2014-02-03 19:23:00 +00:00
|
|
|
# Copyright (C) 2012-2014 Zammad Foundation, http://zammad-foundation.org/
|
2015-05-04 18:58:28 +00:00
|
|
|
# rubocop:disable Rails/Output
|
2013-03-05 08:38:58 +00:00
|
|
|
class Scheduler < ApplicationModel
|
|
|
|
|
2015-05-20 23:07:13 +00:00
|
|
|
# rubocop:disable Style/ClassVars
|
|
|
|
@@jobs_started = {}
|
|
|
|
# rubocop:enable Style/ClassVars
|
|
|
|
|
|
|
|
# start threads
|
|
|
|
def self.threads
|
2013-03-05 08:38:58 +00:00
|
|
|
|
|
|
|
Thread.abort_on_exception = true
|
2013-06-12 15:59:58 +00:00
|
|
|
|
2015-05-20 23:07:13 +00:00
|
|
|
# start worker for background jobs
|
|
|
|
worker
|
|
|
|
|
|
|
|
# start loop to execute scheduler jobs
|
2015-05-05 14:10:06 +00:00
|
|
|
loop do
|
2015-05-20 23:07:13 +00:00
|
|
|
logger.info 'Scheduler running...'
|
2013-05-07 20:45:00 +00:00
|
|
|
|
2014-03-27 13:07:57 +00:00
|
|
|
# reconnect in case db connection is lost
|
|
|
|
begin
|
|
|
|
ActiveRecord::Base.connection.reconnect!
|
|
|
|
rescue => e
|
2015-07-03 15:18:01 +00:00
|
|
|
logger.error "Can't reconnect to database #{e.inspect}"
|
2014-03-27 13:07:57 +00:00
|
|
|
end
|
|
|
|
|
2013-05-07 20:45:00 +00:00
|
|
|
# read/load jobs and check if it is alredy started
|
2015-06-30 22:26:02 +00:00
|
|
|
jobs = Scheduler.where( 'active = ?', true ).order('prio ASC')
|
2013-05-07 20:45:00 +00:00
|
|
|
jobs.each {|job|
|
2015-05-20 23:07:13 +00:00
|
|
|
|
|
|
|
# ignore job is still running
|
|
|
|
next if @@jobs_started[ job.id ]
|
|
|
|
|
|
|
|
# check job.last_run
|
|
|
|
next if job.last_run && job.period && job.last_run > ( Time.zone.now - job.period )
|
|
|
|
|
|
|
|
# run job as own thread
|
|
|
|
@@jobs_started[ job.id ] = true
|
2015-06-30 22:26:02 +00:00
|
|
|
start_job(job)
|
|
|
|
sleep 10
|
2013-05-07 20:45:00 +00:00
|
|
|
}
|
2015-05-20 23:07:13 +00:00
|
|
|
sleep 60
|
2013-03-05 08:38:58 +00:00
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2015-05-20 23:07:13 +00:00
|
|
|
def self.start_job( job )
|
2013-03-05 08:38:58 +00:00
|
|
|
|
|
|
|
Thread.new {
|
2015-05-20 23:07:13 +00:00
|
|
|
|
2015-05-21 06:37:21 +00:00
|
|
|
logger.info "Started job thread for '#{job.name}' (#{job.method})..."
|
2015-05-20 23:07:13 +00:00
|
|
|
|
|
|
|
# start loop for periods under 5 minutes
|
|
|
|
if job.period && job.period <= 300
|
2015-05-05 14:10:06 +00:00
|
|
|
loop do
|
2015-06-30 22:26:02 +00:00
|
|
|
_start_job(job)
|
|
|
|
job = Scheduler.lookup(id: job.id)
|
2013-05-07 20:45:00 +00:00
|
|
|
|
|
|
|
# exit is job got deleted
|
2013-03-05 08:38:58 +00:00
|
|
|
break if !job
|
2013-05-07 20:45:00 +00:00
|
|
|
|
|
|
|
# exit if job is not active anymore
|
2013-03-05 08:38:58 +00:00
|
|
|
break if !job.active
|
2013-05-07 20:45:00 +00:00
|
|
|
|
|
|
|
# exit if there is no loop period defined
|
2013-03-05 08:38:58 +00:00
|
|
|
break if !job.period
|
2013-05-07 20:45:00 +00:00
|
|
|
|
|
|
|
# wait until next run
|
2013-03-05 08:38:58 +00:00
|
|
|
sleep job.period
|
|
|
|
end
|
|
|
|
else
|
2015-06-30 22:26:02 +00:00
|
|
|
_start_job(job)
|
2013-03-05 08:38:58 +00:00
|
|
|
end
|
|
|
|
job.pid = ''
|
|
|
|
job.save
|
2014-10-26 12:17:00 +00:00
|
|
|
logger.info " ...stopped thread for '#{job.method}'"
|
2015-02-16 12:41:46 +00:00
|
|
|
ActiveRecord::Base.connection.close
|
2015-05-20 23:07:13 +00:00
|
|
|
|
|
|
|
# release thread lock
|
|
|
|
@@jobs_started[ job.id ] = false
|
2013-03-05 08:38:58 +00:00
|
|
|
}
|
|
|
|
end
|
|
|
|
|
2015-05-20 23:07:13 +00:00
|
|
|
def self._start_job( job, try_count = 0, try_run_time = Time.zone.now )
|
2014-03-27 13:07:57 +00:00
|
|
|
begin
|
2015-05-08 10:20:33 +00:00
|
|
|
job.last_run = Time.zone.now
|
2015-05-20 23:07:13 +00:00
|
|
|
job.pid = Thread.current.object_id
|
2014-03-27 13:07:57 +00:00
|
|
|
job.save
|
2015-05-20 23:07:13 +00:00
|
|
|
logger.info "execute #{job.method} (try_count #{try_count})..."
|
2015-05-07 12:33:04 +00:00
|
|
|
eval job.method() # rubocop:disable Lint/Eval
|
2014-03-27 13:07:57 +00:00
|
|
|
rescue => e
|
2015-07-03 15:18:01 +00:00
|
|
|
logger.error "execute #{job.method} (try_count #{try_count}) exited with error #{e.inspect}"
|
2014-03-27 13:07:57 +00:00
|
|
|
|
|
|
|
# reconnect in case db connection is lost
|
|
|
|
begin
|
|
|
|
ActiveRecord::Base.connection.reconnect!
|
|
|
|
rescue => e
|
2015-07-03 15:18:01 +00:00
|
|
|
logger.error "Can't reconnect to database #{e.inspect}"
|
2014-03-27 13:07:57 +00:00
|
|
|
end
|
|
|
|
|
|
|
|
try_run_max = 10
|
|
|
|
try_count += 1
|
|
|
|
|
|
|
|
# reset error counter if to old
|
2015-05-08 10:20:33 +00:00
|
|
|
if try_run_time + ( 60 * 5 ) < Time.zone.now
|
2014-03-27 13:07:57 +00:00
|
|
|
try_count = 0
|
2014-10-26 12:17:00 +00:00
|
|
|
end
|
2015-05-08 10:20:33 +00:00
|
|
|
try_run_time = Time.zone.now
|
2014-03-27 13:07:57 +00:00
|
|
|
|
|
|
|
# restart job again
|
|
|
|
if try_run_max > try_count
|
2015-05-20 23:07:13 +00:00
|
|
|
_start_job( job, try_count, try_run_time)
|
2014-03-27 13:07:57 +00:00
|
|
|
else
|
2015-05-20 23:07:13 +00:00
|
|
|
raise "STOP thread for #{job.method} after #{try_count} tries"
|
2014-03-27 13:07:57 +00:00
|
|
|
end
|
|
|
|
end
|
2013-03-05 08:38:58 +00:00
|
|
|
end
|
2013-08-04 21:56:02 +00:00
|
|
|
|
2014-02-03 18:26:41 +00:00
|
|
|
def self.worker
|
2015-05-20 23:07:13 +00:00
|
|
|
wait = 8
|
2014-02-03 18:26:41 +00:00
|
|
|
|
2015-05-20 23:07:13 +00:00
|
|
|
Thread.new {
|
|
|
|
sleep wait
|
2014-02-03 18:26:41 +00:00
|
|
|
|
2015-05-21 06:37:21 +00:00
|
|
|
logger.info "Starting worker thread #{Delayed::Job}"
|
2014-02-03 18:26:41 +00:00
|
|
|
|
2015-05-20 23:07:13 +00:00
|
|
|
loop do
|
|
|
|
result = nil
|
2014-02-03 18:26:41 +00:00
|
|
|
|
2015-05-20 23:07:13 +00:00
|
|
|
realtime = Benchmark.realtime do
|
|
|
|
result = Delayed::Worker.new.work_off
|
|
|
|
end
|
|
|
|
|
|
|
|
count = result.sum
|
|
|
|
|
|
|
|
if count.zero?
|
|
|
|
sleep wait
|
|
|
|
logger.debug '*** worker thread loop'
|
|
|
|
else
|
|
|
|
format "*** #{count} jobs processed at %.4f j/s, %d failed ...\n", count / realtime, result.last
|
|
|
|
end
|
2014-02-03 18:26:41 +00:00
|
|
|
end
|
2015-05-20 23:07:13 +00:00
|
|
|
|
|
|
|
logger.info ' ...stopped worker thread'
|
|
|
|
ActiveRecord::Base.connection.close
|
|
|
|
}
|
|
|
|
|
2014-02-03 18:26:41 +00:00
|
|
|
end
|
|
|
|
|
2013-08-04 21:56:02 +00:00
|
|
|
def self.check( name, time_warning = 10, time_critical = 20 )
|
2015-05-08 10:20:33 +00:00
|
|
|
time_warning_time = Time.zone.now - time_warning.minutes
|
|
|
|
time_critical_time = Time.zone.now - time_critical.minutes
|
2015-05-07 10:15:40 +00:00
|
|
|
scheduler = Scheduler.find_by( name: name )
|
2013-08-04 21:56:02 +00:00
|
|
|
if !scheduler
|
|
|
|
puts "CRITICAL - no such scheduler jobs '#{name}'"
|
|
|
|
return true
|
|
|
|
end
|
2015-05-04 18:58:28 +00:00
|
|
|
logger.debug scheduler.inspect
|
2013-08-04 21:56:02 +00:00
|
|
|
if !scheduler.last_run
|
|
|
|
puts "CRITICAL - scheduler jobs never started '#{name}'"
|
|
|
|
exit 2
|
|
|
|
end
|
|
|
|
if scheduler.last_run < time_critical_time
|
2015-05-06 09:30:39 +00:00
|
|
|
puts "CRITICAL - scheduler jobs was not running in last '#{time_critical}' minutes - last run at '#{scheduler.last_run}' '#{name}'"
|
2013-08-04 21:56:02 +00:00
|
|
|
exit 2
|
|
|
|
end
|
|
|
|
if scheduler.last_run < time_warning_time
|
2015-05-06 09:30:39 +00:00
|
|
|
puts "CRITICAL - scheduler jobs was not running in last '#{time_warning}' minutes - last run at '#{scheduler.last_run}' '#{name}'"
|
2013-08-04 21:56:02 +00:00
|
|
|
exit 2
|
|
|
|
end
|
2015-05-06 09:30:39 +00:00
|
|
|
puts "ok - scheduler jobs was running at '#{scheduler.last_run}' '#{name}'"
|
2013-08-04 21:56:02 +00:00
|
|
|
exit 0
|
|
|
|
end
|
2013-06-12 15:59:58 +00:00
|
|
|
end
|