xrange is not supported in python 3, so replace it with range since no
large size lists involved.
Closes-Bug: #1237717
Change-Id: I4e5e0782153d32c8beee03f7d871722ed4352859
Added a retry_on_ratelimit parameter to the Connection
class so that ratelimited requests can be retried.
DocImpact
Change-Id: I2817a7ea0ed2d69a7659e80111fbd2c91a75d530
As far as proxies usage is concerned, keystone-client API and swift-client API behave differently because the former uses python Request library while the latter uses raw httplib. As a result, Keystone authentication honors environment variables http_proxy, https_proxy and no_proxy while Swift doesn't.
This patch, which code is mainly borrowed from Python Requests, makes Swift
data connections and Swift authentication connections behaving homogeneously.
Change-Id: Ic8a0089c35c458d7ed96e572e22429014298fe4c
Before commit 7d88d14d (http://git.io/hhJWdQ) swifclient used to accept
header value with the type None :
{'Headers': None}
It would just be happy with it with those None headers and not process
them, reinstate the old behavior.
Closes-Bug: 1256441
Change-Id: Ic7f80c38ef2ce9ef1687ed0d6d65521f2e754905
We noticed a difference in download bandwidth while using
authentication details and pre-auth information for download.
Download using authentication details included authentication
time into the download bandwidth at each call of the function.
This time should have been excluded from download bandwidth.
This patch set is adding a timer variable to the client library.
That allows to accurately measure and store the time of each action.
Then while printing the download bandwith in the swift command
authentication time is excluded from bandwidth.
Change-Id: I63df9023e169f637f120ae9e25dac9c90a4e75a0
Fixes: bug #1197443
* log to debug for successes and info for errors
* remove dead code, neither body or raw_body are sent in kwargs
* include resp.reason and response headers in logging
* remove trailing \n in logging. users who want them can set logging.basicConfig(format=FORMAT)
* add --info option to swift cli
Change-Id: If07af46cb377f3f3d70f6c4284037241d360a8b7
Appends the container's name to the URL displayed, to be used to
replay the queries with curl.
Replaying the curl commands displayed by "swift --debug list
<container>" now works properly.
Example:
$ swift --debug list test-container
...
DEBUG:swiftclient:REQ: curl -i http://10.0.48.134:8080/v1/AUTH_6c554b8dd3d74e44878eddb92caf8687/test-container?format=json -X GET -H ...
...
Change-Id: I7ec33d185fedc44a529c016d38f841fde39d20d0
Closes-Bug: #1238612
This lets you build python-swiftclient packages that don't require pbr
to be installed at all. You would need pbr on the machine running
rpmbuild / debuild, but not on the machines that install the packages.
Unfortunately, this does not make python-swiftclient able to be
installed via pip 0.3.1 on Lucid; you'll need to uninstall the system
python-pip package and install a new pip some other way. Given that
pip < 1.3 doesn't perform SSL certificate validation for pypi (trivial
MITM attack, anyone?), you'd probably want to get a new pip anyway.
Change-Id: I85d4d77aacf094e48d39e48e750594b95dbc7af0
When you stat a container or object with the verbose flag the full path of the
reousrce will be displayed with the token similarlly to how an account stat
displays the auth url and token.
* move some logic out of bin/swift.st_stat to test it
* new module swiftclient.commnad_helpers for code you want to test
* moved prt_bytes into swiftclient.utils to test it
* fixed IndexError with prt_bytes on sizes >= 1024Y
Change-Id: Iaaa96e0308b08c554205b0055b8a04de581fefa4
I was gunna write a test, but mocking ImportErrors is a PITA, so I decided to
just get my tox py26 working instead. Tests seem to fail/pass as expected
without/with the inline NullHanlder.
Closes-Bug: 1212861
Change-Id: I20207374a77ee9969ab201c3a57e4caf40c02a61
This patch extracts the multi-threading code from bin/swift into
swiftclient/multithreading and adds tests. In particular, this new way
of doing it (with context managers) will prevent non-daemonic threads
from wedging the process when unexpected exceptions happen.
I enabled reporting of which lines, specifically, are not covered by
unit tests (added -m option to "coverage report" in .unittests).
This patch includes a drive-by fix for uploading a segmented file with
--use-slo when that object already exists. A key of "name" was used
instead of "path", raising KeyError.
There's also another drive-by fix for uploading segmented objects with
--use-slo. Commit 874e0e4427b80e1b15b74a1557b73ba9d61443ca regressed
this by removing the capturing of thread-worker results in
QueueFunctionThread.run(). This patch restores that functionality and
the feature (uploading SLO objects).
Change-Id: I0b4f677e4a734e83d1a25088d9a74f7d46384e53
Added log statements in swiftclient/client.py:_retry where exceptions
are being raised.
Fixes bug# 1202229
Change-Id: I727537177849d08bb9603aa884152bdebc62fb85
There were two undocumented parameters in there, one introduced
recently in 6411fd3, and the other some time before.
Change-Id: I3a7d59e1c272b382b035b88ba5495bc1c57a3154
See bug #1188896.
Comparing with Curl and Django, they both refuse carriage returns in
header values, so the request() method on the HTTP(S)Connection instance
returned by swiftclient.client.http_connection() will raise an
InvalidHeadersException if any of the headers to be sent contain a
newline.
Drive-by fix for a couple of header values which were integers instead
of strings (Content-Length getting set to zero).
Fixes bug #1188896
Change-Id: Ic6afdb92882284f843aacb06d20f682ddcb47151
The max-backoff concept prevents the backoff used in _retry to double infinitely (which in turn causes wait times of an hour or more between retries).
As per review, changed code to use min instead.
Fixes: bug #1183542
Change-Id: Ic084f54069b7fa7a33604741487045c5e697ff06
When --os-storage-url is specified on the command-line to bin/swift, it
will override the used storage URL regardless of authentication for both
authentication version 1 and version 2. This can be used to bypass a
load-balancer to hit a specific proxy server for testing/debugging
purposes.
Within the client library, this feature is accessed by passing the
desired storage URL into swiftclient.client.Conection.__init__() via the
os_options keyword argument. For example:
conn = Connection(auth_url, user, key, os_options={
'object_storage_url': 'http://overridden.storage.url/AUTH_foo'})
This patch also adds a dependency on mock>=0.8.0, which is the same as
openstack/swift.
Change-Id: Id2a36ed6abffd65e7762b6beea5bbfc6c036e848
The re-authentication of expired tokens assumed that the 401 would
always occur on the first HTTP at tempt. This fix changes that so that
re-auth is still only attempted once, but no longer has to be on the
first attempt.
Change-Id: I93a1187da3637287a803a59c146256d4f543c9d5
Fixes: bug #1131142
Allows optionally disabling SSL compression. This can significantly
improve HTTPS upload/download performance in some cases -- in particular
when the object is not compressible and you have very high network
bandwidth.
Implements blueprint ssl-compression.
Change-Id: I1260055f9c2e83cdabfeb51aed11b3899bed4d55
The bug is simple: whenever swift uploads to a Swift with SSL,
it uses 100% CPU. It happens because we use HTTPSConnection from
eventlet that loops like that, while holding the interpreter lock.
Now, it could be fixed in eventlet, but let's try something more
natural: drop the eventlet's HTTP client. We do not use green
threads in the client anymore, so it's not like we need it for that.
Note that in most cases clients do not use the BufferedHTTPConnection
either, because it's only installed on Swift server nodes, not on
workstations. Get rid of that too.
bug: 959221
Change-Id: I1eb932779d4171598b3efaa043f817b9c6c995c4
Some mechancical replacement of the deprecated except x,y:
construct with except x as y, which works with any Python
version >= 2.6
Change-Id: Ic245049dc7b408a5c89b9e27dfd2bd7c08acc5b5
If you use the API and supply only preauth values (preauthurl and
preauthtoken), and if either of these are incorrect, a non-descriptive
AttributeError will be thrown and uncaught:
'NoneType' object has no attribute 'find'
The _retry() will fail on the first pass (getting 401), then try
to reauthenticate with non-preauth values. If those are not given,
particularly the auth url, then the urlparse() call will be supplied
None for the url in http_connection() causing the exception above.
This change makes sure that we have an auth url, user and key before
retrying authentication. Given the situation above, a '401 Unauthorized'
ClientException will now be thrown instead of the AttributeError.
Change-Id: Ie1b5bde1e8ff321aa18c0f23dbd2960d6e482236
Also fixed bug with current large objects with segment listing prefixes.
Allow retry for object PUTs when possible.
Change-Id: I0edff127fd5d5c53da33aa7cb76a4f4dc85bf6e6
Changed documentation for chunk_size parameter to indicate
that it can be used only with file like objects.
Also added a UserWarning when using a string contents to inform
that chunk_size will be ignored.
Added a unit test to check whether the warning is working
correctly.
Fixes: bug #1147232
Change-Id: I618ec45520ba81905ce2ead4d61f192d21ae3489
