client: properly trim unlinked inode #7297
Conversation
Client should trim inode from its cache when receiving a cap message with nlink == 0. But the corresponding code has a bug, it does nothing when inode's nlink is already 0. Fixes: ceph#13903 Signed-off-by: Yan, Zheng <zyan@redhat.com>
|
test locally with jscp's test case ceph/ceph-qa-suite#787 |
|
@ukernel just out of curiosity, how did you test the teuthology task locally? i assume you don't have a teuthology testbed "locally"? as i am debugging my test suite at ceph/ceph-qa-suite#795, which is also a teuthology task. and found that it's time consuming to test it in sepia, and waste the precious sepia lab resource when doing so. |
|
@tchaikov it's the test runner for our cephfs tests (things in ceph-qa-suite/tasks/cephfs) that works with a vstart cluster (http://comments.gmane.org/gmane.comp.file-systems.ceph.devel/26151) -- I had exactly the same issue of developing tests in sepia being too slow, so wrote this. Unfortunately it doesn't apply to most of the non-cephfs tests, because they're not written against the python interface (CephFSTestCase). |
|
What was this nlink test trying to do and why is it okay to just eliminate it? |
|
@jcsp thanks. seems there is long way to go to follow the cephfs CephFSTestCase's model. the rados tests are quite different. |
|
I think purpose of the test is saving a few cpu cycles. On Thu, Jan 21, 2016 at 1:19 PM, Gregory Farnum notifications@github.com
|
…-fs-testing #7297 Reviewed-by: Greg Farnum <gfarnum@redhat.com>
client: properly trim unlinked inode Reviewed-by: Greg Farnum <gfarnum@redhat.com>
Client should trim inode from its cache when receiving a cap message
with nlink == 0. But the corresponding code has a bug, it does nothing
when inode's nlink is already 0.
Fixes: #13903
Signed-off-by: Yan, Zheng zyan@redhat.com