Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.
Sign upAdd AWS4 support in the S3_RGW_OUTSCALE provider #736
Conversation
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
tonybaloney
Apr 5, 2016
Contributor
Hey @jmunhoz I remember the AWS4 auth algorithm works with GET and POST, but I'm not so sure about PUT. Can you please check that you can upload files using these changes
|
Hey @jmunhoz I remember the AWS4 auth algorithm works with GET and POST, but I'm not so sure about PUT. Can you please check that you can upload files using these changes |
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
jmunhoz
Apr 6, 2016
Contributor
@tonybaloney I would expect S3RGWOutscaleConnectionAWS4 (https://github.com/jmunhoz/libcloud/blob/s3-rgw-outscale-aws4/libcloud/storage/drivers/s3.py#L1021) having the same behaviour as S3APNE2Connection (https://github.com/jmunhoz/libcloud/blob/s3-rgw-outscale-aws4/libcloud/storage/drivers/s3.py#L987). I ran a smoke test listing objects (GET) and the testing suite. They were ok.
As suggested I will check the PUT case explicitly ASAP. Thanks!
|
@tonybaloney I would expect S3RGWOutscaleConnectionAWS4 (https://github.com/jmunhoz/libcloud/blob/s3-rgw-outscale-aws4/libcloud/storage/drivers/s3.py#L1021) having the same behaviour as S3APNE2Connection (https://github.com/jmunhoz/libcloud/blob/s3-rgw-outscale-aws4/libcloud/storage/drivers/s3.py#L987). I ran a smoke test listing objects (GET) and the testing suite. They were ok. As suggested I will check the PUT case explicitly ASAP. Thanks! |
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
jmunhoz
May 1, 2016
Contributor
@tonybaloney I reviewed the PUT case. The Libcloud code required one minor fix to handle the different behaviour related to ETags in the Amazon S3 and Ceph RGW implementations. Ceph RGW is not shipping ETag headers in this case so the code needs to handle the case where ETags don't exist.
BTW, Ceph RGW was not handling the case for unsigned payloads in the PUT method so I pushed this support in Ceph and it is upstream now with Jewel 10.2.0
The PUT method is working now. Tests are passing. The header dumps for Amazon S3 and Ceph RGW using the new patch follow...
Amazon S3 --------- PUT /my-container-12345678901234/my-name-12345678901234 HTTP/1.1 X-AMZ-Content-SHA256: UNSIGNED-PAYLOAD Content-Length: 12 Accept-Encoding: gzip,deflate X-AMZ-Date: 20160501T164719Z x-amz-storage-class: STANDARD Content-Type: text/plain Host: s3.eu-central-1.amazonaws.com Authorization: AWS4-HMAC-SHA256 Credential=12345678901234567890/20160502/eu-central-1/s3/aws4_request, SignedHeaders=accept-encoding;content-length;content-type;host;user-agent;x-amz-content-sha256;x-amz-date;x-amz-storage-class, Signature=3200c1e45af4ef5acde510877d1b78541337b7c90f420012d694a04a13cebcd5 User-Agent: libcloud/1.0.0-rc2 (OUTSCALE Ceph RGW S3 (eu-central-1)) Hello test! HTTP/1.1 200 OK x-amz-id-2: FiEe9rH1Rmgb7+palW18yvo6cVNt1TE90UbLwC3syxRiu0DNuDWMwzGSZMQT0ao7srkGyqOmQi4= x-amz-request-id: CC7007FFB4A71C44 Date: Sun, 01 May 2016 16:47:20 GMT ETag: "8ac91dd17dd64e95271601b773859aaa" Content-Length: 0 Server: AmazonS3 Ceph RGW --------- PUT /my-container-12345678901234/my-name-12345678901234 HTTP/1.1 X-AMZ-Content-SHA256: UNSIGNED-PAYLOAD Content-Length: 12 Accept-Encoding: gzip,deflate X-AMZ-Date: 20160501T185440Z x-amz-storage-class: STANDARD Content-Type: text/plain Host: osu.eu-west-1.outscale.com Authorization: AWS4-HMAC-SHA256 Credential=12345678901234567890/20160502/eu-west-1/s3/aws4_request, SignedHeaders=accept-encoding;content-length;content-type;host;user-agent;x-amz-content-sha256;x-amz-date;x-amz-storage-class, Signature=fb7503d5014ce581723d9695a1395bb668a3a5fcee713663de3d7a316d9b4b27 User-Agent: libcloud/1.0.0-rc2 (OUTSCALE Ceph RGW S3 (eu-west-1)) Hello test! HTTP/1.1 200 OK ETag: "8ac91dd17dd64e95271601b773859aaa" Content-Length: 0 Accept-Ranges: bytes x-amz-request-id: tx000000000000000000012-00572650f0-1017-default Date: Sun, 01 May 2016 18:54:40 GMT
jmunhoz
commented
May 1, 2016
•
edited
Edited 1 time
-
jmunhoz
edited May 1, 2016 (most recent)
edited
-
May 1, 2016 (most recent)
|
@tonybaloney I reviewed the PUT case. The Libcloud code required one minor fix to handle the different behaviour related to ETags in the Amazon S3 and Ceph RGW implementations. Ceph RGW is not shipping ETag headers in this case so the code needs to handle the case where ETags don't exist. BTW, Ceph RGW was not handling the case for unsigned payloads in the PUT method so I pushed this support in Ceph and it is upstream now with Jewel 10.2.0 The PUT method is working now. Tests are passing. The header dumps for Amazon S3 and Ceph RGW using the new patch follow... Amazon S3 --------- PUT /my-container-12345678901234/my-name-12345678901234 HTTP/1.1 X-AMZ-Content-SHA256: UNSIGNED-PAYLOAD Content-Length: 12 Accept-Encoding: gzip,deflate X-AMZ-Date: 20160501T164719Z x-amz-storage-class: STANDARD Content-Type: text/plain Host: s3.eu-central-1.amazonaws.com Authorization: AWS4-HMAC-SHA256 Credential=12345678901234567890/20160502/eu-central-1/s3/aws4_request, SignedHeaders=accept-encoding;content-length;content-type;host;user-agent;x-amz-content-sha256;x-amz-date;x-amz-storage-class, Signature=3200c1e45af4ef5acde510877d1b78541337b7c90f420012d694a04a13cebcd5 User-Agent: libcloud/1.0.0-rc2 (OUTSCALE Ceph RGW S3 (eu-central-1)) Hello test! HTTP/1.1 200 OK x-amz-id-2: FiEe9rH1Rmgb7+palW18yvo6cVNt1TE90UbLwC3syxRiu0DNuDWMwzGSZMQT0ao7srkGyqOmQi4= x-amz-request-id: CC7007FFB4A71C44 Date: Sun, 01 May 2016 16:47:20 GMT ETag: "8ac91dd17dd64e95271601b773859aaa" Content-Length: 0 Server: AmazonS3 Ceph RGW --------- PUT /my-container-12345678901234/my-name-12345678901234 HTTP/1.1 X-AMZ-Content-SHA256: UNSIGNED-PAYLOAD Content-Length: 12 Accept-Encoding: gzip,deflate X-AMZ-Date: 20160501T185440Z x-amz-storage-class: STANDARD Content-Type: text/plain Host: osu.eu-west-1.outscale.com Authorization: AWS4-HMAC-SHA256 Credential=12345678901234567890/20160502/eu-west-1/s3/aws4_request, SignedHeaders=accept-encoding;content-length;content-type;host;user-agent;x-amz-content-sha256;x-amz-date;x-amz-storage-class, Signature=fb7503d5014ce581723d9695a1395bb668a3a5fcee713663de3d7a316d9b4b27 User-Agent: libcloud/1.0.0-rc2 (OUTSCALE Ceph RGW S3 (eu-west-1)) Hello test! HTTP/1.1 200 OK ETag: "8ac91dd17dd64e95271601b773859aaa" Content-Length: 0 Accept-Ranges: bytes x-amz-request-id: tx000000000000000000012-00572650f0-1017-default Date: Sun, 01 May 2016 18:54:40 GMT |
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
|
Great work. LGTM |
Kami
reviewed
May 2, 2016
libcloud/storage/drivers/s3.py
| def _get_s3_rgw_outscale_conn_by_sign_version(signature_version): | ||
| # aws4 connection support | ||
| class S3RGWOutscaleConnectionAWS4(SignedAWSConnection, BaseS3Connection): |
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
Kami
May 2, 2016
Member
I would prefer to make those classes top-level and define them outside of this function, but that's mostly a style thing.
Kami
May 2, 2016
Member
I would prefer to make those classes top-level and define them outside of this function, but that's mostly a style thing.
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
Kami
reviewed
May 2, 2016
libcloud/storage/drivers/s3.py
| @@ -1029,7 +1056,9 @@ def __init__(self, key, secret=None, secure=True, host=None, port=None, | ||
| self.name = 'OUTSCALE Ceph RGW S3 (%s)' % (region) | ||
| self.ex_location_name = region | ||
| self.region_name = region | ||
| self.connectionCls = S3RGWOutscaleConnection | ||
| signature_version = kwargs.get('signature_version', 'aws2') | ||
| self.connectionCls =\ |
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
Kami
May 2, 2016
Member
Since connectionCls is a class-level attribute (and not an instance one), I believe this won't work correctly if multiple instances of this class are instantiated and each one uses a different signature version.
For example:
cls1 = S3RGWOutscaleStorageDriver(..., signature_version='aws2')
cls2 = S3RGWOutscaleStorageDriver(..., signature_version='aws4')The version of the class which was instantiated last wins so connectionCls of cls1 would also be set to S3RGWOutscaleConnectionAWS4 after the second class is instantiated.
Another option, which I believe should work correctly, is to utilize ex_connection_class_kwargs method and pass correct value for the version argument to the connectionCls depending on the requested signature version.
Kami
May 2, 2016
Member
Since connectionCls is a class-level attribute (and not an instance one), I believe this won't work correctly if multiple instances of this class are instantiated and each one uses a different signature version.
For example:
cls1 = S3RGWOutscaleStorageDriver(..., signature_version='aws2')
cls2 = S3RGWOutscaleStorageDriver(..., signature_version='aws4')The version of the class which was instantiated last wins so connectionCls of cls1 would also be set to S3RGWOutscaleConnectionAWS4 after the second class is instantiated.
Another option, which I believe should work correctly, is to utilize ex_connection_class_kwargs method and pass correct value for the version argument to the connectionCls depending on the requested signature version.
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
Kami
May 2, 2016
Member
Here is an example of how we handle region connection class argument in the same manner - https://github.com/apache/libcloud/blob/trunk/libcloud/compute/drivers/rackspace.py#L142
Kami
May 2, 2016
Member
Here is an example of how we handle region connection class argument in the same manner - https://github.com/apache/libcloud/blob/trunk/libcloud/compute/drivers/rackspace.py#L142
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
jmunhoz
May 2, 2016
Contributor
@Kami I added the suggested improvements. After some tests I finished adding the dispatching code between the connection classes as part of the _ex_connection_class_kwargs method. It is working with the case you mentioned.
|
@Kami I added the suggested improvements. After some tests I finished adding the dispatching code between the connection classes as part of the _ex_connection_class_kwargs method. It is working with the case you mentioned. |
Kami
reviewed
May 10, 2016
| super(S3RGWOutscaleStorageDriver, self).__init__(key, secret, | ||
| secure, host, port, | ||
| api_version, region, | ||
| **kwargs) | ||
| def _ex_connection_class_kwargs(self): |
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
Kami
May 10, 2016
Member
With this approach, the same issue exists as before - you need to do something along the lines of:
def _ex_connection_class_kwargs(self):
kwargs = {}
kwargs['signature_version'] = self.signature_version
return kwargsAnd you also need to modify connection class to take signature_version argument (or similar) and use this argument to correctly set host attribute on the connection class inside the constructor.
I'm also fine with merging this approach for now, as long as we add some test cases for it (a test case where you instantiate two drivers with different signature versions and verify that the connectionCls and connectionCls.host is set correctly) so it's easy to refactor and fix in the near future.
Kami
May 10, 2016
•
edited
Edited 1 time
-
Kami
edited May 10, 2016 (most recent)
Member
edited
-
May 10, 2016 (most recent)
With this approach, the same issue exists as before - you need to do something along the lines of:
def _ex_connection_class_kwargs(self):
kwargs = {}
kwargs['signature_version'] = self.signature_version
return kwargsAnd you also need to modify connection class to take signature_version argument (or similar) and use this argument to correctly set host attribute on the connection class inside the constructor.
I'm also fine with merging this approach for now, as long as we add some test cases for it (a test case where you instantiate two drivers with different signature versions and verify that the connectionCls and connectionCls.host is set correctly) so it's easy to refactor and fix in the near future.
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
jmunhoz
May 12, 2016
Contributor
@Kami done. Added the modifications. I tested with the shared snippet and it looks okay now.
jmunhoz
May 12, 2016
Contributor
@Kami done. Added the modifications. I tested with the shared snippet and it looks okay now.
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
Kami
May 13, 2016
Member
@jmunhoz Thanks for the update and sorry to bug you again, but this still won't work correctly.
You are still manipulating connectionCls inside the driver constructor. You should get rid of this code and as mentioned above, update connection class constructor to take signature_version argument and correctly set host inside the connection class constructor.
This _ex_connection_class_kwargs method doesn't actually do anything at this point, it doesn't throw because connection class declares **kwargs. So in short, connection class constructor accepts this argument, but doesn't do anything with it.
In any case, I can go ahead, merge this into master as is and add a (failing) test case so this can be fixed in the future.
Kami
May 13, 2016
Member
@jmunhoz Thanks for the update and sorry to bug you again, but this still won't work correctly.
You are still manipulating connectionCls inside the driver constructor. You should get rid of this code and as mentioned above, update connection class constructor to take signature_version argument and correctly set host inside the connection class constructor.
This _ex_connection_class_kwargs method doesn't actually do anything at this point, it doesn't throw because connection class declares **kwargs. So in short, connection class constructor accepts this argument, but doesn't do anything with it.
In any case, I can go ahead, merge this into master as is and add a (failing) test case so this can be fixed in the future.
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
jmunhoz
May 13, 2016
Contributor
@Kami Thanks for the review. It is very appreciated.
Yes, please. Feel free to merge. We can continue improving the code over the upstream version. I guess my next patch would build on a similar bug so they could be fixed together.
I guess the example code I used at #736 (comment) is not catching the issue properly although it looks working ok? I am manipulating the connectionCls inside the driver constructor cause it looks the only place to choose between the two different connection classes without a further modification.
jmunhoz
May 13, 2016
Contributor
@Kami Thanks for the review. It is very appreciated.
Yes, please. Feel free to merge. We can continue improving the code over the upstream version. I guess my next patch would build on a similar bug so they could be fixed together.
I guess the example code I used at #736 (comment) is not catching the issue properly although it looks working ok? I am manipulating the connectionCls inside the driver constructor cause it looks the only place to choose between the two different connection classes without a further modification.
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
jmunhoz
May 11, 2016
Contributor
@Kami updated!
I tested the 'last connectionCls wins' issue with the following code:
$ cat ex.py
#!/usr/bin/env python
from libcloud.storage.types import Provider, ContainerDoesNotExistError
from libcloud.storage.providers import get_driver
import libcloud
container_name = 'container-example'
object_name = 'object-example'
cls = get_driver(Provider.S3_RGW_OUTSCALE)
driver1 = cls('id', 'secret', region='eu-west-1', secure=False, signature_version='4')
driver2 = cls('id', 'secret', region='us-east-2', secure=False, signature_version='2')
print "driver1.connectionCls = %s" % (driver1.connectionCls)
print "driver1.connectionCls.host = %s" %( driver1.connectionCls.host)
print "driver2.connectionCls = %s" % (driver2.connectionCls)
print "driver2.connectionCls.host = %s" %( driver2.connectionCls.host)
$ ./ex.py
driver1.connectionCls = <class 'libcloud.storage.drivers.s3.S3RGWOutscaleConnectionAWS4'>
driver1.connectionCls.host = osu.eu-west-1.outscale.com
driver2.connectionCls = <class 'libcloud.storage.drivers.s3.S3RGWOutscaleConnectionAWS2'>
driver2.connectionCls.host = osu.us-east-2.outscale.com
jmunhoz
commented
May 11, 2016
•
edited
Edited 1 time
-
jmunhoz
edited
May 11, 2016 (most recent)
edited
-
May 11, 2016 (most recent)
|
@Kami updated! I tested the 'last connectionCls wins' issue with the following code: $ cat ex.py
#!/usr/bin/env python
from libcloud.storage.types import Provider, ContainerDoesNotExistError
from libcloud.storage.providers import get_driver
import libcloud
container_name = 'container-example'
object_name = 'object-example'
cls = get_driver(Provider.S3_RGW_OUTSCALE)
driver1 = cls('id', 'secret', region='eu-west-1', secure=False, signature_version='4')
driver2 = cls('id', 'secret', region='us-east-2', secure=False, signature_version='2')
print "driver1.connectionCls = %s" % (driver1.connectionCls)
print "driver1.connectionCls.host = %s" %( driver1.connectionCls.host)
print "driver2.connectionCls = %s" % (driver2.connectionCls)
print "driver2.connectionCls.host = %s" %( driver2.connectionCls.host)
$ ./ex.py
driver1.connectionCls = <class 'libcloud.storage.drivers.s3.S3RGWOutscaleConnectionAWS4'>
driver1.connectionCls.host = osu.eu-west-1.outscale.com
driver2.connectionCls = <class 'libcloud.storage.drivers.s3.S3RGWOutscaleConnectionAWS2'>
driver2.connectionCls.host = osu.us-east-2.outscale.com
|
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
Kami
May 13, 2016
Member
Made a small change (09638ed) and merged it into trunk - thanks!
P.S. I forgot to close the PR in the commit message, can you please close this PR?
|
Made a small change (09638ed) and merged it into trunk - thanks! P.S. I forgot to close the PR in the commit message, can you please close this PR? |
This comment has been minimized.
Show comment
Hide comment
This comment has been minimized.
|
Thanks! closing this PR |
jmunhoz
closed this
May 13, 2016
added a commit
to DimensionDataCBUSydney/libcloud
that referenced
this pull request
May 25, 2016
jmunhoz commentedApr 4, 2016
This patch adds AWS4 auth protocol support in the S3_RGW_OUTSCALE
provider. It is needed to use AWS4 with the upcoming version of Ceph RGW
(Jewel)
Ceph Jewel will ship with AWS2 and AWS4 enabled by default.
In the case of regions and signature binding, Ceph does not enforce any
signature version per region. Every region supports AWS2 and AWS4. Ceph
detects the signature version per request in order to authenticate
properly.
More information on the Ceph's AWS4 implementation:
http://docs.ceph.com/docs/master/release-notes/#v10-1-0-jewel-release-candidate
http://blogs.igalia.com/jmunhoz/blog/2016/03/01/aws-signature-version-4-goes-upstream-in-ceph.html