Awesome

QuasiQueue

QuasiQueue is a MultiProcessing library for Python that makes it super easy to have long running MultiProcess jobs. QuasiQueue handles process creation and cleanup, signal management, cross process communication, and all the other garbage that makes people hate dealing with multiprocessing.

QuasiQueue works by splitting the work into two components- the main process whose job it is to feed a Queue with work, and then read processes that take work off of the Queue to run. All the developers have to do is create two functions-

• writer is called when the queue gets low. It should return an iterable (list, generator) that QuasiQueue uses to grow the multiprocess Queue.
• reader is called once for each item in the Queue. It runs in completely different processes from the writer.

flowchart LR
  writer(writer)-->queue((queue))
  queue-->reader1(reader)
  queue-->reader2(reader)
  queue-->reader3(reader)
  queue-->reader4(reader)

These functions can be as simple or complex as you need.

import asyncio

from quasiqueue import QuasiQueue


async def writer(desired: int):
  """Feeds data to the Queue when it is low.
  """
  for x in range(0, desired):
    yield x



async def reader(identifier: int|str):
  """Receives individual items from the queue.

  Args:
      identifier (int | str): Comes from the output of the Writer function
  """
  print(f"{identifier}")


runner = QuasiQueue(
  "hello_world",
  reader=reader,
  writer=writer,
)

if __name__ == '__main__':
  asyncio.run(runner.main())

Use Cases

There are a ton of use cases for QuasiQueue.

WebServer

QuasiQueue could be the basis for a web server. The write function would need to feed sockets to the Queue, would would be picked up by the reader for handling.

flowchart LR

  subgraph Configuration
  http
  end

  subgraph Server
  http-->writer
  writer(writer)--socket-->queue((queue))
  queue--socket-->reader1(reader)
  queue--socket-->reader2(reader)
  queue--socket-->reader3(reader)
  queue--socket-->reader4(reader)
  end

Website Image Crawler

QuasiQueue could be used to crawl a website, or series of websites, to download data.

flowchart RL

  subgraph Crawler
  writer(writer)-->queue((queue))
  queue-->reader1(reader)
  end
  database(Links)--Stale or Unread Links-->writer
  reader1(reader)--Images-->FileSystem
  reader1(reader)--Found Links-->database

As new pages are found they get added to a database. The write pulls pages out of the database as the queue gets smaller, and the reader adds new pages that it finds to the database. The writer function can pull links that haven't been crawled at all, and once it runs out of those it can recrawl links based on their age.

Image Processor

QuasiQueue can be used to run large one off jobs as well, such as processing a list of images. If someone has several thousand images to process they can have the writer function feed the list into the Queue, and reader processes can take the files from the queue and run the processing needed.

flowchart LR

  subgraph Configuration
  filelist
  end

  subgraph ImageProcessor
  filelist-->writer
  writer(writer)-->queue((queue))
  queue-->reader1(reader)
  end
  reader1(reader)-->ProcessedFiles

Installation

pip install quasiqueue

Arguments

Name

The first argument when initializing QuasiQueue is the name of the queue. This is used when naming new processes (which makes logging and ps commands a lot more useful)

Reader

The reader function is called once per item in the queue.

async def reader(identifier: int|str):
  """Receives individual items from the queue.

  Args:
      identifier (int | str): Comes from the output of the Writer function
  """
  print(f"{identifier}")

The reader can be extremely simple, as this one liner shows, or it can be extremely complex.

The reader can be asynchronous or synchronous. When the reader is an async function the concurrent_tasks_per_process setting can be used to control how many reader tasks will run per process. For example, if you have four processes running and allow ten concurrent tasks then the reader function will have 40 instances running. This can be beneficial if your reader function requires a lot of independent IO (such as disk writes or HTTP lookups), but if your reader is primarily running calculations then it's less likely to benefit.

Writer

The write function is called whenever the Queue is low. It has to return an iterator of items that can be pickled (strings, integers, or sockets are common examples) that will be feed to the Reader. Generators are a great option to reduce memory usage, but even simple lists can be returned. The writer function has to be asynchronous.

The writer function only has one argument- the desired number of items that QuasiQueue would like to retrieve and add to the Queue. This number is meant to allow for optimization on behalf of the developers- it can be completely ignored, but QuasiQueue will run more efficiently if you keep it as close the desired as possible.

async def writer(desired: int):
  """Feeds data to the Queue when it is low.
  """
  return range(0, desired)

In the event that there are no items available to put in the Queue the write function should return None. This will signal to QuasiQueue that there is nothing for it, and it will add a slight (configurable) delay before attempting to retrieve more items.

QuasiQueue will prevent items that were recently placed in the Queue from being requeued within a configurable time frame. This is meant to make the write function more lenient- if it happens to return duplicates between calls QuasiQueue will just discard them.

Context

The context function is completely optional. It runs once, and only once, when a new reader process is launched. It is used to initialize resources such as database pools so they can be reused between reader calls.

If the function is provided it should return a dictionary. The reader function will need to have a context argument, which will be the results from the context function. The context function can be asynchronous or synchronous.

def context():
  ctx = {}
  ctx['http'] = get_http_connection_pool()
  ctx['dbengine'] = get_db_engine_pool()
  return ctx

def reader(identifier: int|str, ctx: Dict[str, Any]):
  """Receives individual items from the queue.

  Args:
      identifier (int | str): Comes from the output of the Writer function
      ctx (Dict[str, Any]): Comes from the output of the Context function
  """
  ctx['dbengine'].execute("get item")
  ctx['http'].get("url")
  print(f"{identifier}")


runner = QuasiQueue(
  "hello_world",
  reader=reader,
  writer=writer,
  context=context
)

Although this function is not required it can have amazing performance implications. Connection pooling of databases and websites can save a remarkable amount of resources on SSL handshakes alone.

Settings

QuasiQueue has a variety of optimization settings that can be tweaked depending on usage.

Settings can be configured programmatically, via environment variables, or both.

Environment Variables

All Settings can be configured via environment variables. The variables should start with the QuasiQueue name and an underscore. For example, if you named your QuasiQueue Acme then ACME_NUM_PROCESS would be used to set the number of processes.

Programmatic

There are two methods to programmatically define the settings.

The first one is to initialize the settings and override the specific ones.

from quasiqueue import Settings, QuasiQueue

QuasiQueue(
  "MyQueue",
  reader=reader,
  writer=writer,
  settings=Settings(lookup_block_size=50)
)

This method is simple, but the downside is that you lose the environment variable prefixes. So when using this method you have to set NUM_PROCESSES rather than MYQUEUE_NUM_PROCESSES. The work around is to extend the Settings object to give it your desired prefix.

from quasiqueue import Settings, QuasiQueue
from pydantic_settings import SettingsConfigDict

class MySettings(Settings)
  model_config = SettingsConfigDict(env_prefix="MY_QUEUE_")
  lookup_block_size: int = 50

QuasiQueue(
  "MyQueue",
  reader=reader,
  writer=writer,
  settings=MySettings()
)

Accessing Settings

The reader, writer, and context functions can optionally retrieve a copy of the QuasiQueue settings in use by defining a settings argument in their function.

async def reader(item: Any, settings: Dict[str, Any])
  print(settings['project_name'])

If you create a custom Settings class, as in the programmatic example, you can add your own fields that will be passed to your QuasiQueue functions.