On jewel it will have stolen the exclusive lock. Instead, ensure that
object map and fast diff are already disabled on the clone before the
start of the test.
Fixes: http://tracker.ceph.com/issues/19716
Signed-off-by: Mykola Golub <mgolub@mirantis.com>
This is usually rocksdb on bluefs, but it should work with whatever kv
backend bluestore is configured to use.
Signed-off-by: Sage Weil <sage@redhat.com>
This change happened a while back, but it got rolled back
when the generic objectstore/ dir had its filestore
entry split out into xfs and btrfs in 208675af.
Signed-off-by: John Spray <john.spray@redhat.com>
The "recovery" sub suite was originally tests for
client/mds recovery in certain failure cases, it has
since grown to include lots of unit testing of
various features using CephFSTestCase.
The "basic" suite is now specifically just running workloads
now that I've moved out the smaller functional tests.
Signed-off-by: John Spray <john.spray@redhat.com>
Most of what's in basic/ is "workload" type testing
(i.e. a simple cluster cluster configuration and then
running a script inside the client), which gets
permuted in various ways. Move the simpler
functional tests out with the others like themselves.
Signed-off-by: John Spray <john.spray@redhat.com>
These are unit tests for specific CephFS functionality,
it is gratuitous to repeat them with different underlying
RADOS object stores.
We retain coverage of XFS vs. bluestore in the workload tests.
Signed-off-by: John Spray <john.spray@redhat.com>
The following warning appears during make. Fixed the review comments too.
[ 28%] Building CXX object src/mds/CMakeFiles/mds.dir/__/osdc/Journaler.cc.o
./src/osdc/Journaler.cc: In member function ‘void Journaler::_do_flush(unsigned int)’:
./src/osdc/Journaler.cc:643:16: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
if (newlen < len) {
~~~~~~~^~~~~
Signed-off-by: Jos Collin <jcollin@redhat.com>
create bucket in secondary zonegroup will forward to master. The master may have different num_shards option.
So when create bucket in local, should use master's num_shards instead of local num_shards option.
Fixes: http://tracker.ceph.com/issues/19745
Signed-off-by: Shasha Lu <lu.shasha@eisoo.com>
after running 'radosgw-admin data sync init', the sync_marker's state will change to FullSync, but the marker didn't clean. So 'sync status' command will wrongly inset shards_behind.
Signed-off-by: Shasha Lu <lu.shasha@eisoo.com>
Due to handle_get_locker is executed in async context, shut_down could
start before handle_get_locker.
Fixes: http://tracker.ceph.com/issues/19405
Signed-off-by: Mykola Golub <mgolub@mirantis.com>
commit d788bae60d renamed config.rst to config-fcgi.rst
Reported-by: fusl in #ceph on irc.oftc.net
Signed-off-by: Peter Maloney <peter.maloney@brockmann-consult.de>
there is chance that the acting set of a PG being created becomes empty,
and then OSDs joins the acting set. in that case, we need also update
the "create" epoch of the creating_pgs, so the new primary can be
updated with the MOSDPCreate message.
Fixes: http://tracker.ceph.com/issues/19744
Signed-off-by: Kefu Chai <kchai@redhat.com>
