It is only possible to remove image from rbd_mirroring object if the
mirroring image state is set to DISABLING.
Signed-off-by: Ricardo Dias <rdias@suse.com>
At least one S3 client is overly strict about requiring the xmlns to be present
in server responses. Let's be strict about what we send, and continue to be lax
about what we receive.
- Put correct XMLNS in a define instead of repeating it.
- Correct XMLNS hostname: s/doc.s3.amazonaws.com/s3.amazonaws.com/
- Add XMLNS where it was missing:
-- CORSConfiguration
-- CopyObjectResult (see S3QL#190)
-- ListMultipartUploadsResult
S3QL bug #190: https://bitbucket.org/nikratio/s3ql/issues/190/sq3l-should-not-rely-on-xmlns-declarations
Signed-off-by: Robin H. Johnson <robin.johnson@dreamhost.com>
* pass the environment variables using `env` to pass environment
varibles with space(s) in them to the COMMAND in add_custom_target.
otherwise, cmake will try to quote the space with "\". this breaks the
generated command line.
* add a comment for ccache to note that we do not expect ccache to speed
up linking. we use it as the linker's launcher to workaround
https://bugs.python.org/issue8027. to be specific,
distutils.UnixCCompiler.link overwrites the first linker CLI's arg
using the the first C++ compiler's CLI arg, if "env" is not used to
launch the linker. this breaks the cythonization of our pybind APIs.
Signed-off-by: Kefu Chai <kchai@redhat.com>
'source' is not sh command and executing bash commands in sh won't
work. We could try to replace it with:
'. ../qa/workunits/ceph-helpers.sh'
but the above file uses bash.
Signed-off-by: Michal Jarzabek <stiopa@gmail.com>
Disabling exclusive lock will now acquire the exclusive lock before
proceeding. If this resulted in another client releasing the lock
and re-requesting it, support for exclusive locking might be disabled
while a request is pending.
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
If object map or journaling is dynamically disabled, we want to
ensure that no in-flight IO could be updating the objects while
we attempt to delete them.
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
Need to clear any in-flight IO before the journal is closed to
avoid IO completions getting stuck in odd state.
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
If the exclusive lock is enabled/disabled during an IO-driven
image refresh, we need to avoid deadlocking the queue.
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
if the first journal object got removed successfully while later objects
failed then the final result should be set to fail
Signed-off-by: runsisi <runsisi@zte.com.cn>
All current callers (except list_objects) already expect
ret < 0 for error. Adjust list_objects to skip on EINVAL
instead of 0.
Signed-off-by: Samuel Just <sjust@redhat.com>
We leave skip_chain_cleanup = false since the object may have been
written using code that split the lfn attr over multiple attrs.
Signed-off-by: Samuel Just <sjust@redhat.com>
onechunk was a deceptive name since it didn't actually ensure that
a single chunk would be used. Rename to ensure_single_attr. Also,
add a parameter to ensure that we use a single attribute. We need
these to be distinct since we have LFN attrs which have been split
over 254 byte xattrs since after hammer which we need to correctly
clean up. However, we need to ensure going forward that those attrs
are never split over more than one chunk -- it's not atomic.
Further, skip_chain_cleanup should imply ensure_single_attr, so make
them template params and add a static check to make sure we don't
mess this up in the future.
Signed-off-by: Samuel Just <sjust@redhat.com>
Enforce locator length vs the max name length and also
introduce a namespace length limit.
In addition to these checks, also pass the head object to the
ObjectStore implementation to validate. This allows LFNIndex to account
for the idiosyncracies of its filename escaping and for different xattr
value max sizes.
Signed-off-by: Samuel Just <sjust@redhat.com>
