ceph: periodically send perf metrics to MDSes
authorXiubo Li <xiubli@redhat.com>
Thu, 16 Jul 2020 14:05:57 +0000 (10:05 -0400)
committerIlya Dryomov <idryomov@gmail.com>
Mon, 3 Aug 2020 09:05:26 +0000 (11:05 +0200)
commit18f473b384a64cef69f166a3e2b73d3d2eca82c6
tree3d9626cc829877510c1ac5a557f88c7b5f855bbf
parentaaf5a476201bf93bdab75d6922340516ee63f7e2
ceph: periodically send perf metrics to MDSes

This will send the caps/read/write/metadata metrics to any available MDS
once per second, which will be the same as the userland client.  It will
skip the MDS sessions which don't support the metric collection, as the
MDSs will close socket connections when they get an unknown type
message.

We can disable the metric sending via the disable_send_metrics module
parameter.

[ jlayton: fix up endianness bug in ceph_mdsc_send_metrics() ]

URL: https://tracker.ceph.com/issues/43215
Signed-off-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
fs/ceph/mds_client.c
fs/ceph/mds_client.h
fs/ceph/metric.c
fs/ceph/metric.h
fs/ceph/super.c
fs/ceph/super.h
include/linux/ceph/ceph_fs.h