Django生成Excel并用芹菜和RabbitMQ保存为对象
我正在使用Django 2.2.5,芹菜5.2.6和Rabbitmq,并且是最后2个新手。
我想生成Excel表并将其存储在新创建的对象中(不下载)中,这是我做了什么:
project/settings.py:
CELERY_RESULT_BACKEND = "django-db"
CELERY_CACHE_BACKEND = "django-cache"
CELERY_BROKER_URL = "amqp://localhost"
CELERY_ACCEPT_CONTENT = ["application/json"]
CELERY_TASK_SERIALIZER = "json"
CELERY_RESULT_SERIALIZER = "json"
CELERY_TIMEZONE = "Africa/Casablanca"
project/celery.py:
from __future__ import absolute_import, unicode_literals
import os
from celery import Celery
os.environ.setdefault("DJANGO_SETTINGS_MODULE", "verautonoapi.settings")
app = Celery("verautonoapi")
app.config_from_object("django.conf:settings", namespace="CELERY")
app.autodiscover_tasks()
@app.task(bind=True)
def debug_task(self):
print("Request: {0!r}".format(self.request))
project/ init .py:
from __future__ import absolute_import, unicode_literals
# This will make sure the app is always imported when
# Django starts so that shared_task will use this app.
from .celery import app as celery_app
__all__ = ("celery_app",)
app/tasks.py:
from __future__ import absolute_import, unicode_literals
from django.http import HttpResponse, FileResponse
from celery import shared_task
from celery.utils.log import get_task_logger
from time import sleep
from openpyxl import Workbook
from datetime import date
from .models import Dossier, Export
logger = get_task_logger(__name__)
@shared_task(bind=True, track_started=True)
def exp_en_cours(request):
print("task activated")
sleep(7)
print("now starting generating the file")
dossier_queryset = Dossier.objects.filter(status_admin="Validation OR Requise")
today = str(date.today())
response = FileResponse(
content_type="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
)
response["Content-Disposition"] = (
"attachment; filename=" + today + "-dossiers_en_cours.xlsx"
)
workbook = Workbook()
# Get active worksheet/tab
worksheet = workbook.active
worksheet.title = "Dossiers en cours"
# Define the titles for columns
columns = [
"Numéro",
"Créé le",
"Assurance",
"Prestataire",
"Matricule",
]
row_num = 1
# Assign the titles for each cell of the header
for col_num, column_title in enumerate(columns, 1):
cell = worksheet.cell(row=row_num, column=col_num)
cell.value = column_title
# Iterate through all movies
for dossier in dossier_queryset:
row_num += 1
# Define the data for each cell in the row
row = [
dossier.numero,
str(dossier.date_posted)[:10],
dossier.assurance.name,
dossier.created_by.user.username,
dossier.matricule,
]
# Assign the data for each cell of the row
for col_num, cell_value in enumerate(row, 1):
cell = worksheet.cell(row=row_num, column=col_num)
cell.value = cell_value
workbook.save(response)
Export.objects.create(status="Dossier en cours", file=response)
return "OK"
app/models.py:
class Export(models.Model):
STATUS = [
("Dossier en cours", "Dossier en cours"),
("Dossier validés", "Dossier validés"),
("Dossier rejetés", "Dossier rejetés"),
("Dossier à clôturer", "Dossier à clôturer"),
("Dossier clôturés", "Dossier clôturés"),
("Dossier facturés", "Dossier facturés"),
("Dossier archivés", "Dossier archivés"),
]
status = models.CharField(max_length=20, choices=STATUS, null=True, blank=True)
timestamp = models.DateTimeField(auto_now_add=True, editable=False)
file = models.FileField(
upload_to=get_rapport_filename, verbose_name="Fichiers Excel"
)
def __str__(self):
return "%s - %s" % (self.dossier, self.timestamp)
class Meta:
verbose_name = "Rapport Excel"
verbose_name_plural = "Rapports Excel"
我还安装了django_celery_results 并将其添加到已安装的应用程序中。
当我执行任务时,它会产生以下错误:
This FileResponse instance is not writable
你能告诉我我在做什么错吗? Excel功能不正确吗?因为我确实进行了更改,以防止它下载(用fileerSponse替换了httpresponse)。还是兔子和芹菜?
更新:问题在于Excel功能,而不是芹菜/兔子。我复制了任务并将其作为常规视图,并获得了相同的错误。
更新2 :我修改了视图以使该文件作为字节流并使用ContentFile将其放入模型中,我能够下载该文件,但是它不能打开:
Views.py:
dossier_queryset = Dossier.objects.filter(status_admin="Validation OR Requise")
# today = str(date.today())
# response = FileResponse(
# content_type="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
# )
# response["Content-Disposition"] = (
# "attachment; filename=" + today + "-dossiers_en_cours.xlsx"
# )
workbook = Workbook()
# Get active worksheet/tab
worksheet = workbook.active
worksheet.title = "Dossiers en cours"
# Define the titles for columns
columns = [
"Numéro",
"Créé le",
"Assurance",
"Prestataire",
"Matricule",
]
row_num = 1
# Assign the titles for each cell of the header
for col_num, column_title in enumerate(columns, 1):
cell = worksheet.cell(row=row_num, column=col_num)
cell.value = column_title
# Iterate through all movies
for dossier in dossier_queryset:
row_num += 1
# Define the data for each cell in the row
row = [
dossier.numero,
str(dossier.date_posted)[:10],
dossier.assurance.name,
dossier.created_by.user.username,
dossier.matricule,
]
# Assign the data for each cell of the row
for col_num, cell_value in enumerate(row, 1):
cell = worksheet.cell(row=row_num, column=col_num)
cell.value = cell_value
vworkbook = BytesIO()
workbook.save(vworkbook)
content = vworkbook.getvalue()
b64 = base64.b64encode(content)
file = (
"data:application/vnd.openxmlformats-officedocument.spreadsheetml.sheet;base64,"
+ str(b64)
)
format, filestr = file.split(";base64,")
# ext = format.split("/")[-1]
ext = ".xlsx"
Export.objects.create(
status="Dossier en cours",
file=ContentFile(base64.b64decode(filestr), name="temp." + ext),
)
return redirect("homepage")
I am using Django 2.2.5, Celery 5.2.6 and RabbitMQ and I am new to the last 2.
I want to generate an Excel sheet and store it in a FileField inside a newly created object (not download it), and this is what I did:
project/settings.py:
CELERY_RESULT_BACKEND = "django-db"
CELERY_CACHE_BACKEND = "django-cache"
CELERY_BROKER_URL = "amqp://localhost"
CELERY_ACCEPT_CONTENT = ["application/json"]
CELERY_TASK_SERIALIZER = "json"
CELERY_RESULT_SERIALIZER = "json"
CELERY_TIMEZONE = "Africa/Casablanca"
project/celery.py:
from __future__ import absolute_import, unicode_literals
import os
from celery import Celery
os.environ.setdefault("DJANGO_SETTINGS_MODULE", "verautonoapi.settings")
app = Celery("verautonoapi")
app.config_from_object("django.conf:settings", namespace="CELERY")
app.autodiscover_tasks()
@app.task(bind=True)
def debug_task(self):
print("Request: {0!r}".format(self.request))
project/init.py :
from __future__ import absolute_import, unicode_literals
# This will make sure the app is always imported when
# Django starts so that shared_task will use this app.
from .celery import app as celery_app
__all__ = ("celery_app",)
app/tasks.py:
from __future__ import absolute_import, unicode_literals
from django.http import HttpResponse, FileResponse
from celery import shared_task
from celery.utils.log import get_task_logger
from time import sleep
from openpyxl import Workbook
from datetime import date
from .models import Dossier, Export
logger = get_task_logger(__name__)
@shared_task(bind=True, track_started=True)
def exp_en_cours(request):
print("task activated")
sleep(7)
print("now starting generating the file")
dossier_queryset = Dossier.objects.filter(status_admin="Validation OR Requise")
today = str(date.today())
response = FileResponse(
content_type="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
)
response["Content-Disposition"] = (
"attachment; filename=" + today + "-dossiers_en_cours.xlsx"
)
workbook = Workbook()
# Get active worksheet/tab
worksheet = workbook.active
worksheet.title = "Dossiers en cours"
# Define the titles for columns
columns = [
"Numéro",
"Créé le",
"Assurance",
"Prestataire",
"Matricule",
]
row_num = 1
# Assign the titles for each cell of the header
for col_num, column_title in enumerate(columns, 1):
cell = worksheet.cell(row=row_num, column=col_num)
cell.value = column_title
# Iterate through all movies
for dossier in dossier_queryset:
row_num += 1
# Define the data for each cell in the row
row = [
dossier.numero,
str(dossier.date_posted)[:10],
dossier.assurance.name,
dossier.created_by.user.username,
dossier.matricule,
]
# Assign the data for each cell of the row
for col_num, cell_value in enumerate(row, 1):
cell = worksheet.cell(row=row_num, column=col_num)
cell.value = cell_value
workbook.save(response)
Export.objects.create(status="Dossier en cours", file=response)
return "OK"
app/models.py:
class Export(models.Model):
STATUS = [
("Dossier en cours", "Dossier en cours"),
("Dossier validés", "Dossier validés"),
("Dossier rejetés", "Dossier rejetés"),
("Dossier à clôturer", "Dossier à clôturer"),
("Dossier clôturés", "Dossier clôturés"),
("Dossier facturés", "Dossier facturés"),
("Dossier archivés", "Dossier archivés"),
]
status = models.CharField(max_length=20, choices=STATUS, null=True, blank=True)
timestamp = models.DateTimeField(auto_now_add=True, editable=False)
file = models.FileField(
upload_to=get_rapport_filename, verbose_name="Fichiers Excel"
)
def __str__(self):
return "%s - %s" % (self.dossier, self.timestamp)
class Meta:
verbose_name = "Rapport Excel"
verbose_name_plural = "Rapports Excel"
I also installed django_celery_results
and added it to installed apps.
When I execute the task, it gives the following error:
This FileResponse instance is not writable
I tried on both Windows and Linux (on digitalocean) and I get the same error. Can you tell me what am I doing wrong? is the excel function not correct? Because I did make a change in it to prevent it from downloading (replaced HttpResponse with FileResponse). Or is it RabbitMQ and Celery?
UPDATE: The problem is in the Excel function, not Celery/RabbitMQ. I copied the task and put it as a regular view and I obtain the same error.
UPDATE 2: I modified the view to be able to get the file as a BytesIO stream and use ContentFile to put it inside a model, and I was able to download the file, but it cannot be open:
views.py:
dossier_queryset = Dossier.objects.filter(status_admin="Validation OR Requise")
# today = str(date.today())
# response = FileResponse(
# content_type="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
# )
# response["Content-Disposition"] = (
# "attachment; filename=" + today + "-dossiers_en_cours.xlsx"
# )
workbook = Workbook()
# Get active worksheet/tab
worksheet = workbook.active
worksheet.title = "Dossiers en cours"
# Define the titles for columns
columns = [
"Numéro",
"Créé le",
"Assurance",
"Prestataire",
"Matricule",
]
row_num = 1
# Assign the titles for each cell of the header
for col_num, column_title in enumerate(columns, 1):
cell = worksheet.cell(row=row_num, column=col_num)
cell.value = column_title
# Iterate through all movies
for dossier in dossier_queryset:
row_num += 1
# Define the data for each cell in the row
row = [
dossier.numero,
str(dossier.date_posted)[:10],
dossier.assurance.name,
dossier.created_by.user.username,
dossier.matricule,
]
# Assign the data for each cell of the row
for col_num, cell_value in enumerate(row, 1):
cell = worksheet.cell(row=row_num, column=col_num)
cell.value = cell_value
vworkbook = BytesIO()
workbook.save(vworkbook)
content = vworkbook.getvalue()
b64 = base64.b64encode(content)
file = (
"data:application/vnd.openxmlformats-officedocument.spreadsheetml.sheet;base64,"
+ str(b64)
)
format, filestr = file.split(";base64,")
# ext = format.split("/")[-1]
ext = ".xlsx"
Export.objects.create(
status="Dossier en cours",
file=ContentFile(base64.b64decode(filestr), name="temp." + ext),
)
return redirect("homepage")
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。

绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论
评论(1)
这是解决方案:
Here's the solution: