I am trying to run multiple tasks, each task access the database, and I am trying to run the tasks into separate execution wires.
I played around, tried allow_concurrency which I have set to true, or config.thread_safe! but it I get un-deterministic errors, for example sometimes a class is missing, or a constant ...
here is some code
grabbers = get_grabber_name_list
threads = []
grabbers.each { |grabber|
threads << Thread.new {
ARGV[0] = grabber
if (##last_run_timestamp[grabber.to_sym].blank? || (##last_run_timestamp[grabber.to_sym] >= AbstractGrabber.aff_net_accounts(grabber, "grab_interval").seconds.ago))
Rake::Task["aff_net:import:" + grabber].execute
##last_run_timestamp.merge!({grabber.to_sym => Time.now})
end
}
}
threads.each {|t| t.join }
thanks
I've recently implemented a Rails application that uses threads and made a few discoveries:
First, if you're writing to any arrays or hashes (i.e., complex types) outside your thread, wrap them in a mutex. It looks to me like hash and array references may not be thread safe. It seems unlikely that hash/array element indexing isn't thread safe but all I know is that after I put the external data structures in a mutex before writing, problems disappeared.
Second, close your ActiveRecord connection when the thread terminates, otherwise you can end up creating a large number of stale connections. Here's a post about how to do this. I don't know if it still applies for Rails versions > 2.2 but after I started closing connections explicitly, my problems disappeared. The author suggests monkey-patching ActiveRecord to do this automatically but I decided to release connections explicitly in my code.
Here's a sample of code that's working for me:
mutex = Mutex.new
my_array = []
threads = []
1.upto(10) do |i|
threads << Thread.new {
begin
do_some_stuff
mutex.synchronize {
# You'd think that each thread would only touch its own personal
# array element but without a mutex, I run into problems.
my_array[i] = some_computed_value
}
ensure
ActiveRecord::Base.connection_pool.release_connection
end
}
}
threads.each {|t| t.join}
By the way, if you're using threads to take advantage of multi-core CPUs, you'll need to use JRuby. As far as I know, JRuby is the only implementation that can take advantage of native CPU threads. If you use threads so you can do other things while waiting on network connections or some other non-CPU task, this isn't an issue.
You should probably do this using background workers. There are a few options for background worker libraries, but my favourite is delayed_job (http://github.com/tobi/delayed_job).
It should be pretty easy to convert the code you posted into background jobs.
Related
I'm using sidekiq for background job and I enqueue a job like this:
SampleJob.set(wait: waiting_time.to_i.seconds).perform_later(***) ・・・ ①
When waiting_time is nil,
it becomes
SampleJob.set(wait: 0.seconds).perform_later(***)
Of course it works well, but I'm worried about performance because worker enqueued with wait argument is derived by poller,
so I wonder if I should remove set(wait: waiting_time.to_i.seconds) when
waiting_time is nil.
i.e.)
if waiting_time.present?
SampleJob.set(wait: waiting_time.to_i.seconds).perform_later(***)
else
SampleJob.perform_later(***)
end ・・・ ②
Is there any differences in performance or speed between ① and ②?
Thank you in advance.
There is no difference. It looks like this is already considered in the Sidekiq library.
https://github.com/mperham/sidekiq/blob/main/lib/sidekiq/worker.rb#L261
# Optimization to enqueue something now that is scheduled to go out now or in the past
#opts["at"] = ts if ts > now
I want to run graphs/futures on my distributed cluster which all have a 'load data' root task and then a bunch of training tasks that run on that data. A simplified version would look like this:
from dask.distributed import Client
client = Client(scheduler_ip)
load_data_future = client.submit(load_data_func, 'path/to/data/')
train_task_futures = [client.submit(train_func, load_data_future, params)
for params in train_param_set]
Running this as above the scheduler gets one worker to read the file, then it spills that data to disk to share it with the other workers. However, loading the data is usually reading from a large HDF5 file, which can be done concurrently, so I was wondering if there was a way to force all workers to read this file concurrently (they all compute the root task) instead of having them wait for one worker to finish then slowly transferring the data from that worker.
I know there is the client.run() method which I can use to get all workers to read the file concurrently, but how would you then get the data you've read to feed into the downstream tasks?
I cannot use the dask data primitives to concurrently read HDF5 files because I need things like multi-indexes and grouping on multiple columns.
Revisited this question and found a relatively simple solution, though it uses internal API methods and involves a blocking call to client.run(). Using the same variables as in the question:
from distributed import get_worker
client_id = client.id
def load_dataset():
worker = get_worker()
data = {'load_dataset-0': load_data_func('path/to/data')}
info = worker.update_data(data=data, report=False)
worker.scheduler.update_data(who_has={key: [worker.address] for key in data},
nbytes=info['nbytes'], client=client_id)
client.run(load_dataset)
Now if you run client.has_what() you should see that each worker holds the key load_dataset-0. To use this in downstream computations you can simply create a future for the key:
from distributed import Future
load_data_future = Future('load_dataset-0', client=client)
and this can be used with client.compute() or dask.delayed as usual. Indeed the final line from the example in the question would work fine:
train_task_futures = [client.submit(train_func, load_data_future, params)
for params in train_param_set]
Bear in mind that it uses internal API methods Worker.update_data and Scheduler.update_data and works fine as of distributed.__version__ == 1.21.6 but could be subject to change in future releases.
As of today (distributed.__version__ == 1.20.2) what you ask for is not possible. The closest thing would be to compute once and then replicate the data explicitly
future = client.submit(load, path)
wait(future)
client.replicate(future)
You may want to raise this as a feature request at https://github.com/dask/distributed/issues/new
Let's say we want to test that the database is being locked..
$transaction = Thread.new {
Rails.logger.debug 'transaction process start'
Inventory.transaction do
inventory.lock!
Thread.stop
inventory.units_available=99
inventory.save
end
}
$race_condition = Thread.new {
Rails.logger.debug 'race_condition process start'
config = ActiveRecord::Base.configurations[Rails.env].symbolize_keys
config[:flags] = 65536 | 131072 | Mysql2::Client::FOUND_ROWS
begin
connection = Mysql2::Client.new(config)
$transaction.run
$transaction.join
rescue NoMethodError
ensure
connection.close if connection
end
}
Rails.logger.debug 'main process start'
$transaction.join
Rails.logger.debug 'main process after transaction.join'
sleep 0.1 while $transaction.status!='sleep'
Rails.logger.debug 'main process after sleep'
$race_condition.join
Rails.logger.debug 'main process after race_condition.join'
In theory, I'd think it would do the transaction thread, then wait( Thread.stop ), then the main process would see that it's sleeping, and start the race condition thread(which will be trying to alter data in the locked table when it actually works). Then the race condition would continue the transaction thread after it was done.
what's weird is the trace
main process start
transaction process start
race_condition process start
Coming from nodejs, it seems like threads aren't exactly as user friendly.. though, there has to be a way to get this done.
Is there an easier way to lock the database, then try to change it with a different thread?
Thread.new automatically starts the Thread.
But that does not mean that it is executing.
That depends on Operations system, ruby or jruby, how many cores, etc.
In your example the main thread runs until
$transaction.join,
and only then your transaction thread starts, just by chance.
It runs still Thread.stop, then your '$race_condition' Thread starts, because both other are blocked (it might have started before)
So that explains your log.
You have two $transaction.join
they wait until the thread exits, but a thread can only exit once...
I don't know what is happen then, maybe the second call waits forever.
For your test, you need some sort of explicit synchronization, so that our race_thread writes exactly when the transaction_thread is in the middle of the transaction. You can do this with Mutex, but better would be some sort of message passing. The following blog post may help:
http://www.engineyard.com/blog/2011/a-modern-guide-to-threads/
For any resource to make it a "Mutually Exclusive", you need to use Mutex class and use a synchronize method to make the resources locked while one thread is using them. You have to do something like this:
semaphore = Mutex.new
and use it inside the Thread instance.
$transaction = Thread.new {
semaphore.synchronize{
# Do whatever you want with the *your shared resource*
}
}
This way you can prevent any deadlocks.
Hope this helps.
I have a Ruby process that listens on a given device. I would like to spin up/down instances of it for different devices with a rails app. Everything I can find for Ruby daemons seems to be based around a set number of daemons running or background processing with message queues.
Should I just be doing this with Kernel.spawn and storing the PIDs in the database? It seems a bit hacky but if there isn't an existing framework that allows me to bring up/down daemons it seems I may not have much choice.
Instead of spawning another script and keeping the PIDs in the database, you can do it all within the same script, using fork, and keeping PIDs in memory. Here's a sample script - you add and delete "worker instances" by typing commands "add" and "del" in console, exiting with "quit":
#pids = []
#counter = 0
def add_process
#pids.push(Process.fork {
loop do
puts "Hello from worker ##{#counter}"
sleep 1
end
})
#counter += 1
end
def del_process
return false if #pids.empty?
pid = #pids.pop
Process.kill('SIGTERM', pid)
true
end
def kill_all
while del_process
end
end
while cmd = gets.chomp
case cmd.downcase
when 'quit'
kill_all
exit
when 'add'
add_process
when 'del'
del_process
end
end
Of course, this is just an example, and for sending comands and/or monitoring instances you can replace this simple gets loop with a small Sinatra app, or socket interface, or named pipes etc.
I am using lock! in my code and want to catch the exception thrown if lock! fails for some reason (e.g. cannot get the lock). What kind of exceptions can lock! throw? I checked the ruby docs but couldn't find the specific Exception classes.
Thanks.
When in doubt, probe.
Consider the following pair of functions:
def long_hold
ActiveRecord::Base.transaction do
u = User.find(220)
u.lock!
sleep 100.seconds
u.email="foo#bar.com"
u.save!
end
end
def short_hold
ActiveRecord::Base.transaction do
u = User.find(220)
u.lock!
u.email="foo#bar.com"
u.save!
end
end
In my setup (OSX 10.11, ruby 2.2.4, rails 4.2, postgres 9.5), running long_hold in one rails console and then running short_hold in a second console, I observe short_hold blocks until long_hold completes; moreover, instrumenting the code with puts, we see that while long_hold is sleeping, short_hold is waiting to acquire the lock.
Assuming no caveats about the independence of rails consoles, this suggests that no exceptions are thrown if a second process tries to lock a row that is already locked, but that process blocks until the first completes.
Here is the source for that locking call. It calls reload and its source looks like this:
# File lib/active_record/base.rb, line 2333
2333: def reload(options = nil)
2334: clear_aggregation_cache
2335: clear_association_cache
2336: #attributes.update(self.class.find(self.id, options).instance_variable_get('#attributes'))
2337: #attributes_cache = {}
2338: self
2339: end
so when you call reload(:lock => lock) as the call to lock does it it really updating the attributes of that record.
There are a lot of different situations here. You could try to lock a record that dosnt exist, or lock one that has been locked elsewhere. What error are you interested in catching?