Develop Reference

ElasticBeanstalk suddenly starts failing to deploy Django App with "Cannot use ImageField because Pillow is not installed" exception

I am using elastic beanstalk to deploy my Django application. Today it suddenly stopped working without any breaking changes from the application side (I've changed some templates, nothing more).
The deployment time outs after 10 minutes of trying to deploy the app and nothing happens.
The only more or less useful hints I can see in the log is this:
[2020-02-20T15:00:20.437Z] INFO [19057] - [Application update .../postbuild_0_myproject/Command 01_migrate] : Activity execution failed, because: SystemCheckError: System check identified some issues:
ERRORS:
education.Author.photo: (fields.E210) Cannot use ImageField because Pillow is not installed.
HINT: Get Pillow at https://pypi.org/project/Pillow/ or run command "pip install Pillow".
education.Course.cover_image: (fields.E210) Cannot use ImageField because Pillow is not installed.
HINT: Get Pillow at https://pypi.org/project/Pillow/ or run command "pip install Pillow".
education.CourseCategory.icon_image: (fields.E210) Cannot use ImageField because Pillow is not installed.
HINT: Get Pillow at https://pypi.org/project/Pillow/ or run command "pip install Pillow".
Using staging settings
App receivers connected
(ElasticBeanstalk::ExternalInvocationError)
[2020-02-20T15:00:20.437Z] INFO [19057] - [Application update .../postbuild_0_myproject/Command 01_migrate] : Activity failed.
[2020-02-20T15:00:20.437Z] INFO [19057] - [Application update .../postbuild_0_myproject] : Activity failed.
[2020-02-20T15:00:20.437Z] INFO [19057] - [Application update ...] : Activity failed.
[2020-02-20T15:00:20.507Z] INFO [19057] - [Application update app-9a24-200220_145942-stage-200220_145942#142/AppDeployStage0/EbExtensionPostBuild] : Activity failed.
[2020-02-20T15:00:20.507Z] INFO [19057] - [Application update app-9a24-200220_145942-stage-200220_145942#142/AppDeployStage0] : Activity failed.
[2020-02-20T15:00:20.508Z] INFO [19057] - [Application update app-9a24-200220_145942-stage-200220_145942#142] : Completed activity. Result:
Application update - Command CMD-AppDeploy failed
But I already have Pillow in requirements.txt and the log above says:
Requirement already satisfied: Pillow==6.2.1 in /opt/python/run/venv/lib64/python3.6/site-packages (from -r /opt/python/ondeck/app/requirements.txt (line 51))
How can I troubleshoot and fix this? And how can I avoid similar issues in the future? I am really frightened that the same problem may randomly pop out on production environment.
Here's some more info about the configuration:
Here's what I have in .ebextensions:
01_packages.config:
packages:
yum:
git: []
postgresql93-devel: []
db-migrate.config
container_commands:
01_migrate:
command: "django-admin.py migrate"
leader_only: true
option_settings:
aws:elasticbeanstalk:application:environment:
DJANGO_SETTINGS_MODULE: myproject.settings
django.config
option_settings:
aws:elasticbeanstalk:container:python:
WSGIPath: myproject/wsgi.py
wsgi_custom.config
files:
"/etc/httpd/conf.d/wsgihacks.conf":
mode: "000644"
owner: root
group: root
content: |
WSGIPassAuthorization On

This one is a pain and a known issue with Django when using the ImageField model/form. Due to Pythons dynamic import system it will suddenly appear and annoyed the hell out of me when I first came across it.
The way I normally fix this is by using conda and its equivalent of a virtualenv to ensure the right interpreter (the one with my packages) is used.
If you are not using a virtualenv or equivalent, set one up now, if you are already using one then check you are installing pillow with pip3 install pillow - the pip3 being important here as on debian (and many other) systems normal pip will only install for python 2.x.
Using conda will ensure this doesnt happen in production, but I would also add it to your checklist of things to test when deploying - check correct version of pillow setup and working.

I had two Elastic Beanstalk environments with the same issue (one web tier env and a worker env).
On one of them the issue was resolved by restarting the environment.
The other one failed to restart and timed out every time on any operation. This one I managed to fix by going to configuration > capacity and changing the minimum and maximum number of instances to 0. I've applied the changes, waited for them to apply and then returned the previous values for min and max instance numbers.
That fixed the issue.
I still have no idea what caused the issue in the first place and would love to receive some comment on that.

Python 3 flask install wkhtmltopdf on heroku

I have a problem to install the wkhtmltopdf binary on my heroku python app (flask).
A year ago (python 2) I already had an issue, but I was able to solve it by first adding the wkhtmltopdf-pack to the requirements and installing it on heroku and then setting the config var to WKHTMLTOPDF_BINARY=wkhtmltopdf-pack. Here is my old thread
Problem now:
I am trying to use the same approach for python 3, but no version of the wkhtmltopdf-pack works, every push gets rejected and I cant install it.
I tried these versions in the requirements:
wkhtmltopdf-pack==0.12.5
wkhtmltopdf-pack==0.12.4
wkhtmltopdf-pack==0.12.3
wkhtmltopdf-pack==0.12.3.0.post1
wkhtmltopdf-pack==0.12.2.4
I get these errors:
No matching distribution
or
error: can't copy 'bin/wkhtmltopdf-pack': doesn't exist or not a regular file
and I remember once it told me there was a SyntaxError and it could not decode something.
Alternative approach:
It seems it is also possible to use a buildpack, so I tried adding a buildpack:
heroku buildpacks:add https://github.com/dscout/wkhtmltopdf-buildpack.git
I see that the buildpack has been added, but there was no installation and there is also no config var for wkhtmltopdf. I dont understand how to trigger the installation, in all documantations for buildpacks its written "add the buildpack and you are ready to go".
Trying to create a PDF gives me a server error here:
OSError: No wkhtmltopdf executable found: "b''"
EDIT:
I managed to install the buildpack:
The push was successful, but no config var has been created and I have no clue what the path to the binary is.
EDIT
I was able to find the files through heroku bash:
app bin dev etc lib lib64 lost+found proc sbin sys tmp usr var
/ $ cd app
~ $ cd vendor
~/vendor $ dir
wkhtmltox
~/vendor $ cd wkhtmltox
~/vendor/wkhtmltox $ dir
lib
~/vendor/wkhtmltox $ cd lib
~/vendor/wkhtmltox/lib $ dir
libwkhtmltox.so libwkhtmltox.so.0 libwkhtmltox.so.0.12 libwkhtmltox.so.0.12.3
~/vendor/wkhtmltox/lib $ exit
Now I tried to all these files but all give an error:
OSError: wkhtmltopdf exited with non-zero code -11. error
Here is how I set the path:
# WKHTMLTOPDF config
if 'DYNO' in os.environ:
print ('loading wkhtmltopdf path on heroku')
MYDIR = os.path.dirname(__file__)
WKHTMLTOPDF_CMD = os.path.join(MYDIR + "/vendor/wkhtmltox/lib/", "libwkhtmltox.so")
else:
print ('loading wkhtmltopdf path on localhost')
MYDIR = os.path.dirname(__file__)
WKHTMLTOPDF_CMD = os.path.join(MYDIR + "/static/executables/bin/", "wkhtmltopdf.exe")

The best approach to get installed wkhtmltopdf on Heroku by getting the binary of wkhtmltopdf for python 3 instead of wkhtmltopdf-pack and you can achieve this by using pydf.
You can install it simply using pip like:
pip install python-pdf
or for Python 2:
pip install python-pdf==0.30.0
Unlike the buildpack based approach pydf installs with the wkhtmltopdf binary included making it very easy to use, and this is the right approach for Heroku.
But if you still want to stick with build-pack wkhtmltopdf, here's another solution you can give it a try:
Via: CLI Installation
$ heroku create --buildpack https://github.com/homelight/wkhtmltox-buildpack.git
Or Manually:
Add the following line to your .buildpacks file
https://github.com/homelight/wkhtmltox-buildpack.git
Please note that this buildpack is only compatible with the cedar-14 stack. You can use heroku stack:set cedar-14 to set the correct stack.

I was able to solve the problem on my own, following my first approach.
I found an other wkhtmltopdf-pack on pypi and added it to my requirements.txt:
wkhtmltopdf-pack-ng==0.12.3.0
Heroku was able to install this pack.
After that I added the config var for wkhtmltopdf:
heroku config:set WKHTMLTOPDF_BINARY=wkhtmltopdf-pack
The installation is now complete. I need to use the correct path now on my app:
if 'DYNO' in os.environ:
print ('loading wkhtmltopdf path on heroku')
WKHTMLTOPDF_CMD = subprocess.Popen(
['which', os.environ.get('WKHTMLTOPDF_BINARY', 'wkhtmltopdf-pack')], # Note we default to 'wkhtmltopdf' as the binary name
stdout=subprocess.PIPE).communicate()[0].strip()
else:
print ('loading wkhtmltopdf path on localhost')
MYDIR = os.path.dirname(__file__)
WKHTMLTOPDF_CMD = os.path.join(MYDIR + "/static/executables/bin/", "wkhtmltopdf.exe")
Thats it.

Heroku python Tutorial Won't run locally in windows

I'm trying to run the python heroku tutorial and it won't work in windows. This is from this repository.
I posted this previously but I was able to get a more descriptive error message. It should be said that I've installed postgres.
Furthermore, I can't run it locally using the method defined in the git respository. Both the createdb and foreman commands don't work. This is despite installing foreman.
django.core.exceptions.ImproperlyConfigured: 'django_postgrespool' isn't an available database backend.
Try using 'django.db.backends.XXX', where XXX is one of:
u'base', u'mysql', u'oracle', u'postgresql_psycopg2', u'sqlite3'
Error was: DLL load failed: The specified module could not be found.

Looks like python does not know what django-postgrespool is.
Perhaps it did not install properly. Check the output of pip install -r requirements.txt

DATABASES['default']['ENGINE'] = 'django_postgrespool'
in your settings.py is what this is referring to. For me i still dont know why this is causing a problem. I've install psycopg2 and I even tried installing 'pip install django-postgrespool'. It worked once i commented out:
DATABASES['default'] = dj_database_url.config()
DATABASES['default']['ENGINE'] = 'django_postgrespool'
this helps me run the app locally by using the
heroku local web -f Procfile.windows

I was following the same tutorial and what worked for me was changing the settings.py file, this line:
# Enable Connection Pooling (if desired)
DATABASES['default']['ENGINE'] = 'django_postgrespool'
To this:
# Enable Connection Pooling (if desired)
DATABASES['default']['ENGINE'] = 'django.db.backends.postgresql_psycopg2'

Python web application with OpenCV in Heroku

I am building a web application that uses OpenCV in its back-end. I have built the application on Ubuntu (and I tried it on Windows, too) and it works fine. Currently, I am trying to configure OpenCV to work on Heroku. As OpenCV is not possible to be loaded using pip, I read about using heroku buildpacks which provide customization for the server environment.
The following is my attempt to test two of OpenCV buildpacks:
I build simple web server with Flask that tries to import OpenCV:
#hello.py
import os
from flask import Flask
app = Flask(__name__)
#app.route("/")
def hello():
text = ''
try:
import cv2
text = 'success'
except:
text = 'fail'
pass
return text + ' to load openCV'
if __name__ == "__main__":
port = int(os.environ.get("PORT", 5000))
app.run(host='0.0.0.0', port=port)
The above code should return either success or fail in loading OpenCV.
Then I configured Heroku to use (heroku multi buildpack) by running the following command:
heroku buildpacks:set https://github.com/ddollar/heroku-buildpack-multi
In the .buildpacks file (that is required by multi buildpack) I put the https://github.com/heroku/heroku-buildpack-python and https://github.com/slobdell/heroku-buildpack-python-opencv-scipy buildpacks.
The first one is for compiling a python application and for installing other modules (e.g., Flask) through pip. The second buildpack is the one that is supposed to load OpenCV.
After all, the whole application did not work!
I got (Application Error) page in Heroku as following screenshot:
I tried to use other buildpack (https://github.com/diogojc/heroku-buildpack-python-opencv-scipy) but I got the same result.
My questions are:
What is wrong in the steps I did?
How should I call (or use) OpenCV within my application in heroku?Should I use import statement or some other commands?

I could install by doing as follows:
cd /path/to/your/dir && git init
heroku create MYAPP (start from scratch)
heroku config:add BUILDPACK_URL=https://github.com/ddollar/heroku-buildpack-multi.git --app MYAPP
create .buildpacks as follows:
https://github.com/heroku/heroku-buildpack-python
https://github.com/diogojc/heroku-buildpack-python-opencv-scipy#cedar14
git add . && git commit -m 'MESSAGE' && git push heroku master

For anyone seeing this today and having the same issue, switch opencv-python in your requirements.txt to opencv-python-headless. This sidesteps the problem with the problematic library file.

The following steps should solve the problem of openCV which you are facing -
Add the heroku-buildpack-apt to the BuildPack by pasting - https://github.com/heroku/heroku-buildpack-apt to add buildpack in dasboard.
ScreenShot -
Adding through Dashboard -> Settings -> Add BuildPacks
Then add the Aptfile in your Github base directory which contains -
libsm6
libxrender1
libfontconfig1
libice6
- one library in each line. See Example Github Link
Now build and deploy and you are ready to go!

Global paster command not found in virtualenv

I created a custom paster command as described in http://pythonpaste.org/script/developer.html#what-do-commands-look-like. In my setup.py I have defined the entry point like this:
entry_points={
'paste.global_paster_command' : [
'xxx_new = xxxconf.main:NewXxx'
]
}
I'm inside an activated virtualenv and have installed my package via
python setup.py develop
If I run paster while inside my package folder, I see my custom command and I can run it via paster xxx .... But if I leave my package folder paster does not display my command anymore. I checked which paster and it's the version of my virtualenv. I also started a python interpreter and imported xxxconf and it works fine.
I have no idea why my global command is not recognized when I'm outside my package folder!?

You are doing something wrong, it should work. This is the minimal working example, you can test it with your virtualenv:
blah/setup.py:
from setuptools import setup, find_packages
setup(name='blah',
version='0.1',
packages=find_packages(exclude=['ez_setup', 'examples', 'tests']),
include_package_data=True,
zip_safe=False,
entry_points={'paste.global_paster_command': [ "xxx_new = blah.xxx:NewXxx", ] },
)
blah/blah/xxx.py:
from paste.script import command
class NewXxx(command.Command):
usage = "PREFIX"
summary = "some command"
group_name = "my group"
blah/blah/__init__.py: empty.
Now testing:
$ pwd
/tmp
$ virtualenv paster
New python executable in paster/bin/python
Installing setuptools............done.
Installing pip...............done.
$ . paster/bin/activate
(paster)$ pip install PasteScript
Downloading/unpacking PasteScript
[... skipping long pip output here ...]
(paster)$ paster
[...]
Commands:
create Create the file layout for a Python distribution
help Display help
make-config Install a package and create a fresh config file/directory
points Show information about entry points
post Run a request for the described application
request Run a request for the described application
serve Serve the described application
setup-app Setup an application, given a config file
(paster)$ cd blah/
(paster)$ python setup.py develop
running develop
[... skipping setup.py output...]
(paster)$ paster
[...]
Commands:
create Create the file layout for a Python distribution
help Display help
make-config Install a package and create a fresh config file/directory
points Show information about entry points
post Run a request for the described application
request Run a request for the described application
serve Serve the described application
setup-app Setup an application, given a config file
my group:
xxx_new some command
(paster)$ cd ~
(paster)$ paster
[...]
Commands:
[...]
setup-app Setup an application, given a config file
my group:
xxx_new some command

You should install your paster_script in the active virtualenv. Then you can use it anywhere.