directory fragmentation generates extra osd ops, which affects checks
in the test.
Fixes: http://tracker.ceph.com/issues/19892
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
This was sending lots of metadata ops to MDSs to persuade
them to migrate some subtrees, but that was flaky. Use
the shiny new rank pinning functionality instead.
Signed-off-by: John Spray <john.spray@redhat.com>
Don't assume that test_data_scan will be run on exactly 2 MDS nodes.
Fixes: http://tracker.ceph.com/issues/19893
Signed-off-by: Douglas Fuller <dfuller@redhat.com>
When selinux is enabled, kernel client may releases inodes (without
uptodate xattr) in readdir reply immediately after processing the reply.
The reason is that linking the inode to dentry causes deadlock if xattr
is not uptodate.
We can use stat(2) syscall to guarantee that kernel client caches an
inode.
Fixes: http://tracker.ceph.com/issues/19912
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Also catches corner-case found by Zheng where an unjournaled directory will
cause export pinning to fail because it cannot be made a subtree until its
parent is stable.
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
Idea here is that a pinned inode should not be exported when its parent is.
Setting the pinned inode's dirfrags to aux subtrees prevents them from being
merged with a parent subtree.
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
This allows the client/admin to pin a directory tree to a particular rank,
preventing its export by the dynamic balancer.
Fixes: http://tracker.ceph.com/issues/17834
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
Previously, errors stuck indelibly to the inode, which
meant that a close call would see an error even if the
user already dutifully fsync()'d and handled it.
We should emit each error only once per file handle.
Signed-off-by: John Spray <john.spray@redhat.com>
There were some cases where we would leave a mountpoint
that would cause the teuthology teardown to get hung up
when it tried to look inside cephtest/
Signed-off-by: John Spray <john.spray@redhat.com>
The umount process can get stuck, in which case
we want to fail the test rather than waiting around for it.
During teardown of the kclient task catch this
timeout explicitly so that we will powercycle the node if
needed.
Signed-off-by: John Spray <john.spray@redhat.com>
Do the write after opening the file, so that we get good
behaviour wrt the change in Mount.open_background that uses
file existence to confirm that the open happened.
Signed-off-by: John Spray <john.spray@redhat.com>
Previously a later remote call could end up executing
before the remote python program in open_background
had actually got as far as opening the file.
Fixes: http://tracker.ceph.com/issues/18661
Signed-off-by: John Spray <john.spray@redhat.com>
