#!/usr/bin/env ruby require 'log4r' require 'log4r/outputter/fileoutputter' require 'flapjack' require 'flapjack/filters/acknowledgement' require 'flapjack/filters/ok' require 'flapjack/filters/scheduled_maintenance' require 'flapjack/filters/unscheduled_maintenance' require 'flapjack/filters/detect_mass_client_failures' require 'flapjack/filters/delays' require 'flapjack/data/contact' require 'flapjack/data/entity_check' require 'flapjack/data/notification' require 'flapjack/data/event' require 'flapjack/notification/sms' require 'flapjack/notification/email' require 'flapjack/pikelet' require 'flapjack/redis_pool' module Flapjack class Executive include Flapjack::GenericPikelet alias_method :generic_bootstrap, :bootstrap alias_method :generic_cleanup, :cleanup def bootstrap(opts = {}) generic_bootstrap(opts) @redis = Flapjack::RedisPool.new(:config => opts[:redis_config], :size => 1) @queues = {:email => @config['email_queue'], :sms => @config['sms_queue'], :jabber => @config['jabber_queue'], :pagerduty => @config['pagerduty_queue']} notifylog = @config['notification_log_file'] || 'log/notify.log' @notifylog = Log4r::Logger.new("executive") @notifylog.add(Log4r::FileOutputter.new("notifylog", :filename => notifylog)) # FIXME: Put loading filters into separate method # FIXME: should we make the filters more configurable by the end user? options = { :log => @logger, :persistence => @redis } @filters = [] @filters << Flapjack::Filters::Ok.new(options) @filters << Flapjack::Filters::ScheduledMaintenance.new(options) @filters << Flapjack::Filters::UnscheduledMaintenance.new(options) @filters << Flapjack::Filters::DetectMassClientFailures.new(options) @filters << Flapjack::Filters::Delays.new(options) @filters << Flapjack::Filters::Acknowledgement.new(options) @boot_time = Time.now @fqdn = `/bin/hostname -f`.chomp @pid = Process.pid @instance_id = "#{@fqdn}:#{@pid}" # FIXME: all of the below keys assume there is only ever one executive running; # we could generate a fuid and save it to disk, and prepend it from that # point on... # TODO unset on exit? @redis.set('boot_time', @boot_time.to_i) # FIXME: add an administrative function to reset all event counters if @redis.hget('event_counters', 'all').nil? @redis.hset('event_counters', 'all', 0) @redis.hset('event_counters', 'ok', 0) @redis.hset('event_counters', 'failure', 0) @redis.hset('event_counters', 'action', 0) end @redis.zadd('executive_instances', @boot_time.to_i, @instance_id) @redis.hset("event_counters:#{@instance_id}", 'all', 0) @redis.hset("event_counters:#{@instance_id}", 'ok', 0) @redis.hset("event_counters:#{@instance_id}", 'failure', 0) @redis.hset("event_counters:#{@instance_id}", 'action', 0) end def cleanup @redis.empty! if @redis generic_cleanup end def main @logger.info("Booting main loop.") until should_quit? && @received_shutdown @logger.info("Waiting for event...") event = Flapjack::Data::Event.next(:redis => @redis) process_event(event) unless event.nil? end @logger.info("Exiting main loop.") end # this must use a separate connection to the main Executive one, as it's running # from a different fiber while the main one is blocking. def add_shutdown_event(opts = {}) return unless redis = opts[:redis] redis.rpush('events', JSON.generate('type' => 'shutdown', 'host' => '', 'service' => '', 'state' => '')) end private def process_event(event) pending = Flapjack::Data::Event.pending_count(:redis => @redis) @logger.debug("#{pending} events waiting on the queue") @logger.debug("Raw event received: #{event.inspect}") time_at = event.time time_at_str = time_at ? ", #{Time.at(time_at).to_s}" : '' @logger.info("Processing Event: #{event.id}, #{event.type}, #{event.state}, #{event.summary}#{time_at_str}") entity_check = (event.type == 'shutdown') ? nil : Flapjack::Data::EntityCheck.for_event_id(event.id, :redis => @redis) result = update_keys(event, entity_check) return if result[:shutdown] skip_filters = result[:skip_filters] blocker = @filters.find {|filter| filter.block?(event) } unless skip_filters if skip_filters @logger.info("#{Time.now}: Not sending notifications for event #{event.id} because filtering was skipped") return end if blocker blocker_names = [ blocker.name ] @logger.info("#{Time.now}: Not sending notifications for event #{event.id} because these filters blocked: #{blocker_names.join(', ')}") return end @logger.info("#{Time.now}: Sending notifications for event #{event.id}") send_notification_messages(event, entity_check) end def update_keys(event, entity_check) result = { :skip_filters => false } timestamp = Time.now.to_i @event_count = @redis.hincrby('event_counters', 'all', 1) @redis.hincrby("event_counters:#{@instance_id}", 'all', 1) # FIXME skip if entity_check.nil? # FIXME: validate that the event is sane before we ever get here # FIXME: create an event if there is dodgy data case event.type # Service events represent changes in state on monitored systems when 'service' # Track when we last saw an event for a particular entity:check pair entity_check.last_update = timestamp if event.ok? @redis.hincrby('event_counters', 'ok', 1) @redis.hincrby("event_counters:#{@instance_id}", 'ok', 1) elsif event.failure? @redis.hincrby('event_counters', 'failure', 1) @redis.hincrby("event_counters:#{@instance_id}", 'failure', 1) @redis.hset('unacknowledged_failures', @event_count, event.id) end event.previous_state = entity_check.state @logger.info("No previous state for event #{event.id}") if event.previous_state.nil? # If there is a state change, update record with: the time, the new state if event.state != event.previous_state entity_check.update_state(event.state, :timestamp => timestamp, :summary => event.summary, :client => event.client, :count => @event_count) end # No state change, and event is ok, so no need to run through filters # OR # If the service event's state is ok and there was no previous state, don't alert. # This stops new checks from alerting as "recovery" after they have been added. if !event.previous_state && event.ok? @logger.debug("setting skip_filters to true because there was no previous state and event is ok") result[:skip_filters] = true end entity_check.update_current_scheduled_maintenance # Action events represent human or automated interaction with Flapjack when 'action' # When an action event is processed, store the event. @redis.hset(event.id + ':actions', timestamp, event.state) @redis.hincrby('event_counters', 'action', 1) @redis.hincrby("event_counters:#{@instance_id}", 'action', 1) if event.acknowledgement? && event.acknowledgement_id @redis.hdel('unacknowledged_failures', event.acknowledgement_id) end when 'shutdown' # should this be logged as an action instead? being minimally invasive for now result[:shutdown] = @received_shutdown = true end result end # takes an event for which a notification needs to be generated, works out the type of # notification, updates the notification history in redis, sends the notifications def send_notification_messages(event, entity_check) timestamp = Time.now.to_i notification_type = 'unknown' case event.type when 'service' case event.state when 'ok', 'unknown' notification_type = 'recovery' when 'warning', 'critical' notification_type = 'problem' end when 'action' case event.state when 'acknowledgement' notification_type = 'acknowledgement' when 'test_notifications' notification_type = 'test' end end @redis.set("#{event.id}:last_#{notification_type}_notification", timestamp) @redis.rpush("#{event.id}:#{notification_type}_notifications", timestamp) @logger.debug("Notification of type #{notification_type} is being generated for #{event.id}.") contacts = entity_check.contacts if contacts.empty? @notifylog.info("#{Time.now.to_s} | #{event.id} | #{notification_type} | NO CONTACTS") return end notification = Flapjack::Data::Notification.for_event(event, :type => notification_type) notification.messages(:contacts => contacts).each do |msg| media_type = msg.medium.to_sym @notifylog.info("#{Time.now.to_s} | #{event.id} | " + "#{notification_type} | #{msg.contact.id} | #{media_type.to_s} | #{msg.address}") unless @queues[media_type] # TODO log error next end contents = msg.contents # TODO consider changing Resque jobs to use raw blpop like the others case media_type when :sms Resque.enqueue_to(@queues[:sms], Flapjack::Notification::Sms, contents) when :email Resque.enqueue_to(@queues[:email], Flapjack::Notification::Email, contents) when :jabber # TODO move next line up into other notif value setting above? contents['event_count'] = @event_count if @event_count @redis.rpush(@queues[:jabber], Yajl::Encoder.encode(contents)) when :pagerduty @redis.rpush(@queues[:pagerduty], Yajl::Encoder.encode(contents)) end end end end end