django-storages
django-storages copied to clipboard
Error upload file. Use zappa+ lambda + s3 + django2 + RDS.
When I save in local it keeps me well in S3 but when it is running with lamda I get the following error. `
Request Method: | POST |
---|---|
https://s5h4s4t98f.execute-api.us-west-2.amazonaws.com/dev/admin/publicidad/publicidad/add/ | |
2.1.1 | |
ClientError | |
An error occurred (InvalidToken) when calling the PutObject operation: The provided token is malformed or otherwise invalid. | |
/var/runtime/botocore/client.py in _make_api_call, line 612 | |
/var/lang/bin/python3.6 | |
3.6.1 | |
['/var/task', '/var/runtime/awslambda', '/var/runtime', '/var/lang/lib/python36.zip', '/var/lang/lib/python3.6', '/var/lang/lib/python3.6/lib-dynload', '/var/lang/lib/python3.6/site-packages', '/var/task/setuptools-39.1.0-py3.6.egg', '/var/task'] | |
Lun, 24 Sep 2018 00:27:27 -0500 |
`
Try to downgrade the installed version, it works for me.
pip install django-storages==1.6.6
How can I reproduce the issue?
The issue appears to be the new behaviour of django-storages to automatically pull the AWS security token from environment variables.
AWS Lambda provides AWS_SESSION_TOKEN
and AWS_SECURITY_TOKEN
as environment variables, taken from the execution role for Lambda which may not be the same credentials required by django-storages for S3 access. https://docs.aws.amazon.com/lambda/latest/dg/current-supported-versions.html
I was able to fix the issue by subclassing S3Boto3Storage
:
from storages.backends.s3boto3 import S3Boto3Storage
class SecurityTokenWorkaroundS3Boto3Storage(S3Boto3Storage):
def _get_security_token(self):
return None
I suppose the solution needs to be a django setting e.g. AWS_SECURITY_TOKEN_IGNORE_ENVIRONMENT
which, when set to True
won't try and load the security token / session token from environment variables?
While AWS Lambda also populates the environment with AWS_ACCESS_KEY_ID
and AWS_SECRET_ACCESS_KEY
, django-storages allows these to be overridden in django settings. My setup generally has these loaded from the environment, but with a different variable name (e.g. AWS_APP_ACCESS_KEY_ID
). However, it doesn't feel right setting AWS_SESSION_TOKEN
or AWS_SECURITY_TOKEN
in django settings as these will change over time, so probably we should provide a django setting which instead allows the behaviour of loading the security token from the environment to be overridden.
Thanks for the investigation. I’m trying to figure out the cleanest way for this to work for everyone. My use case is obviously different but it’s clear that there are a lot of different way this can be setup. As an initial step I am leaning towards partially reverting b13efd92b3bf3e9967b8e7819224bfcf9. Would you be willing to write up some docs for some ways to configure access?
A different user opened #458 which would remove all of the explicit settings which it seems would break your described use case as well.
@jschneier certainly happy to help.
Looking at b13efd9 -- it seems that the issue had not been a problem in my case because I had explicitly set AWS_ACCESS_KEY_ID
and AWS_SECRET_ACCESS_KEY
in django settings , so at
https://github.com/jschneier/django-storages/commit/b13efd92b3bf3e9967b8e7819224bfcf9abb977e#diff-2be7bb15fa1145a1443e92f28ed4549bL249
if not self.access_key and not self.secret_key:
...
self.security_token = self._get_security_token()
was never called because self.access_key
and self.secret_key
had already been set.
So, for me reverting b13efd9 would fix the issue, but it doesn't fix the underlying problem -- I think other users who aren't explicitly setting AWS_ACCESS_KEY_ID
or AWS_SECRET_ACCESS_KEY
in django settings would see this issue prior to b13efd9. Although, probably most people would have those values set in django settings because they're listed in docs/backends/amazon-S3.rst
without being specified as optional.
Whether or not my original idea of AWS_SECURITY_TOKEN_IGNORE_ENVIRONMENT
is a good approach, it seems like that solution should rather be AWS_IGNORE_ENVIRONMENT_CREDENTIALS
and not load any of the env variables populated by AWS Lambda (AWS_ACCESS_KEY
m AWS_ACCESS_KEY_ID
, AWS_SECRET_KEY
, AWS_SECRET_ACCESS_KEY
, AWS_SESSION_TOKEN
, AWS_SECURITY_TOKEN
).
Alternatively, and probably the proper approach, is to provide permission to the S3 bucket in the AWS Lambda execution role. This might fix my issue without needing change in django-storages, and it would mean that #458 also wouldn't cause any issues. For me it would require updating quite a few production sites, which I don't mind if it's the proper approach, though I wonder how many other people are in the same boat now that Zappa is quite popular.
I'll open a PR and add a section in docs/backends/amazon-S3.rst
documenting any special configuration required for AWS Lambda, but I'll do some testing with AWS Lambda execution roles first.
I ran into this issue too, using the same services/frameworks as @davidllauce. I had AWS_ACCESS_KEY_ID
and AWS_SECRET_ACCESS_KEY
defined in my Django settings.py
, however these were populated based on environment variables that had different names (e.g. DJANGO_AWS_ACCESS_KEY_ID
).
The solution posted by @dpretty to override _get_security_token
in a subclass of S3Boto3Storage
worked for me, but I feel like a more straightforward solution should be provided by django-storages
.
I was able to resolve this by setting the four public access settings seen below to False
.
I'm not sure all four were needed to be set to
False
but I'll look further into it shortly and update here.
The issue appears to be the new behaviour of django-storages to automatically pull the AWS security token from environment variables.
AWS Lambda provides
AWS_SESSION_TOKEN
andAWS_SECURITY_TOKEN
as environment variables, taken from the execution role for Lambda which may not be the same credentials required by django-storages for S3 access. https://docs.aws.amazon.com/lambda/latest/dg/current-supported-versions.htmlI was able to fix the issue by subclassing
S3Boto3Storage
:from storages.backends.s3boto3 import S3Boto3Storage class SecurityTokenWorkaroundS3Boto3Storage(S3Boto3Storage): def _get_security_token(self): return None
this didn't work for my case: zappa + role. Each time the request fails.
Can anyone say if this is connected to setting the following header as named by AWS docs:
Using Temporary Security Credentials If you are signing your request using temporary security credentials (see Making Requests), you must include the corresponding security token in your request by adding the x-amz-security-token header.
When you obtain temporary security credentials using the AWS Security Token Service API, the response includes temporary security credentials and a session token. You provide the session token value in the x-amz-security-token header when you send requests to Amazon S3. For information about the AWS Security Token Service API provided by IAM, go to Action in the AWS Security Token Service API Reference Guide .
from this page: https://docs.aws.amazon.com/AmazonS3/latest/dev/RESTAuthentication.html
Also, my research showed that
AWS_SECRET_ACCESS_KEY
can be different for each AWS lambda instance
AWS_ACCESS_KEY_ID
is the same always
AWS_SESSION_TOKEN
= AWS_SECURITY_TOKEN
are the same always
I was able to resolve this by setting the four public access settings seen below to
False
.I'm not sure all four were needed to be set to
False
but I'll look further into it shortly and update here.
This is exactly how my settings are set for the bucket I have problems with. The problem still appears, but only once in a while. it may happen today, tomorrow or the next month. At the beginning I thought it was related to new versions of boto, but that wasn't the case also.
Is possible to be a problem with old and new credentials of the lambda-role. If the credentials are changed during the instance live span then the old credentials might not be the proper one.
How about adding a new custom class to the lib?
from boto3.session import Session
from storages.utils import lookup_env
from storages.backends.s3boto3 import S3Boto3Storage
class S3Boto3StorageForZappaAWSRole(S3Boto3Storage):
"""
This is required, because AWS Role changes its credentials
from 1 to 12 hours and if the Lambda instance has stored
expired version of them we can't have a successful request
until the Lambda Instance's live has expired
"""
def _get_security_token(self):
"""
Gets the security token to use when accessing S3. Get it from
the environment variables.
"""
return lookup_env(S3Boto3Storage.security_token_names)
def _get_access_keys(self):
"""
Gets the access keys to use when accessing S3. If none is
provided in the settings then get them from the environment
variables.
"""
access_key = lookup_env(S3Boto3Storage.access_key_names)
secret_key = lookup_env(S3Boto3Storage.secret_key_names)
return access_key, secret_key
@property
def connection(self):
connection = getattr(self._connections, 'connection', None)
if connection is None:
access_key, secret_key = self._get_access_keys()
security_token = self._get_security_token()
session = Session()
self._connections.connection = session.resource(
's3',
aws_access_key_id=access_key,
aws_secret_access_key=secret_key,
aws_session_token=security_token,
region_name=self.region_name,
use_ssl=self.use_ssl,
endpoint_url=self.endpoint_url,
config=self.config,
verify=self.verify,
)
return self._connections.connection
Or if my PR is accepted:
from storages.utils import lookup_env
from storages.backends.s3boto3 import S3Boto3Storage
class S3Boto3StorageForZappaAWSRole(S3Boto3Storage):
def _get_security_token(self):
return lookup_env(S3Boto3Storage.security_token_names)
def _get_access_keys(self):
access_key = lookup_env(S3Boto3Storage.access_key_names)
secret_key = lookup_env(S3Boto3Storage.secret_key_names)
return access_key, secret_key
Hi I am facing the same issue even when I have AWS_ACCESS_KEY_ID
and AWS_SECRET_ACCESS_KEY
defined in my django settings, works perfectly on my local machine, fails when run in lambda. Started working after downgrading from django_storage 1.7.1 to django-storages==1.6.6.
I'm facing the same error using Django + Zappa + S3 Storages
I can get this work only by removing the AWS_ACCESS_KEY_ID
and AWS_SECRET_ACCESS_KEY
from my settings.py
The issue appears to be the new behaviour of django-storages to automatically pull the AWS security token from environment variables.
AWS Lambda provides
AWS_SESSION_TOKEN
andAWS_SECURITY_TOKEN
as environment variables, taken from the execution role for Lambda which may not be the same credentials required by django-storages for S3 access. https://docs.aws.amazon.com/lambda/latest/dg/current-supported-versions.htmlI was able to fix the issue by subclassing
S3Boto3Storage
:from storages.backends.s3boto3 import S3Boto3Storage class SecurityTokenWorkaroundS3Boto3Storage(S3Boto3Storage): def _get_security_token(self): return None
This worked for me. My requirements.txt:
boto3==1.14.8
Django==3.0.7
djangorestframework==3.11.0
django-storages==1.9.1
Pillow==7.1.2
psycopg2-binary==2.8.5
sorl-thumbnail==12.6.3
zappa==0.51.0
Thanks @dpretty
It works for me using the tip that @guilhermej said. I am going to use a different settings.py file for local developement and for production environment. Maybe that could be an option
The issue appears to be the new behaviour of django-storages to automatically pull the AWS security token from environment variables.
AWS Lambda provides
AWS_SESSION_TOKEN
andAWS_SECURITY_TOKEN
as environment variables, taken from the execution role for Lambda which may not be the same credentials required by django-storages for S3 access. https://docs.aws.amazon.com/lambda/latest/dg/current-supported-versions.htmlI was able to fix the issue by subclassing
S3Boto3Storage
:from storages.backends.s3boto3 import S3Boto3Storage class SecurityTokenWorkaroundS3Boto3Storage(S3Boto3Storage): def _get_security_token(self): return None
Where do I put this code in django?
Where do I put this code in django?
Did you find a solution to this?
removing AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY solve this problem in my case.
removing AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY solve this problem in my case.
if these attributes removed from settings how would django-storages access s3.
The issue appears to be the new behaviour of django-storages to automatically pull the AWS security token from environment variables.
AWS Lambda provides
AWS_SESSION_TOKEN
andAWS_SECURITY_TOKEN
as environment variables, taken from the execution role for Lambda which may not be the same credentials required by django-storages for S3 access. https://docs.aws.amazon.com/lambda/latest/dg/current-supported-versions.htmlI was able to fix the issue by subclassing
S3Boto3Storage
:from storages.backends.s3boto3 import S3Boto3Storage class SecurityTokenWorkaroundS3Boto3Storage(S3Boto3Storage): def _get_security_token(self): return None
I have added this code in custom_storages.py
and it doesn't work. Where should I add this code?
The issue appears to be the new behaviour of django-storages to automatically pull the AWS security token from environment variables. AWS Lambda provides
AWS_SESSION_TOKEN
andAWS_SECURITY_TOKEN
as environment variables, taken from the execution role for Lambda which may not be the same credentials required by django-storages for S3 access. https://docs.aws.amazon.com/lambda/latest/dg/current-supported-versions.html I was able to fix the issue by subclassingS3Boto3Storage
:from storages.backends.s3boto3 import S3Boto3Storage class SecurityTokenWorkaroundS3Boto3Storage(S3Boto3Storage): def _get_security_token(self): return None
I have added this code in
custom_storages.py
and it doesn't work. Where should I add this code?
You need to add this in whatever file you are mentioning in your settings file as storage backend.
@dpretty you saved the day!
Hi all. I just re-read through this thread. It seems the main thing is we don't want to automatically pull the security token because it can be wrong sometimes. However, other users of Lambda rely on pulling the security token.
Does anyone have a link to documentation about why there are competing use-cases? We can add another storage or a setting once I understand the root problem.
The issue appears to be the new behaviour of django-storages to automatically pull the AWS security token from environment variables.
AWS Lambda provides
AWS_SESSION_TOKEN
andAWS_SECURITY_TOKEN
as environment variables, taken from the execution role for Lambda which may not be the same credentials required by django-storages for S3 access. https://docs.aws.amazon.com/lambda/latest/dg/current-supported-versions.htmlI was able to fix the issue by subclassing
S3Boto3Storage
:from storages.backends.s3boto3 import S3Boto3Storage class SecurityTokenWorkaroundS3Boto3Storage(S3Boto3Storage): def _get_security_token(self): return None
thanks , it work for me
i was able to fix by adding custom storages.py
from django.conf import settings
from storages.backends.s3boto3 import S3Boto3Storage
class SecurityTokenWorkaroundS3Boto3Storage(S3Boto3Storage):
def _get_security_token(self):
return None
class MediaStorage(SecurityTokenWorkaroundS3Boto3Storage):
location = settings.MEDIAFILES_LOCATION
class StaticStorage(SecurityTokenWorkaroundS3Boto3Storage):
location = settings.STATICFILES_LOCATION
I found this to be a policy issue within AWS. I erased my original policy, recreated it with the policy generator, then pasted. Seemed to do the trick for me. Also the setting AWS_QUERYSTRING_AUTH = False
in the settings.py file may help.
The above solution won't work anymore as S3Boto3Storage
has been totally refactored and moved to S3Storage
. The problem with using lambda is AWS_ACCESS_KEY_ID
, AWS_SECRET_ACCESS_KEY
, and other keys are reserved. So the workaround i did is just simply set the credentials using the custom OPTIONS
and never used those reserved keys.
For example:
STORAGES = {
"staticfiles": {
"BACKEND": "storages.backends.s3.S3Storage",
"OPTIONS": {
"access_key": ENV("APP_S3_ACCESS_KEY"),
"secret_key": ENV("APP_S3_SECRET_KEY"),
"bucket_name": ENV("APP_S3_BUCKET_NAME"),
},
},
}