Job scheduler for Ruby (at, cron, in and every jobs).
It uses threads.
Note: maybe are you looking for the README of rufus-scheduler 2.x? (especially if you're using Dashing which is stuck on rufus-scheduler 2.0.24)
Quickstart:
# quickstart.rb
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
scheduler.in '3s' do
puts 'Hello... Rufus'
end
scheduler.join
#
# let the current thread join the scheduler thread
#
# (please note that this join should be removed when scheduling
# in a web application (Rails and friends) initializer)
(run with ruby quickstart.rb
)
Various forms of scheduling are supported:
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
# ...
scheduler.in '10d' do
# do something in 10 days
end
scheduler.at '2030/12/12 23:30:00' do
# do something at a given point in time
end
scheduler.every '3h' do
# do something every 3 hours
end
scheduler.every '3h10m' do
# do something every 3 hours and 10 minutes
end
scheduler.cron '5 0 * * *' do
# do something every day, five minutes after midnight
# (see "man 5 crontab" in your terminal)
end
# ...
Rufus-scheduler uses fugit for parsing time strings, et-orbi for pairing time and tzinfo timezones.
Rufus-scheduler (out of the box) is an in-process, in-memory scheduler. It uses threads.
It does not persist your schedules. When the process is gone and the scheduler instance with it, the schedules are gone.
A rufus-scheduler instance will go on scheduling while it is present among the objects in a Ruby process. To make it stop scheduling you have to call its #shutdown
method.
Please note: rufus-scheduler is not a cron replacement.
- ruby-clock - a clock process / job scheduler for Ruby
- Puma-Rufus-Scheduler - Puma plugin to run Rufus-Scheduler
- Whenever - let cron call back your Ruby code, trusted and reliable cron drives your schedule
- Clockwork - rufus-scheduler inspired gem
- Crono - an in-Rails cron scheduler
- PerfectSched - highly available distributed cron built on Sequel and more
It's a complete rewrite of rufus-scheduler.
There is no EventMachine-based scheduler anymore.
I'll drive you right to the tracks.
- As said, no more EventMachine-based scheduler
scheduler.every('100') {
will schedule every 100 seconds (previously, it would have been 0.1s). This aligns rufus-scheduler with Ruby'ssleep(100)
- The scheduler isn't catching the whole of Exception anymore, only StandardError
- The error_handler is #on_error (instead of #on_exception), by default it now prints the details of the error to $stderr (used to be $stdout)
- Rufus::Scheduler::TimeOutError renamed to Rufus::Scheduler::TimeoutError
- Introduction of "interval" jobs. Whereas "every" jobs are like "every 10 minutes, do this", interval jobs are like "do that, then wait for 10 minutes, then do that again, and so on"
- Introduction of a lockfile: true/filename mechanism to prevent multiple schedulers from executing
- "discard_past" is on by default. If the scheduler (its host) sleeps for 1 hour and a
every '10m'
job is on, it will trigger once at wakeup, not 6 times (discard_past was false by default in rufus-scheduler 2.x). - Introduction of Scheduler #on_pre_trigger and #on_post_trigger callback points
So you need help. People can help you, but first help them help you, and don't waste their time. Provide a complete description of the issue. If it works on A but not on B and others have to ask you: "so what is different between A and B" you are wasting everyone's time.
"hello", "please" and "thanks" are not swear words.
Go read how to report bugs effectively, twice.
Update: help_help.md might help help you.
Yes, issues can be reported in rufus-scheduler issues, I'd actually prefer bugs in there. If there is nothing wrong with rufus-scheduler, a Stack Overflow question is better.
- It doesn't work...
- I want a refund
- Passenger and rufus-scheduler
- Passenger and rufus-scheduler (2)
- Passenger in-depth spawn methods
- Passenger in-depth spawn methods (smart spawning)
- The scheduler comes up when running the Rails console or a Rake task
- The job triggers twice
- I don't get any of this, I just want it to work in my Rails application
- I get "zotime.rb:41:in `initialize': cannot determine timezone from nil"
Rufus-scheduler supports five kinds of jobs. in, at, every, interval and cron jobs.
Most of the rufus-scheduler examples show block scheduling, but it's also OK to schedule handler instances or handler classes.
In and at jobs trigger once.
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
scheduler.in '10d' do
puts "10 days reminder for review X!"
end
scheduler.at '2014/12/24 2000' do
puts "merry xmas!"
end
In jobs are scheduled with a time interval, they trigger after that time elapsed. At jobs are scheduled with a point in time, they trigger when that point in time is reached (better to choose a point in the future).
Every, interval and cron jobs trigger repeatedly.
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
scheduler.every '3h' do
puts "change the oil filter!"
end
scheduler.interval '2h' do
puts "thinking..."
puts sleep(rand * 1000)
puts "thought."
end
scheduler.cron '00 09 * * *' do
puts "it's 9am! good morning!"
end
Every jobs try hard to trigger following the frequency they were scheduled with.
Interval jobs trigger, execute and then trigger again after the interval elapsed. (every jobs time between trigger times, interval jobs time between trigger termination and the next trigger start).
Cron jobs are based on the venerable cron utility (man 5 crontab
). They trigger following a pattern given in (almost) the same language cron uses.
By default, rufus-scheduler relies on Ruby to parse strings like "Wed at 2pm". But most of the time, when calling this on Thursday, the day before is "invoked". Relying on Chronic might solve the issue.
Compare
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
scheduler.at('Wed at 2pm') do
# ... might point in the past and thus get triggered immediately
end
with
require 'chronic'
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
scheduler.at('Wed at 2pm') do
# ... will point to a time in the future and trigger appropriately
end
schedule_in, schedule_at, schedule_cron, etc will return the new Job instance.
in, at, cron will return the new Job instance's id (a String).
job_id =
scheduler.in '10d' do
# ...
end
job = scheduler.job(job_id)
# versus
job =
scheduler.schedule_in '10d' do
# ...
end
# also
job =
scheduler.in '10d', job: true do
# ...
end
Sometimes it pays to be less verbose.
The #schedule
methods schedules an at, in or cron job. It just decides based on its input. It returns the Job instance.
scheduler.schedule '10d' do; end.class
# => Rufus::Scheduler::InJob
scheduler.schedule '2013/12/12 12:30' do; end.class
# => Rufus::Scheduler::AtJob
scheduler.schedule '* * * * *' do; end.class
# => Rufus::Scheduler::CronJob
The #repeat
method schedules and returns an EveryJob or a CronJob.
scheduler.repeat '10d' do; end.class
# => Rufus::Scheduler::EveryJob
scheduler.repeat '* * * * *' do; end.class
# => Rufus::Scheduler::CronJob
(Yes, no combination here gives back an IntervalJob).
A schedule block may be given 0, 1 or 2 arguments.
The first argument is "job", it's simply the Job instance involved. It might be useful if the job is to be unscheduled for some reason.
scheduler.every '10m' do |job|
status = determine_pie_status
if status == 'burnt' || status == 'cooked'
stop_oven
takeout_pie
job.unschedule
end
end
The second argument is "time", it's the time when the job got cleared for triggering (not Time.now).
Note that time is the time when the job got cleared for triggering. If there are mutexes involved, now = mutex_wait_time + time...
It's OK to change the next_time of an every job in-flight:
scheduler.every '10m' do |job|
# ...
status = determine_pie_status
job.next_time = Time.now + 30 * 60 if status == 'burnt'
#
# if burnt, wait 30 minutes for the oven to cool a bit
end
It should work as well with cron jobs, not so with interval jobs whose next_time is computed after their block ends its current run.
It's OK to pass any object, as long as it responds to #call(), when scheduling:
class Handler
def self.call(job, time)
p "- Handler called for #{job.id} at #{time}"
end
end
scheduler.in '10d', Handler
# or
class OtherHandler
def initialize(name)
@name = name
end
def call(job, time)
p "* #{time} - Handler #{name.inspect} called for #{job.id}"
end
end
oh = OtherHandler.new('Doe')
scheduler.every '10m', oh
scheduler.in '3d5m', oh
The call method must accept 2 (job, time), 1 (job) or 0 arguments.
Note that time is the time when the job got cleared for triggering. If there are mutexes involved, now = mutex_wait_time + time...
One can pass a handler class to rufus-scheduler when scheduling. Rufus will instantiate it and that instance will be available via job#handler.
class MyHandler
attr_reader :count
def initialize
@count = 0
end
def call(job)
@count += 1
puts ". #{self.class} called at #{Time.now} (#{@count})"
end
end
job = scheduler.schedule_every '35m', MyHandler
job.handler
# => #<MyHandler:0x000000021034f0>
job.handler.count
# => 0
If you want to keep that "block feeling":
job_id =
scheduler.every '10m', Class.new do
def call(job)
puts ". hello #{self.inspect} at #{Time.now}"
end
end
The scheduler can be paused via the #pause and #resume methods. One can determine if the scheduler is currently paused by calling #paused?.
While paused, the scheduler still accepts schedules, but no schedule will get triggered as long as #resume isn't called.
Sets the name of the job.
scheduler.cron '*/15 8 * * *', name: 'Robert' do |job|
puts "A, it's #{Time.now} and my name is #{job.name}"
end
job1 =
scheduler.schedule_cron '*/30 9 * * *', n: 'temporary' do |job|
puts "B, it's #{Time.now} and my name is #{job.name}"
end
# ...
job1.name = 'Beowulf'
By default, jobs are triggered in their own, new threads. When blocking: true
, the job is triggered in the scheduler thread (a new thread is not created). Yes, while a blocking job is running, the scheduler is not scheduling.
Since, by default, jobs are triggered in their own new threads, job instances might overlap. For example, a job that takes 10 minutes and is scheduled every 7 minutes will have overlaps.
To prevent overlap, one can set overlap: false
. Such a job will not trigger if one of its instances is already running.
The :overlap
option is considered before the :mutex
option when the scheduler is reviewing jobs for triggering.
When a job with a mutex triggers, the job's block is executed with the mutex around it, preventing other jobs with the same mutex from entering (it makes the other jobs wait until it exits the mutex).
This is different from overlap: false
, which is, first, limited to instances of the same job, and, second, doesn't make the incoming job instance block/wait but give up.
:mutex
accepts a mutex instance or a mutex name (String). It also accept an array of mutex names / mutex instances. It allows for complex relations between jobs.
Array of mutexes: original idea and implementation by Rainux Luo
Note: creating lots of different mutexes is OK. Rufus-scheduler will place them in its Scheduler#mutexes hash... And they won't get garbage collected.
The :overlap
option is considered before the :mutex
option when the scheduler is reviewing jobs for triggering.
It's OK to specify a timeout when scheduling some work. After the time specified, it gets interrupted via a Rufus::Scheduler::TimeoutError.
scheduler.in '10d', timeout: '1d' do
begin
# ... do something
rescue Rufus::Scheduler::TimeoutError
# ... that something got interrupted after 1 day
end
end
The :timeout option accepts either a duration (like "1d" or "2w3d") or a point in time (like "2013/12/12 12:00").
This option is for repeat jobs (cron / every) only.
It's used to specify the first time after which the repeat job should trigger for the first time.
In the case of an "every" job, this will be the first time (modulo the scheduler frequency) the job triggers. For a "cron" job as well, the :first will point to the first time the job has to trigger, the following trigger times are then determined by the cron string.
scheduler.every '2d', first_at: Time.now + 10 * 3600 do
# ... every two days, but start in 10 hours
end
scheduler.every '2d', first_in: '10h' do
# ... every two days, but start in 10 hours
end
scheduler.cron '00 14 * * *', first_in: '3d' do
# ... every day at 14h00, but start after 3 * 24 hours
end
:first, :first_at and :first_in all accept a point in time or a duration (number or time string). Use the symbol you think makes your schedule more readable.
Note: it's OK to change the first_at (a Time instance) directly:
job.first_at = Time.now + 10
job.first_at = Rufus::Scheduler.parse('2029-12-12')
The first argument (in all its flavours) accepts a :now or :immediately value. That schedules the first occurrence for immediate triggering. Consider:
require 'rufus-scheduler'
s = Rufus::Scheduler.new
n = Time.now; p [ :scheduled_at, n, n.to_f ]
s.every '3s', first: :now do
n = Time.now; p [ :in, n, n.to_f ]
end
s.join
that'll output something like:
[:scheduled_at, 2014-01-22 22:21:21 +0900, 1390396881.344438]
[:in, 2014-01-22 22:21:21 +0900, 1390396881.6453865]
[:in, 2014-01-22 22:21:24 +0900, 1390396884.648807]
[:in, 2014-01-22 22:21:27 +0900, 1390396887.651686]
[:in, 2014-01-22 22:21:30 +0900, 1390396890.6571937]
...
In some heavy-duty configurations, the :first_at
setting might be set on a point of time before the actual scheduling/triggering occurs and will result in an error "cannot set first[_at|_in] in the past..."
. To prevent that, the :first_at_no_error
option may be set to true.
scheduler.every '10h', first_at: Time.now + 10, first_at_no_error: true do
# ...
end
As introduced in gh-342.
This option is for repeat jobs (cron / every) only.
It indicates the point in time after which the job should unschedule itself.
scheduler.cron '5 23 * * *', last_in: '10d' do
# ... do something every evening at 23:05 for 10 days
end
scheduler.every '10m', last_at: Time.now + 10 * 3600 do
# ... do something every 10 minutes for 10 hours
end
scheduler.every '10m', last_in: 10 * 3600 do
# ... do something every 10 minutes for 10 hours
end
:last, :last_at and :last_in all accept a point in time or a duration (number or time string). Use the symbol you think makes your schedule more readable.
Note: it's OK to change the last_at (nil or a Time instance) directly:
job.last_at = nil
# remove the "last" bound
job.last_at = Rufus::Scheduler.parse('2029-12-12')
# set the last bound
One can tell how many times a repeat job (CronJob or EveryJob) is to execute before unscheduling by itself.
scheduler.every '2d', times: 10 do
# ... do something every two days, but not more than 10 times
end
scheduler.cron '0 23 * * *', times: 31 do
# ... do something every day at 23:00 but do it no more than 31 times
end
It's OK to assign nil to :times to make sure the repeat job is not limited. It's useful when the :times is determined at scheduling time.
scheduler.cron '0 23 * * *', times: (nolimit ? nil : 10) do
# ...
end
The value set by :times is accessible in the job. It can be modified anytime.
job =
scheduler.cron '0 23 * * *' do
# ...
end
# later on...
job.times = 10
# 10 days and it will be over
in
and at
accept a discard_past:
option since rufus-scheduler 3.9.0:
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
scheduler.in(-3600, discard_past: true) {}
# the job will never get scheduled
scheduler.in(-3600, discard_past: false) {}
# the job will trigger immediately
scheduler.in(-3600, discard_past: :fail) {}
# will raise an error...
Please note that discard_past
can be set at the scheduler level:
require 'rufus-scheduler'
s0 = Rufus::Scheduler.new(discard_past: true) # default
s1 = Rufus::Scheduler.new(discard_past: false)
# or
s1 = Rufus::Scheduler.new
s1.discard_past = false
s2 = Rufus::Scheduler.new(discard_past: :fail)
# or
s2 = Rufus::Scheduler.new
s2.discard_past = :fail
When calling a schedule method, the id (String) of the job is returned. Longer schedule methods return Job instances directly. Calling the shorter schedule methods with the job: true
also returns Job instances instead of Job ids (Strings).
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
job_id =
scheduler.in '10d' do
# ...
end
job =
scheduler.schedule_in '1w' do
# ...
end
job =
scheduler.in '1w', job: true do
# ...
end
Those Job instances have a few interesting methods / properties:
Returns the job id.
job = scheduler.schedule_in('10d') do; end
job.id
# => "in_1374072446.8923042_0.0_0"
Returns the scheduler instance itself.
Returns the options passed at the Job creation.
job = scheduler.schedule_in('10d', tag: 'hello') do; end
job.opts
# => { :tag => 'hello' }
Returns the original schedule.
job = scheduler.schedule_in('10d', tag: 'hello') do; end
job.original
# => '10d'
callable() returns the scheduled block (or the call method of the callable object passed in lieu of a block)
handler() returns nil if a block was scheduled and the instance scheduled otherwise.
# when passing a block
job =
scheduler.schedule_in('10d') do
# ...
end
job.handler
# => nil
job.callable
# => #<Proc:0x00000001dc6f58@/home/jmettraux/whatever.rb:115>
and
# when passing something else than a block
class MyHandler
attr_reader :counter
def initialize
@counter = 0
end
def call(job, time)
@counter = @counter + 1
end
end
job = scheduler.schedule_in('10d', MyHandler.new)
job.handler
# => #<Method: MyHandler#call>
job.callable
# => #<MyHandler:0x0000000163ae88 @counter=0>
Added to rufus-scheduler 3.8.0.
Returns the array [ 'path/to/file.rb', 123 ]
like Proc#source_location
does.
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
job = scheduler.schedule_every('2h') { p Time.now }
p job.source_location
# ==> [ '/home/jmettraux/rufus-scheduler/test.rb', 6 ]
Returns the Time instance when the job got created.
job = scheduler.schedule_in('10d', tag: 'hello') do; end
job.scheduled_at
# => 2013-07-17 23:48:54 +0900
Returns the last time the job triggered (is usually nil for AtJob and InJob).
job = scheduler.schedule_every('10s') do; end
job.scheduled_at
# => 2013-07-17 23:48:54 +0900
job.last_time
# => nil (since we've just scheduled it)
# after 10 seconds
job.scheduled_at
# => 2013-07-17 23:48:54 +0900 (same as above)
job.last_time
# => 2013-07-17 23:49:04 +0900
Returns the previous #next_time
scheduler.every('10s') do |job|
puts "job scheduled for #{job.previous_time} triggered at #{Time.now}"
puts "next time will be around #{job.next_time}"
puts "."
end
The job keeps track of how long its work was in the last_work_time
attribute. For a one time job (in, at) it's probably not very useful.
The attribute mean_work_time
contains a computed mean work time. It's recomputed after every run (if it's a repeat job).
Returns an array of EtOrbi::EoTime
instances (Time instances with a designated time zone), listing the n
next occurrences for this job.
Please note that for "interval" jobs, a mean work time is computed each time and it's used by this #next_times(n)
method to approximate the next times beyond the immediate next time.
Unschedule the job, preventing it from firing again and removing it from the schedule. This doesn't prevent a running thread for this job to run until its end.
Returns the list of threads currently "hosting" runs of this Job instance.
Interrupts all the work threads currently running for this job instance. They discard their work and are free for their next run (of whatever job).
Note: this doesn't unschedule the Job instance.
Note: if the job is pooled for another run, a free work thread will probably pick up that next run and the job will appear as running again. You'd have to unschedule and kill to make sure the job doesn't run again.
Returns true if there is at least one running Thread hosting a run of this Job instance.
Returns true if the job is scheduled (is due to trigger). For repeat jobs it should return true until the job gets unscheduled. "at" and "in" jobs will respond with false as soon as they start running (execution triggered).
These four methods are only available to CronJob, EveryJob and IntervalJob instances. One can pause or resume such jobs thanks to these methods.
job =
scheduler.schedule_every('10s') do
# ...
end
job.pause
# => 2013-07-20 01:22:22 +0900
job.paused?
# => true
job.paused_at
# => 2013-07-20 01:22:22 +0900
job.resume
# => nil
Returns the list of tags attached to this Job instance.
By default, returns an empty array.
job = scheduler.schedule_in('10d') do; end
job.tags
# => []
job = scheduler.schedule_in('10d', tag: 'hello') do; end
job.tags
# => [ 'hello' ]
Threads have thread-local variables, similarly Rufus-scheduler jobs have job-local variables. Those are more like a dict with thread-safe access.
job =
@scheduler.schedule_every '1s' do |job|
job[:timestamp] = Time.now.to_f
job[:counter] ||= 0
job[:counter] += 1
end
sleep 3.6
job[:counter]
# => 3
job.key?(:timestamp) # => true
job.has_key?(:timestamp) # => true
job.keys # => [ :timestamp, :counter ]
Locals can be set at schedule time:
job0 =
@scheduler.schedule_cron '*/15 12 * * *', locals: { a: 0 } do
# ...
end
job1 =
@scheduler.schedule_cron '*/15 13 * * *', l: { a: 1 } do
# ...
end
One can fetch the Hash directly with Job#locals
. Of course, direct manipulation is not thread-safe.
job.locals.entries do |k, v|
p "#{k}: #{v}"
end
Job instances have a #call method. It simply calls the scheduled block or callable immediately.
job =
@scheduler.schedule_every '10m' do |job|
# ...
end
job.call
Warning: the Scheduler#on_error handler is not involved. Error handling is the responsibility of the caller.
If the call has to be rescued by the error handler of the scheduler, call(true)
might help:
require 'rufus-scheduler'
s = Rufus::Scheduler.new
def s.on_error(job, err)
if job
p [ 'error in scheduled job', job.class, job.original, err.message ]
else
p [ 'error while scheduling', err.message ]
end
rescue
p $!
end
job =
s.schedule_in('1d') do
fail 'again'
end
job.call(true)
#
# true lets the error_handler deal with error in the job call
Returns when the job will trigger (hopefully).
An alias for time.
Returns the next time the job will trigger (hopefully).
Returns how many times the job fired.
It returns the scheduling frequency. For a job scheduled "every 20s", it's 20.
It's used to determine if the job frequency is higher than the scheduler frequency (it raises an ArgumentError if that is the case).
Returns the interval scheduled between each execution of the job.
Every jobs use a time duration between each start of their execution, while interval jobs use a time duration between the end of an execution and the start of the next.
An expensive method to run, it's brute. It caches its results. By default it runs for 2017 (a non leap-year).
require 'rufus-scheduler'
Rufus::Scheduler.parse('* * * * *').brute_frequency
#
# => #<Fugit::Cron::Frequency:0x00007fdf4520c5e8
# @span=31536000.0, @delta_min=60, @delta_max=60,
# @occurrences=525600, @span_years=1.0, @yearly_occurrences=525600.0>
#
# Occurs 525600 times in a span of 1 year (2017) and 1 day.
# There are least 60 seconds between "triggers" and at most 60 seconds.
Rufus::Scheduler.parse('0 12 * * *').brute_frequency
# => #<Fugit::Cron::Frequency:0x00007fdf451ec6d0
# @span=31536000.0, @delta_min=86400, @delta_max=86400,
# @occurrences=365, @span_years=1.0, @yearly_occurrences=365.0>
Rufus::Scheduler.parse('0 12 * * *').brute_frequency.to_debug_s
# => "dmin: 1D, dmax: 1D, ocs: 365, spn: 52W1D, spnys: 1, yocs: 365"
#
# 365 occurrences, at most 1 day between each, at least 1 day.
The CronJob#frequency
method found in rufus-scheduler < 3.5 has been retired.
The scheduler #job(job_id)
method can be used to look up Job instances.
require 'rufus-scheduler'
scheduler = Rufus::Scheduler.new
job_id =
scheduler.in '10d' do
# ...
end
# later on...
job = scheduler.job(job_id)
Are methods for looking up lists of scheduled Job instances.
Here is an example:
#
# let's unschedule all the at jobs
scheduler.at_jobs.each(&:unschedule)
When scheduling a job, one can specify one or more tags attached to the job. These can be used to look up the job later on.
scheduler.in '10d', tag: 'main_process' do
# ...
end
scheduler.in '10d', tags: [ 'main_process', 'side_dish' ] do
# ...
end
# ...
jobs = scheduler.jobs(tag: 'main_process')
# find all the jobs with the 'main_process' tag
jobs = scheduler.jobs(tags: [ 'main_process', 'side_dish' ]
# find all the jobs with the 'main_process' AND 'side_dish' tags
Returns the list of Job instance that have currently running instances.
Whereas other "_jobs" method scan the scheduled job list, this method scans the thread list to find the job. It thus comprises jobs that are running but are not scheduled anymore (that happens for at and in jobs).
Unschedule a job given directly or by its id.
Shuts down the scheduler, ceases any scheduler/triggering activity.
Shuts down the scheduler, waits (blocks) until all the jobs cease running.
Shuts down the scheduler, waits (blocks) at most n seconds until all the jobs cease running. (Jobs are killed after n seconds have elapsed).
Kills all the job (threads) and then shuts the scheduler down. Radical.
Returns true if the scheduler has been shut down.
Returns the Time instance at which the scheduler got started.
Returns since the count of seconds for which the scheduler has been running.
#uptime_s
returns this count in a String easier to grasp for humans, like "3d12m45s123"
.
Lets the current thread join the scheduling thread in rufus-scheduler. The thread comes back when the scheduler gets shut down.
#join
is mostly used in standalone scheduling script (or tiny one file examples). Calling #join
from a web application initializer will probably hijack the main thread and prevent the web application from being served. Do not put a #join
in such a web application initializer file.
Returns all the threads associated with the scheduler, including the scheduler thread itself.
Lists the work threads associated with the scheduler. The query option defaults to :all.
- :all : all the work threads
- :active : all the work threads currently running a Job
- :vacant : all the work threads currently not running a Job
Note that the main schedule thread will be returned if it is currently running a Job (ie one of those blocking: true
jobs).
Returns true if the arg is a currently scheduled job (see Job#scheduled?).
Returns a hash { job => [ t0, t1, ... ] }
mapping jobs to their potential trigger time within the [ time0, time1 ]
span.
Please note that, for interval jobs, the #mean_work_time
is used, so the result is only a prediction.
Like #occurrences
but returns a list [ [ t0, job0 ], [ t1, job1 ], ... ]
of time + job pairs.
The easy, job-granular way of dealing with errors is to rescue and deal with them immediately. The two next sections show examples. Skip them for explanations on how to deal with errors at the scheduler level.
As said, jobs could take care of their errors themselves.
scheduler.every '10m' do
begin
# do something that might fail...
rescue => e
$stderr.puts '-' * 80
$stderr.puts e.message
$stderr.puts e.stacktrace
$stderr.puts '-' * 80
end
end
Jobs are not only shrunk to blocks, here is how the above would look like with a dedicated class.
scheduler.every '10m', Class.new do
def call(job)
# do something that might fail...
rescue => e
$stderr.puts '-' * 80
$stderr.puts e.message
$stderr.puts e.stacktrace
$stderr.puts '-' * 80
end
end
TODO: talk about callable#on_error (if implemented)
(see scheduling handler instances and scheduling handler classes for more about those "callable jobs")
By default, rufus-scheduler intercepts all errors (that inherit from StandardError) and dumps abundant details to $stderr.
If, for example, you'd like to divert that flow to another file (descriptor), you can reassign $stderr for the current Ruby process
$stderr = File.open('/var/log/myapplication.log', 'ab')
or, you can limit that reassignement to the scheduler itself
scheduler.stderr = File.open('/var/log/myapplication.log', 'ab')
We've just seen that, by default, rufus-scheduler dumps error information to $stderr. If one needs to completely change what happens in case of error, it's OK to overwrite #on_error
def scheduler.on_error(job, error)
Logger.warn("intercepted error in #{job.id}: #{error.message}")
end
On Rails, the on_error
method redefinition might look like:
def scheduler.on_error(job, error)
Rails.logger.error(
"err#{error.object_id} rufus-scheduler intercepted #{error.inspect}" +
" in job #{job.inspect}")
error.backtrace.each_with_index do |line, i|
Rails.logger.error(
"err#{error.object_id} #{i}: #{line}")
end
end
One can bind callbacks before and after jobs trigger:
s = Rufus::Scheduler.new
def s.on_pre_trigger(job, trigger_time)
puts "triggering job #{job.id}..."
end
def s.on_post_trigger(job, trigger_time)
puts "triggered job #{job.id}."
end
s.every '1s' do
# ...
end
The trigger_time
is the time at which the job triggers. It might be a bit before Time.now
.
Warning: these two callbacks are executed in the scheduler thread, not in the work threads (the threads where the job execution really happens).
One can create an around callback which will wrap a job:
def s.around_trigger(job)
t = Time.now
puts "Starting job #{job.id}..."
yield
puts "job #{job.id} finished in #{Time.now-t} seconds."
end
The around callback is executed in the thread.
Returning false
in on_pre_trigger will prevent the job from triggering. Returning anything else (nil, -1, true, ...) will let the job trigger.
Note: your business logic should go in the scheduled block itself (or the scheduled instance). Don't put business logic in on_pre_trigger. Return false for admin reasons (backend down, etc), not for business reasons that are tied to the job itself.
def s.on_pre_trigger(job, trigger_time)
return false if Backend.down?
puts "triggering job #{job.id}..."
end
By default, rufus-scheduler sleeps 0.300 second between every step. At each step it checks for jobs to trigger and so on.
The :frequency option lets you change that 0.300 second to something else.
scheduler = Rufus::Scheduler.new(frequency: 5)
It's OK to use a time string to specify the frequency.
scheduler = Rufus::Scheduler.new(frequency: '2h10m')
# this scheduler will sleep 2 hours and 10 minutes between every "step"
Use with care.
This feature only works on OSes that support the flock (man 2 flock) call.
Starting the scheduler with lockfile: '.rufus-scheduler.lock'
will make the scheduler attempt to create and lock the file .rufus-scheduler.lock
in the current working directory. If that fails, the scheduler will not start.
The idea is to guarantee only one scheduler (in a group of schedulers sharing the same lockfile) is running.
This is useful in environments where the Ruby process holding the scheduler gets started multiple times.
If the lockfile mechanism here is not sufficient, you can plug your custom mechanism. It's explained in advanced lock schemes below.
(since rufus-scheduler 3.0.9)
The scheduler lock is an object that responds to #lock
and #unlock
. The scheduler calls #lock
when starting up. If the answer is false
, the scheduler stops its initialization work and won't schedule anything.
Here is a sample of a scheduler lock that only lets the scheduler on host "coffee.example.com" start:
class HostLock
def initialize(lock_name)
@lock_name = lock_name
end
def lock
@lock_name == `hostname -f`.strip
end
def unlock
true
end
end
scheduler =
Rufus::Scheduler.new(scheduler_lock: HostLock.new('coffee.example.com'))
By default, the scheduler_lock is an instance of Rufus::Scheduler::NullLock
, with a #lock
that returns true.
(since rufus-scheduler 3.0.9)
The trigger lock in an object that responds to #lock
. The scheduler calls that method on the job lock right before triggering any job. If the answer is false, the trigger doesn't happen, the job is not done (at least not in this scheduler).
Here is a (stupid) PingLock example, it'll only trigger if an "other host" is not responding to ping. Do not use that in production, you don't want to fork a ping process for each trigger attempt...
class PingLock
def initialize(other_host)
@other_host = other_host
end
def lock
! system("ping -c 1 #{@other_host}")
end
end
scheduler =
Rufus::Scheduler.new(trigger_lock: PingLock.new('main.example.com'))
By default, the trigger_lock is an instance of Rufus::Scheduler::NullLock
, with a #lock
that always returns true.
As explained in advanced lock schemes, another way to tune that behaviour is by overriding the scheduler's #confirm_lock
method. (You could also do that with an #on_pre_trigger
callback).
In rufus-scheduler 2.x, by default, each job triggering received its own, brand new, thread of execution. In rufus-scheduler 3.x, execution happens in a pooled work thread. The max work thread count (the pool size) defaults to 28.
One can set this maximum value when starting the scheduler.
scheduler = Rufus::Scheduler.new(max_work_threads: 77)
It's OK to increase the :max_work_threads of a running scheduler.
scheduler.max_work_threads += 10
Do not want to store a reference to your rufus-scheduler instance?
Then Rufus::Scheduler.singleton
can help, it returns a singleton instance of the scheduler, initialized the first time this class method is called.
Rufus::Scheduler.singleton.every '10s' { puts "hello, world!" }
It's OK to pass initialization arguments (like :frequency or :max_work_threads) but they will only be taken into account the first time .singleton
is called.
Rufus::Scheduler.singleton(max_work_threads: 77)
Rufus::Scheduler.singleton(max_work_threads: 277) # no effect
The .s
is a shortcut for .singleton
.
Rufus::Scheduler.s.every '10s' { puts "hello, world!" }
As seen above, rufus-scheduler proposes the :lockfile system out of the box. If in a group of schedulers only one is supposed to run, the lockfile mechanism prevents schedulers that have not set/created the lockfile from running.
There are situations where this is not sufficient.
By overriding #lock and #unlock, one can customize how schedulers lock.
This example was provided by Eric Lindvall:
class ZookeptScheduler < Rufus::Scheduler
def initialize(zookeeper, opts={})
@zk = zookeeper
super(opts)
end
def lock
@zk_locker = @zk.exclusive_locker('scheduler')
@zk_locker.lock # returns true if the lock was acquired, false else
end
def unlock
@zk_locker.unlock
end
def confirm_lock
return false if down?
@zk_locker.assert!
rescue ZK::Exceptions::LockAssertionFailedError => e
# we've lost the lock, shutdown (and return false to at least prevent
# this job from triggering
shutdown
false
end
end
This uses a zookeeper to make sure only one scheduler in a group of distributed schedulers runs.
The methods #lock and #unlock are overridden and #confirm_lock is provided, to make sure that the lock is still valid.
The #confirm_lock method is called right before a job triggers (if it is provided). The more generic callback #on_pre_trigger is called right after #confirm_lock.
(introduced in rufus-scheduler 3.0.9).
Another way of providing #lock
, #unlock
and #confirm_lock
to a rufus-scheduler is by using the :scheduler_lock
and :trigger_lock
options.
See :trigger_lock and :scheduler_lock.
The scheduler lock may be used to prevent a scheduler from starting, while a trigger lock prevents individual jobs from triggering (the scheduler goes on scheduling).
One has to be careful with what goes in #confirm_lock
or in a trigger lock, as it gets called before each trigger.
Warning: you may think you're heading towards "high availability" by using a trigger lock and having lots of schedulers at hand. It may be so if you limit yourself to scheduling the same set of jobs at scheduler startup. But if you add schedules at runtime, they stay local to their scheduler. There is no magic that propagates the jobs to all the schedulers in your pack.
(Please note that fugit does the heavy-lifting parsing work for rufus-scheduler).
Rufus::Scheduler provides a class method .parse
to parse time durations and cron strings. It's what it's using when receiving schedules. One can use it directly (no need to instantiate a Scheduler).
require 'rufus-scheduler'
Rufus::Scheduler.parse('1w2d')
# => 777600.0
Rufus::Scheduler.parse('1.0w1.0d')
# => 777600.0
Rufus::Scheduler.parse('Sun Nov 18 16:01:00 2012').strftime('%c')
# => 'Sun Nov 18 16:01:00 2012'
Rufus::Scheduler.parse('Sun Nov 18 16:01:00 2012 Europe/Berlin').strftime('%c %z')
# => 'Sun Nov 18 15:01:00 2012 +0000'
Rufus::Scheduler.parse(0.1)
# => 0.1
Rufus::Scheduler.parse('* * * * *')
# => #<Fugit::Cron:0x00007fb7a3045508
# @original="* * * * *", @cron_s=nil,
# @seconds=[0], @minutes=nil, @hours=nil, @monthdays=nil, @months=nil,
# @weekdays=nil, @zone=nil, @timezone=nil>
It returns a number when the input is a duration and a Fugit::Cron instance when the input is a cron string.
It will raise an ArgumentError if it can't parse the input.
Beyond .parse
, there are also .parse_cron
and .parse_duration
, for finer granularity.
There is an interesting helper method named .to_duration_hash
:
require 'rufus-scheduler'
Rufus::Scheduler.to_duration_hash(60)
# => { :m => 1 }
Rufus::Scheduler.to_duration_hash(62.127)
# => { :m => 1, :s => 2, :ms => 127 }
Rufus::Scheduler.to_duration_hash(62.127, drop_seconds: true)
# => { :m => 1 }
To schedule something at noon every first Monday of the month:
scheduler.cron('00 12 * * mon#1') do
# ...
end
To schedule something at noon the last Sunday of every month:
scheduler.cron('00 12 * * sun#-1') do
# ...
end
#
# OR
#
scheduler.cron('00 12 * * sun#L') do
# ...
end
Such cronlines can be tested with scripts like:
require 'rufus-scheduler'
Time.now
# => 2013-10-26 07:07:08 +0900
Rufus::Scheduler.parse('* * * * mon#1').next_time.to_s
# => 2013-11-04 00:00:00 +0900
L can be used in the "day" slot:
In this example, the cronline is supposed to trigger every last day of the month at noon:
require 'rufus-scheduler'
Time.now
# => 2013-10-26 07:22:09 +0900
Rufus::Scheduler.parse('00 12 L * *').next_time.to_s
# => 2013-10-31 12:00:00 +0900
It's OK to pass negative values in the "day" slot:
scheduler.cron '0 0 -5 * *' do
# do it at 00h00 5 days before the end of the month...
end
Negative ranges (-10--5-
: 10 days before the end of the month to 5 days before the end of the month) are OK, but mixed positive / negative ranges will raise an ArgumentError
.
Negative ranges with increments (-10---2/2
) are accepted as well.
Descending day ranges are not accepted (10-8
or -8--10
for example).
Cron schedules and at schedules support the specification of a timezone.
scheduler.cron '0 22 * * 1-5 America/Chicago' do
# the job...
end
scheduler.at '2013-12-12 14:00 Pacific/Samoa' do
puts "it's tea time!"
end
# or even
Rufus::Scheduler.parse("2013-12-12 14:00 Pacific/Saipan")
# => #<Rufus::Scheduler::ZoTime:0x007fb424abf4e8 @seconds=1386820800.0, @zone=#<TZInfo::DataTimezone: Pacific/Saipan>, @time=nil>
For when you see an error like:
rufus-scheduler/lib/rufus/scheduler/zotime.rb:41:
in `initialize':
cannot determine timezone from nil (etz:nil,tnz:"**标准时间",tzid:nil)
(ArgumentError)
from rufus-scheduler/lib/rufus/scheduler/zotime.rb:198:in `new'
from rufus-scheduler/lib/rufus/scheduler/zotime.rb:198:in `now'
from rufus-scheduler/lib/rufus/scheduler.rb:561:in `start'
...
It may happen on Windows or on systems that poorly hint to Ruby which timezone to use. It should be solved by setting explicitly the ENV['TZ']
before the scheduler instantiation:
ENV['TZ'] = 'Asia/Shanghai'
scheduler = Rufus::Scheduler.new
scheduler.every '2s' do
puts "#{Time.now} Hello #{ENV['TZ']}!"
end
On Rails you might want to try with:
ENV['TZ'] = Time.zone.name # Rails only
scheduler = Rufus::Scheduler.new
scheduler.every '2s' do
puts "#{Time.now} Hello #{ENV['TZ']}!"
end
(Hat tip to Alexander in gh-230)
Rails sets its timezone under config/application.rb
.
Rufus-Scheduler 3.3.3 detects the presence of Rails and uses its timezone setting (tested with Rails 4), so setting ENV['TZ']
should not be necessary.
The value can be determined thanks to https://en.wikipedia.org/wiki/List_of_tz_database_time_zones.
Use a "continent/city" identifier (for example "Asia/Shanghai"). Do not use an abbreviation (not "CST") and do not use a local time zone name (not "**标准时间" nor "Eastern Standard Time" which, for instance, points to a time zone in America and to another one in Australia...).
If the error persists (and especially on Windows), try to add the tzinfo-data
to your Gemfile, as in:
gem 'tzinfo-data'
or by manually requiring it before requiring rufus-scheduler (if you don't use Bundler):
require 'tzinfo/data'
require 'rufus-scheduler'
Currently (3.9.x), rufus-scheduler strives to trigger at the right time. The trigger thread might not yield a Time.now
in the scheduled timezone.
ENV['TZ'] = 'Asia/Tokyo'
require 'rufus-scheduler'
s = Rufus::Scheduler.new
s.cron('*/5 * * * * * Europe/Rome') do
p Time.now # ==> the representation will indicate the time is UTC+0900...
end
s.join
In the wake of gh-341.
Yes, I know, all of the above is boring and you're only looking for a snippet to paste in your Ruby-on-Rails application to schedule...
Here is an example initializer:
#
# config/initializers/scheduler.rb
require 'rufus-scheduler'
# Let's use the rufus-scheduler singleton
#
s = Rufus::Scheduler.singleton
# Stupid recurrent task...
#
s.every '1m' do
Rails.logger.info "hello, it's #{Time.now}"
Rails.logger.flush
end
And now you tell me that this is good, but you want to schedule stuff from your controller.
Maybe:
class ScheController < ApplicationController
# GET /sche/
#
def index
job_id =
Rufus::Scheduler.singleton.in '5s' do
Rails.logger.info "time flies, it's now #{Time.now}"
end
render text: "scheduled job #{job_id}"
end
end
The rufus-scheduler singleton is instantiated in the config/initializers/scheduler.rb
file, it's then available throughout the webapp via Rufus::Scheduler.singleton
.
Warning: this works well with single-process Ruby servers like Webrick and Thin. Using rufus-scheduler with Passenger or Unicorn requires a bit more knowledge and tuning, gently provided by a bit of googling and reading, see Faq above.
(Written in reply to gh-186)
If you don't want rufus-scheduler to trigger anything while running the Ruby on Rails console, running for tests/specs, or running from a Rake task, you can insert a conditional return statement before jobs are added to the scheduler instance:
#
# config/initializers/scheduler.rb
require 'rufus-scheduler'
return if defined?(Rails::Console) || Rails.env.test? || File.split($PROGRAM_NAME).last == 'rake'
#
# do not schedule when Rails is run from its console, for a test/spec, or
# from a Rake task
# return if $PROGRAM_NAME.include?('spring')
#
# see https://github.com/jmettraux/rufus-scheduler/issues/186
s = Rufus::Scheduler.singleton
s.every '1m' do
Rails.logger.info "hello, it's #{Time.now}"
Rails.logger.flush
end
(Beware later version of Rails where Spring takes care pre-running the initializers. Running spring stop
or disabling Spring might be necessary in some cases to see changes to initializers being taken into account.)
(Written in reply to #165 )
There is the handy rails server -d
that starts a development Rails as a daemon. The annoying thing is that the scheduler as seen above is started in the main process that then gets forked and daemonized. The rufus-scheduler thread (and any other thread) gets lost, no scheduling happens.
I avoid running -d
in development mode and bother about daemonizing only for production deployment.
These are two well crafted articles on process daemonization, please read them:
- https://www.mikeperham.com/2014/09/22/dont-daemonize-your-daemons/
- https://www.mikeperham.com/2014/07/07/use-runit/
If, anyway, you need something like rails server -d
, why not try bundle exec unicorn -D
instead? In my (limited) experience, it worked out of the box (well, had to add gem 'unicorn'
to Gemfile
first).
You might benefit from wrapping your scheduled code in the executor or reloader. Read more here: https://guides.rubyonrails.org/threading_and_code_execution.html
see getting help above.
MIT, see LICENSE.txt