is up.", labels: common_labels)
+      @latency_metric = ::Prometheus::Client::Gauge.new(LATENCY_METRIC_NAME, docstring: "Latency of NLNOG Ring Node in microseconds.", labels: common_labels)
+      @refreshed_metric = ::Prometheus::Client::Gauge.new(NODE_LAST_CHECKED_METRIC_NAME, docstring: "Last timestamp this NLNOG Ring Node was checked.", labels: common_labels)
+
+      @registry.register(@state_metric)
+      @registry.register(@latency_metric)
+      @registry.register(@refreshed_metric)
+
+
+      accesslog = []
+      if CFG.prometheus.accesslog?
+        accesslog = [
+          [AccessLog, WEBrick::AccessLog::COMMON_LOG_FORMAT]
+        ]
+      end
+      @server =
+        WEBrick::HTTPServer.new(
+          Port: port,
+          BindAddress: bind,
+          Logger: Log,
+          AccessLog: accesslog
+        )
+      start
+    end
+
+    def start
+      @server.mount_proc "/" do |req, res|
+        res["Content-Type"] = "text/plain; charset=utf-8"
+        if req.path == "/metrics"
+          res.status = 200
+          res.body = export_metrics
+        else
+          res.status = 404
+          res.body = "Not Found"
+        end
+      end
+
+      @runner ||=
+        Thread.start do
+          begin
+            @server.start
+          rescue => e
+            Log.error "Failed to start prometheus metrics web on port #{@port}: #{e}"
+          end
+        end
+    end
+
+    def export_metrics
+      ::Prometheus::Client::Formats::Text.marshal(@registry)
+    end
+  end
+
+end
+end

ring-sqa.gemspec

@@ -16,9 +16,11 @@ Gem::Specification.new do |s|
   s.required_ruby_version =                   '>= 1.9.3'
   s.add_runtime_dependency 'slop',            '~> 3.5'
   s.add_runtime_dependency 'rb-inotify',      '~> 0.9'
-  s.add_runtime_dependency 'sequel',          '~> 4.12'
+  s.add_runtime_dependency 'sequel',          '~> 4.12', '< 4.42.0'
   s.add_runtime_dependency 'sqlite3',         '~> 1.3'
   s.add_runtime_dependency 'asetus',          '~> 0.3'
   s.add_runtime_dependency 'graphite-api',    '~> 0.1.0'
   s.add_runtime_dependency 'influxdb-client', '>= 0.1', '<= 3.2.0'
+  s.add_runtime_dependency 'prometheus-client', '~> 4.2'
+  s.add_runtime_dependency 'webrick',         '~> 1.9'
 end