On 3/3/20 3:53 AM, Jeff Layton wrote:
On Fri, 2020-02-28 at 19:55 +0800, Yan, Zheng wrote:
This series make cephfs client not request caps for open files that
idle for a long time. For the case that one active client and multiple
standby clients open the same file, this increase the possibility that
mds issues exclusive caps to the active client.
Yan, Zheng (4):
ceph: always renew caps if mds_wanted is insufficient
ceph: consider inode's last read/write when calculating wanted caps
ceph: simplify calling of ceph_get_fmode()
ceph: remove delay check logic from ceph_check_caps()
fs/ceph/caps.c | 324 +++++++++++++++--------------------
fs/ceph/file.c | 39 ++---
fs/ceph/inode.c | 19 +-
fs/ceph/ioctl.c | 2 +
fs/ceph/mds_client.c | 5 -
fs/ceph/super.h | 35 ++--
include/linux/ceph/ceph_fs.h | 1 +
7 files changed, 188 insertions(+), 237 deletions(-)
changes since v2
- make __ceph_caps_file_wanted more readable
- add patch 5 and 6, which fix hung write during testing patch 1~4
This patch series causes some serious slowdown in the async dirops
patches that I've not yet fully tracked down, and I suspect that they
may also be the culprit in these bugs:
slow down which tests?
https://tracker.ceph.com/issues/44381
this is because I forgot to check if inode is snap when queue delayed
check. But it can't explain slow down.
https://tracker.ceph.com/issues/44382
I'm going to drop this series from the testing branch for now, until we
can track down the issue.
Thanks,
