A Django app to run new background tasks from either admin or cron, and inspect task history from admin
Contents
- Install Django Task:
pip install django-task
- Add it to your `INSTALLED_APPS`:
INSTALLED_APPS = (
...
'django_rq', # optional (not needed when using TaskThreaded)
'django_task',
...
)
- Add Django Task's URL patterns:
urlpatterns = [
...
path('django_task/', include('django_task.urls', namespace='django_task')),
...
]
- Configure Redis and RQ in settings.py (not needed when using TaskThreaded):
#REDIS_URL = 'redis://localhost:6379/0'
redis_host = os.environ.get('REDIS_HOST', 'localhost')
redis_port = 6379
REDIS_URL = 'redis://%s:%d/0' % (redis_host, redis_port)
CACHES = {
'default': {
'BACKEND': 'redis_cache.RedisCache',
'LOCATION': REDIS_URL
},
}
#
# RQ config
#
RQ_PREFIX = "myproject_"
QUEUE_DEFAULT = RQ_PREFIX + 'default'
QUEUE_HIGH = RQ_PREFIX + 'high'
QUEUE_LOW = RQ_PREFIX + 'low'
RQ_QUEUES = {
QUEUE_DEFAULT: {
'URL': REDIS_URL,
#'PASSWORD': 'some-password',
'DEFAULT_TIMEOUT': 360,
},
QUEUE_HIGH: {
'URL': REDIS_URL,
'DEFAULT_TIMEOUT': 500,
},
QUEUE_LOW: {
'URL': REDIS_URL,
#'ASYNC': False,
},
}
Note: if you plan to install many instances of the project on the same server, for each instance use a specific value for RQ_PREFIX; for example:
INSTANCE_PREFIX = "myproject_"
try:
from project.settings.instance_prefix import *
except Exception as e:
pass
RQ_PREFIX = INSTANCE_PREFIX
QUEUE_DEFAULT = RQ_PREFIX + '_default'
QUEUE_LOW = RQ_PREFIX + '_low'
QUEUE_HIGH = RQ_PREFIX + '_high'
...
- Customize django-task specific settings (optional):
RQ_SHOW_ADMIN_LINK = False
DJANGOTASK_LOG_ROOT = os.path.abspath(os.path.join(BASE_DIR, '..', 'protected', 'tasklog'))
DJANGOTASK_ALWAYS_EAGER = False
DJANGOTASK_JOB_TRACE_ENABLED = False
DJANGOTASK_REJECT_IF_NO_WORKER_ACTIVE_FOR_QUEUE = True
- Optionally, revoke pending tasks at startapp;
file main/apps.py:
class MainConfig(AppConfig):
...
def ready(self):
...
try:
from django_task.utils import revoke_pending_tasks
revoke_pending_tasks()
except Exception as e:
print(e)
Purposes
- create async tasks either programmatically or from admin
- monitor async tasks from admin
- log all tasks in the database for later inspection
- optionally save task-specific logs in a TextField and/or in a FileField
Details
- each specific task is described by a Model derived from either models.TaskRQ or models.TaskThreaded, which
is responsible for:
- selecting the name for the consumer queue among available queues (TaskRQ only)
- collecting and saving all parameters required by the associated job
- running the specific job asyncronously
- a new job can be run either:
- creating a Task from the Django admin
- creating a Task from code, then calling Task.run()
- job execution workflow:
- job execution is triggered by task.run(is_async)
- job will receive the task.id, and retrieve paramerts from it
- on start, job will update task status to 'STARTED' and save job.id for reference
- during execution, the job can update the progress indicator
- on completion, task status is finally updated to either 'SUCCESS' or 'FAILURE'
- See example.jobs.count_beans for an example
- DJANGOTASK_LOG_ROOT
Path for log files.
Default: None
Example: os.path.abspath(os.path.join(BASE_DIR, '..', 'protected', 'tasklog'))
- DJANGOTASK_ALWAYS_EAGER
When True, all task are execute syncronously (useful for debugging and unit testing).
Default: False
- DJANGOTASK_JOB_TRACE_ENABLED
Enables low level tracing in Job.run() - for debugging challenging race conditions
Default: False
- DJANGOTASK_REJECT_IF_NO_WORKER_ACTIVE_FOR_QUEUE
Rejects task if not active worker is available for the specific task queue when task.run() is called
Default: False
- REDIS_URL
Redis server to connect to
Default: 'redis://localhost:6379/0'
The verbosity level controls the logging level as follows:
verbosity | log level |
0 | no log |
1 | logging.WARNING |
2 | logging.INFO |
3 | logging.DEBUG |
and can be set by the derived class:
class MyTask(TaskRQ):
...
DEFAULT_VERBOSITY = 2
...
or you can set it on a "per task" basis by adding to the model a task_verbosity field as follows:
task_verbosity = models.PositiveIntegerField(null=False, blank=False, default=2,
choices=((0,'0'), (1,'1'), (2,'2'), (3,'3')),
)
In case, when creating a management command you might want to preserve the default value provided by the class, or override it if and only if the verbosity option has been set by the user:
import sys
def handle(self, *args, **options):
if '-v' in sys.argv or '--verbosity'in sys.argv:
options['task_verbosity'] = options['verbosity']
self.run_task(MyTask, **options)
Does the code actually work?
Running the unit tests from your project:
python manage.py test -v 2 django_task --settings=django_task.tests.settings
Running the unit tests from your local fork:
source <YOURVIRTUALENV>/bin/activate (myenv) $ pip install tox (myenv) $ tox
or:
python ./runtests.py
or:
coverage run --source='.' runtests.py coverage report
The original implementation is based on django-rq and RQ (a Redis based Python queuing library).
On some occasions, using a background queue may be overkill or even inappropriate: if you need to run many short I/O-bound background tasks concurrently, the serialization provided by the queue, while limiting the usage of resources, would cause eccessive delay.
Starting from version 2.0.0, in those cases you can use TaskThreaded instead of TaskRQ; this way, each background task will run in it's own thread.
MIGRATING FROM django-task 1.5.1 to 2.0.0
- derive your queue-based tasks from TaskRQ instead of Task
- or use TaskThreaded
- get_jobclass() overridable replaces get_jobfunc()
Starting from version 0.3.0, some conveniences have been added:
- The @job decorator for job functions is no more required, as Task.run() now uses queue.enqueue() instead of jobfunc.delay(), and retrieves the queue name directly from the Task itself
- each Task can set it's own TASK_TIMEOUT value (expressed in seconds), that when provided overrides the default queue timeout
- a new Job class has been provided to share suggested common logic before and after jobfunc execution; you can either override run() to implement a custom logic, or (in most cases) just supply your own execute() method, and optionally override on_complete() to execute cleanup actions after job completion;
example:
class CountBeansJob(Job):
@staticmethod
def execute(job, task):
num_beans = task.num_beans
for i in range(0, num_beans):
time.sleep(0.01)
task.set_progress((i + 1) * 100 / num_beans, step=10)
@staticmethod
def on_complete(job, task):
print('task "%s" completed with: %s' % (str(task.id), task.status))
# An more realistic example from a real project ...
# if task.status != 'SUCCESS' or task.error_counter > 0:
# task.alarm = BaseTask.ALARM_STATUS_ALARMED
# task.save(update_fields=['alarm', ])
Execute
Run consumer:
python manage.py runserver
Run worker(s):
python manage.py rqworker low high default
python manage.py rqworker low high default
...
Sample Task
from django.db import models
from django.conf import settings
from django_task.models import TaskRQ
class SendEmailTask(TaskRQ):
sender = models.CharField(max_length=256, null=False, blank=False)
recipients = models.TextField(null=False, blank=False,
help_text='put addresses in separate rows')
subject = models.CharField(max_length=256, null=False, blank=False)
message = models.TextField(null=False, blank=True)
TASK_QUEUE = settings.QUEUE_LOW
TASK_TIMEOUT = 60
LOG_TO_FIELD = True
LOG_TO_FILE = False
DEFAULT_VERBOSITY = 2
@staticmethod
def get_jobclass():
from .jobs import SendEmailJob
return SendEmailJob
When using LOG_TO_FILE = True, you might want to add a cleanup handler to remove the log file when the corresponding record is deleted:
import os from django.dispatch import receiver @receiver(models.signals.post_delete, sender=ImportaCantieriTask) def on_sendemailtask_delete_cleanup(sender, instance, **kwargs): """ Autodelete logfile on Task delete """ logfile = instance._logfile() if os.path.isfile(logfile): os.remove(logfile)
Sample Job
import redis
import logging
import traceback
from django.conf import settings
from .models import SendEmailTask
from django_task.job import Job
class SendEmailJob(Job):
@staticmethod
def execute(job, task):
recipient_list = task.recipients.split()
sender = task.sender.strip()
subject = task.subject.strip()
message = task.message
from django.core.mail import send_mail
send_mail(subject, message, sender, recipient_list)
Sample management command
from django_task.task_command import TaskCommand
from django.contrib.auth import get_user_model
class Command(TaskCommand):
def add_arguments(self, parser):
super(Command, self).add_arguments(parser)
parser.add_argument('sender')
parser.add_argument('subject')
parser.add_argument('message')
parser.add_argument('-r', '--recipients', nargs='*')
parser.add_argument('-u', '--user', type=str, help="Specify username for 'created_by' task field")
def handle(self, *args, **options):
from tasks.models import SendEmailTask
# transform the list of recipents into text
# (one line for each recipient)
options['recipients'] = '\n'.join(options['recipients']) if options['recipients'] is not None else ''
# format multiline message
options['message'] = options['message'].replace('\\n', '\n')
if 'user' in options:
created_by = get_user_model().objects.get(username=options['user'])
else:
created_by = None
self.run_task(SendEmailTask, created_by=created_by, **options)
Deferred Task retrieval to avoid job vs. Task race condition
An helper Task.get_task_from_id() classmethod is supplied to retrieve Task object from task_id safely.
Task queues create a new type of race condition. Why ? Because message queues are fast ! How fast ? Faster than databases.
See:
https://speakerdeck.com/siloraptor/django-tasty-salad-dos-and-donts-using-celery
A similar generic helper is available for Job-derived needs:
django_task.utils.get_model_from_id(model_cls, id, timeout=1000, retry_count=10)
Howto schedule jobs with cron
Call management command 'count_beans', which in turn executes the required job.
For example:
SHELL=/bin/bash PATH=/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin 0 * * * * {{username}} timeout 55m {{django.pythonpath}}/python {{django.website_home}}/manage.py count_beans 1000 >> {{django.logto}}/cron.log 2>&1
A base class TaskCommand has been provided to simplify the creation of any specific task-related management commad;
a derived management command is only responsible for:
- defining suitable command-line parameters
- selecting the specific Task class and job function
for example:
from django_task.task_command import TaskCommand
class Command(TaskCommand):
def add_arguments(self, parser):
super(Command, self).add_arguments(parser)
parser.add_argument('num_beans', type=int)
def handle(self, *args, **options):
from tasks.models import CountBeansTask
self.run_task(CountBeansTask, **options)
A few utility views have been supplied for interacting with tasks from javascript.
Retrieve informations about a list of existing tasks
Sample usage:
var tasks = [{
id: 'c50bf040-a886-4aed-bf41-4ae794db0941',
model: 'tasks.devicetesttask'
}, {
id: 'e567c651-c8d5-4dc7-9cbf-860988f55022',
model: 'tasks.devicetesttask'
}];
$.ajax({
url: '/django_task/info/',
data: JSON.stringify(tasks),
cache: false,
type: 'post',
dataType: 'json',
headers: {'X-CSRFToken': getCookie('csrftoken')}
}).done(function(data) {
console.log('data: %o', data);
});
Result:
[ { "id": "c50bf040-a886-4aed-bf41-4ae794db0941", "created_on": "2018-10-11T17:45:14.399491+00:00", "created_on_display": "10/11/2018 19:45:14", "created_by": "4f943f0b-f5a3-4fd8-bb2e-451d2be107e2", "started_on": null, "started_on_display": "", "completed_on": null, "completed_on_display": "", "job_id": "", "status": "PENDING", "status_display": "<div class=\"task_status\" data-task-model=\"tasks.devicetesttask\" data-task-id=\"c50bf040-a886-4aed-bf41-4ae794db0941\" data-task-status=\"PENDING\" data-task-complete=\"0\">PENDING</div>", "log_link_display": "", "failure_reason": "", "progress": null, "progress_display": "-", "completed": false, "duration": null, "duration_display": "", "extra_fields": { } }, ... ]
Create and run a new task based on specified parameters
Expected parameters:
- 'task-model' = "<app_name>.<model_name>"
- ... task parameters ...
Returns the id of the new task.
Sample usage:
function exportAcquisition(object_id) {
if (confirm('Do you want to export data ?')) {
var url = '/django_task/add/';
var data = JSON.stringify({
'task-model': 'tasks.exportdatatask',
'source': 'backend.acquisition',
'object_id': object_id
});
$.ajax({
type: 'POST',
url: url,
data: data,
cache: false,
crossDomain: true,
dataType: 'json',
headers: {'X-CSRFToken': getCookie('csrftoken')}
}).done(function(data) {
console.log('data: %o', data);
alert('New task created: "' + data.task_id + '"');
}).fail(function(jqXHR, textStatus, errorThrown) {
console.log('ERROR: ' + jqXHR.responseText);
alert(errorThrown);
});
}
return;
}
Schedule execution of specified task.
Returns job.id or throws error (400).
Parameters:
- app_label
- model_name
- pk
- is_async (0 or 1, default=1)
Sample usage:
var task_id = 'c50bf040-a886-4aed-bf41-4ae794db0941';
$.ajax({
url: sprintf('/django_task/tasks/devicetesttask/%s/run/', task_id),
cache: false,
type: 'get'
}).done(function(data) {
console.log('data: %o', data);
}).fail(function(jqXHR, textStatus, errorThrown) {
display_server_error(jqXHR.responseText);
});
The list of Tasks in the admin changelist_view is automatically updated to refresh the progess and status of each running Task.
You can obtain the same result in the frontend by calling the DjangoTask.update_tasks() javascript helper, provided you're listing the tasks in an HTML table with a similar layout.
The simplest way to do it is to use the render_task_column_names_as_table_row and render_task_as_table_row template tags.
Example:
{% load i18n django_task_tags %}
{% if not export_data_tasks %}
<div>{% trans 'No recent jobs available' %}</div>
{% else %}
<table id="export_data_tasks" class="table table-striped">
{% with excluded='created_by,created_on,job_id,log_text,mode' %}
<thead>
<tr>
{{ export_data_tasks.0|render_task_column_names_as_table_row:excluded }}
</tr>
</thead>
<tbody>
{% for task in export_data_tasks %}
<tr>
{{ task|render_task_as_table_row:excluded }}
</tr>
{% endfor %}
</tbody>
</table>
{% endwith %}
{% endif %}
{% block extrajs %}
{{ block.super }}
<script type="text/javascript" src="{% static 'js/django_task.js' %}"></script>
<script>
$(document).ready(function() {
DjangoTask.update_tasks(1000, '#export_data_tasks');
});
</script>
{% endblock extrajs %}
For each fieldname included in the table rows, render_task_as_table_row will check if a FIELDNAME_display() method is available in the Task model, and in case will use it for rendering the field value; otherwise, the field value will be simply converted into a string.
If the specific derived Task model defines some additional fields (unknown to the base Task model) which need to be updated regularly by DjangoTask.update_tasks(), include them as "extra_fields" as follows:
def as_dict(self):
data = super(ExportDataTask, self).as_dict()
data['extra_fields'] = {
'result_display': mark_safe(self.result_display())
}
return data
As example project is provided as a convenience feature to allow potential users to try the app straight from the app repo without having to create a django project.
Please follow the instructions detailed in file example/README.rst.
References:
- A simple app that provides django integration for RQ (Redis Queue)
- Asynchronous tasks in django with django-rq
- django-rq redux: advanced techniques and tools
- Benchmark: Shared vs. Dedicated Redis Instances
- Django tasty salad - DOs and DON'Ts using Celery by Roberto Rosario
- Can Django do multi-thread works?