class Roundhouse::Scheduled::Poller
The Poller
checks Redis every N seconds for jobs in the retry or scheduled set have passed their timestamp and should be enqueued. If so, it just pops the job back onto its original queue so the workers can pick it up like any other job.
Constants
- INITIAL_WAIT
Public Class Methods
# File lib/roundhouse/scheduled.rb, line 46 def initialize @enq = (Roundhouse.options[:scheduled_enq] || Roundhouse::Scheduled::Enq).new end
Public Instance Methods
# File lib/roundhouse/scheduled.rb, line 50 def poll(first_time=false) watchdog('scheduling poller thread died!') do initial_wait if first_time begin @enq.enqueue_jobs rescue => ex # Most likely a problem with redis networking. # Punt and try again at the next interval logger.error ex.message logger.error ex.backtrace.first end after(random_poll_interval) { poll } end end
Private Instance Methods
# File lib/roundhouse/scheduled.rb, line 100 def initial_wait begin # Have all processes sleep between 5-15 seconds. 10 seconds # to give time for the heartbeat to register (if the poll interval is going to be calculated by the number # of workers), and 5 random seconds to ensure they don't all hit Redis at the same time. sleep(INITIAL_WAIT) unless Roundhouse.options[:poll_interval_average] sleep(5 * rand) rescue Celluloid::TaskTerminated # Hit Ctrl-C when Roundhouse is finished booting and we have a chance # to get here. end end
We do our best to tune the poll interval to the size of the active Roundhouse
cluster. If you have 30 processes and poll every 15 seconds, that means one Roundhouse
is checking Redis every 0.5 seconds - way too often for most people and really bad if the retry or scheduled sets are large.
Instead try to avoid polling more than once every 15 seconds. If you have 30 Roundhouse
processes, we’ll poll every 30 * 15 or 450 seconds. To keep things statistically random, we’ll sleep a random amount between 225 and 675 seconds for each poll or 450 seconds on average. Otherwise restarting all your Roundhouse
processes at the same time will lead to them all polling at the same time: the thundering herd problem.
We only do this if poll_interval is unset (the default).
# File lib/roundhouse/scheduled.rb, line 87 def poll_interval_average Roundhouse.options[:poll_interval_average] ||= scaled_poll_interval end
Calculates a random interval that is ±50% the desired average.
# File lib/roundhouse/scheduled.rb, line 70 def random_poll_interval poll_interval_average * rand + poll_interval_average.to_f / 2 end
Calculates an average poll interval based on the number of known Roundhouse
processes. This minimizes a single point of failure by dispersing check-ins but without taxing Redis if you run many Roundhouse
processes.
# File lib/roundhouse/scheduled.rb, line 94 def scaled_poll_interval pcount = Roundhouse::ProcessSet.new.size pcount = 1 if pcount == 0 pcount * Roundhouse.options[:average_scheduled_poll_interval] end