2015-12-31 18:33:35 -05:00
|
|
|
# frozen_string_literal: true
|
2012-02-08 20:04:02 -05:00
|
|
|
require 'sidekiq/util'
|
2015-10-07 15:21:10 -04:00
|
|
|
require 'sidekiq/fetch'
|
2017-01-17 17:58:08 -05:00
|
|
|
require 'sidekiq/job_logger'
|
|
|
|
require 'sidekiq/job_retry'
|
2015-10-07 12:42:10 -04:00
|
|
|
require 'thread'
|
2015-10-08 23:59:05 -04:00
|
|
|
require 'concurrent/map'
|
|
|
|
require 'concurrent/atomic/atomic_fixnum'
|
2012-02-08 20:04:02 -05:00
|
|
|
|
2012-01-25 16:32:51 -05:00
|
|
|
module Sidekiq
|
2012-06-13 00:55:06 -04:00
|
|
|
##
|
2015-10-09 18:33:42 -04:00
|
|
|
# The Processor is a standalone thread which:
|
|
|
|
#
|
|
|
|
# 1. fetches a job from Redis
|
|
|
|
# 2. executes the job
|
|
|
|
# a. instantiate the Worker
|
|
|
|
# b. run the middleware chain
|
|
|
|
# c. call #perform
|
|
|
|
#
|
|
|
|
# A Processor can exit due to shutdown (processor_stopped)
|
|
|
|
# or due to an error during job execution (processor_died)
|
|
|
|
#
|
|
|
|
# If an error occurs in the job execution, the
|
|
|
|
# Processor calls the Manager to create a new one
|
|
|
|
# to replace itself and exits.
|
|
|
|
#
|
2012-01-25 16:32:51 -05:00
|
|
|
class Processor
|
2015-10-07 12:40:15 -04:00
|
|
|
|
2012-01-26 15:45:04 -05:00
|
|
|
include Util
|
2012-01-25 16:32:51 -05:00
|
|
|
|
2015-10-05 13:13:00 -04:00
|
|
|
attr_reader :thread
|
2015-10-08 12:48:28 -04:00
|
|
|
attr_reader :job
|
2013-03-27 01:56:49 -04:00
|
|
|
|
2015-10-08 12:37:37 -04:00
|
|
|
def initialize(mgr)
|
2015-10-05 13:13:00 -04:00
|
|
|
@mgr = mgr
|
2015-10-07 15:21:10 -04:00
|
|
|
@down = false
|
2015-10-05 13:13:00 -04:00
|
|
|
@done = false
|
2015-10-07 15:21:10 -04:00
|
|
|
@job = nil
|
2015-10-09 18:33:42 -04:00
|
|
|
@thread = nil
|
2015-10-08 12:37:37 -04:00
|
|
|
@strategy = (mgr.options[:fetch] || Sidekiq::BasicFetch).new(mgr.options)
|
2016-02-01 18:59:20 -05:00
|
|
|
@reloader = Sidekiq.options[:reloader]
|
2017-06-07 13:26:18 -04:00
|
|
|
@logging = (mgr.options[:job_logger] || Sidekiq::JobLogger).new
|
2017-01-17 17:58:08 -05:00
|
|
|
@retrier = Sidekiq::JobRetry.new
|
2015-10-05 13:13:00 -04:00
|
|
|
end
|
|
|
|
|
|
|
|
def terminate(wait=false)
|
|
|
|
@done = true
|
2015-10-08 12:37:37 -04:00
|
|
|
return if !@thread
|
2015-10-06 15:43:01 -04:00
|
|
|
@thread.value if wait
|
|
|
|
end
|
|
|
|
|
|
|
|
def kill(wait=false)
|
2015-10-08 12:37:37 -04:00
|
|
|
@done = true
|
|
|
|
return if !@thread
|
2015-10-05 13:13:00 -04:00
|
|
|
# unlike the other actors, terminate does not wait
|
|
|
|
# for the thread to finish because we don't know how
|
|
|
|
# long the job will take to finish. Instead we
|
|
|
|
# provide a `kill` method to call after the shutdown
|
|
|
|
# timeout passes.
|
2015-10-06 15:43:01 -04:00
|
|
|
@thread.raise ::Sidekiq::Shutdown
|
2015-10-05 13:13:00 -04:00
|
|
|
@thread.value if wait
|
|
|
|
end
|
|
|
|
|
2015-10-06 15:43:01 -04:00
|
|
|
def start
|
|
|
|
@thread ||= safe_thread("processor", &method(:run))
|
2012-02-04 19:53:09 -05:00
|
|
|
end
|
|
|
|
|
2015-10-06 15:43:01 -04:00
|
|
|
private unless $TESTING
|
2015-10-05 13:13:00 -04:00
|
|
|
|
|
|
|
def run
|
|
|
|
begin
|
|
|
|
while !@done
|
2015-10-08 12:37:37 -04:00
|
|
|
process_one
|
2015-10-05 13:13:00 -04:00
|
|
|
end
|
2015-10-09 00:50:45 -04:00
|
|
|
@mgr.processor_stopped(self)
|
2015-10-09 18:33:42 -04:00
|
|
|
rescue Sidekiq::Shutdown
|
|
|
|
@mgr.processor_stopped(self)
|
2015-10-05 13:13:00 -04:00
|
|
|
rescue Exception => ex
|
|
|
|
@mgr.processor_died(self, ex)
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2015-10-08 12:37:37 -04:00
|
|
|
def process_one
|
2015-10-08 12:48:28 -04:00
|
|
|
@job = fetch
|
|
|
|
process(@job) if @job
|
|
|
|
@job = nil
|
2015-10-08 12:37:37 -04:00
|
|
|
end
|
|
|
|
|
2015-10-07 15:21:10 -04:00
|
|
|
def get_one
|
|
|
|
begin
|
|
|
|
work = @strategy.retrieve_work
|
2015-10-08 12:48:28 -04:00
|
|
|
(logger.info { "Redis is online, #{Time.now - @down} sec downtime" }; @down = nil) if @down
|
2015-10-07 15:21:10 -04:00
|
|
|
work
|
2015-10-08 12:37:37 -04:00
|
|
|
rescue Sidekiq::Shutdown
|
2015-10-07 15:21:10 -04:00
|
|
|
rescue => ex
|
|
|
|
handle_fetch_exception(ex)
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
|
|
|
def fetch
|
|
|
|
j = get_one
|
|
|
|
if j && @done
|
|
|
|
j.requeue
|
|
|
|
nil
|
|
|
|
else
|
|
|
|
j
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
|
|
|
def handle_fetch_exception(ex)
|
|
|
|
if !@down
|
|
|
|
@down = Time.now
|
2017-11-29 19:23:49 -05:00
|
|
|
logger.error("Error fetching job: #{ex}")
|
2017-11-30 13:12:39 -05:00
|
|
|
handle_exception(ex)
|
2015-10-07 15:21:10 -04:00
|
|
|
end
|
|
|
|
sleep(1)
|
2016-03-01 11:57:38 -05:00
|
|
|
nil
|
2015-10-07 15:21:10 -04:00
|
|
|
end
|
|
|
|
|
2017-01-17 17:58:08 -05:00
|
|
|
def dispatch(job_hash, queue)
|
|
|
|
# since middleware can mutate the job hash
|
|
|
|
# we clone here so we report the original
|
|
|
|
# job structure to the Web UI
|
|
|
|
pristine = cloned(job_hash)
|
|
|
|
|
2017-05-10 14:56:36 -04:00
|
|
|
Sidekiq::Logging.with_job_hash_context(job_hash) do
|
2017-08-11 16:33:09 -04:00
|
|
|
@retrier.global(pristine, queue) do
|
2017-03-16 14:42:02 -04:00
|
|
|
@logging.call(job_hash, queue) do
|
|
|
|
stats(pristine, queue) do
|
|
|
|
# Rails 5 requires a Reloader to wrap code execution. In order to
|
|
|
|
# constantize the worker and instantiate an instance, we have to call
|
|
|
|
# the Reloader. It handles code loading, db connection management, etc.
|
|
|
|
# Effectively this block denotes a "unit of work" to Rails.
|
|
|
|
@reloader.call do
|
2018-02-16 16:01:25 -05:00
|
|
|
klass = constantize(job_hash['class'])
|
2017-03-16 14:42:02 -04:00
|
|
|
worker = klass.new
|
2018-02-16 16:01:25 -05:00
|
|
|
worker.jid = job_hash['jid']
|
2017-08-11 16:33:09 -04:00
|
|
|
@retrier.local(worker, pristine, queue) do
|
2017-03-16 14:42:02 -04:00
|
|
|
yield worker
|
|
|
|
end
|
2017-01-17 17:58:08 -05:00
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2015-10-06 15:43:01 -04:00
|
|
|
def process(work)
|
2015-10-23 18:05:50 -04:00
|
|
|
jobstr = work.job
|
2013-01-06 00:17:08 -05:00
|
|
|
queue = work.queue_name
|
2012-06-29 23:37:45 -04:00
|
|
|
|
2016-10-28 23:41:06 -04:00
|
|
|
ack = false
|
|
|
|
begin
|
2017-03-14 23:06:04 -04:00
|
|
|
# Treat malformed JSON as a special case: job goes straight to the morgue.
|
2017-01-17 17:58:08 -05:00
|
|
|
job_hash = nil
|
|
|
|
begin
|
|
|
|
job_hash = Sidekiq.load_json(jobstr)
|
|
|
|
rescue => ex
|
2017-03-14 23:06:04 -04:00
|
|
|
handle_exception(ex, { :context => "Invalid JSON for job", :jobstr => jobstr })
|
2018-01-15 13:39:32 -05:00
|
|
|
# we can't notify because the job isn't a valid hash payload.
|
|
|
|
DeadSet.new.kill(jobstr, notify_failure: false)
|
2016-02-01 18:59:20 -05:00
|
|
|
ack = true
|
2017-01-17 17:58:08 -05:00
|
|
|
raise
|
|
|
|
end
|
|
|
|
|
|
|
|
ack = true
|
|
|
|
dispatch(job_hash, queue) do |worker|
|
|
|
|
Sidekiq.server_middleware.invoke(worker, job_hash, queue) do
|
2018-02-16 16:01:25 -05:00
|
|
|
execute_job(worker, cloned(job_hash['args']))
|
2017-01-17 17:58:08 -05:00
|
|
|
end
|
2012-02-11 02:16:12 -05:00
|
|
|
end
|
2016-10-28 23:41:06 -04:00
|
|
|
rescue Sidekiq::Shutdown
|
|
|
|
# Had to force kill this job because it didn't finish
|
|
|
|
# within the timeout. Don't acknowledge the work since
|
|
|
|
# we didn't properly finish it.
|
|
|
|
ack = false
|
|
|
|
rescue Exception => ex
|
2017-02-15 14:13:10 -05:00
|
|
|
e = ex.is_a?(::Sidekiq::JobRetry::Skip) && ex.cause ? ex.cause : ex
|
|
|
|
handle_exception(e, { :context => "Job raised exception", :job => job_hash, :jobstr => jobstr })
|
|
|
|
raise e
|
2016-10-28 23:41:06 -04:00
|
|
|
ensure
|
|
|
|
work.acknowledge if ack
|
2012-02-04 19:53:09 -05:00
|
|
|
end
|
2013-06-08 01:15:13 -04:00
|
|
|
end
|
|
|
|
|
2014-09-09 21:57:39 -04:00
|
|
|
def execute_job(worker, cloned_args)
|
|
|
|
worker.perform(*cloned_args)
|
|
|
|
end
|
|
|
|
|
2015-10-07 17:27:47 -04:00
|
|
|
WORKER_STATE = Concurrent::Map.new
|
|
|
|
PROCESSED = Concurrent::AtomicFixnum.new
|
|
|
|
FAILURE = Concurrent::AtomicFixnum.new
|
|
|
|
|
2017-01-17 17:58:08 -05:00
|
|
|
def stats(job_hash, queue)
|
2018-01-10 13:04:09 -05:00
|
|
|
tid = Sidekiq::Logging.tid
|
2017-01-17 17:58:08 -05:00
|
|
|
WORKER_STATE[tid] = {:queue => queue, :payload => job_hash, :run_at => Time.now.to_i }
|
2012-02-11 02:16:12 -05:00
|
|
|
|
|
|
|
begin
|
|
|
|
yield
|
2012-05-12 16:23:23 -04:00
|
|
|
rescue Exception
|
2015-10-07 17:27:47 -04:00
|
|
|
FAILURE.increment
|
2012-02-11 02:16:12 -05:00
|
|
|
raise
|
|
|
|
ensure
|
2015-10-07 17:27:47 -04:00
|
|
|
WORKER_STATE.delete(tid)
|
|
|
|
PROCESSED.increment
|
2012-02-11 02:16:12 -05:00
|
|
|
end
|
2012-08-04 15:11:46 -04:00
|
|
|
end
|
2012-02-11 02:16:12 -05:00
|
|
|
|
2014-01-27 23:29:19 -05:00
|
|
|
# Deep clone the arguments passed to the worker so that if
|
2015-10-23 18:05:50 -04:00
|
|
|
# the job fails, what is pushed back onto Redis hasn't
|
2012-08-04 15:11:46 -04:00
|
|
|
# been mutated by the worker.
|
2017-01-17 17:58:08 -05:00
|
|
|
def cloned(thing)
|
|
|
|
Marshal.load(Marshal.dump(thing))
|
2012-02-11 02:16:12 -05:00
|
|
|
end
|
2014-02-24 14:58:35 -05:00
|
|
|
|
2017-05-14 23:58:44 -04:00
|
|
|
def constantize(str)
|
|
|
|
names = str.split('::')
|
|
|
|
names.shift if names.empty? || names.first.empty?
|
|
|
|
|
|
|
|
names.inject(Object) do |constant, name|
|
2017-12-20 06:53:31 -05:00
|
|
|
# the false flag limits search for name to under the constant namespace
|
|
|
|
# which mimics Rails' behaviour
|
|
|
|
constant.const_defined?(name, false) ? constant.const_get(name, false) : constant.const_missing(name)
|
2017-05-14 23:58:44 -04:00
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2012-01-25 16:32:51 -05:00
|
|
|
end
|
|
|
|
end
|