erodriguez@lazyk-ranch.com has uploaded this change for review.
FSAL/CEPH: record READ/WRITE stats on async I/O path
With enable_FSALSTATS, sync ceph_ll_read/write are timed via
CEPH_FSAL_TIME_CALL, but the async path (ceph_ll_nonblocking_readv_writev)
never recorded CEPH_STAT_READ/WRITE. Under CephFSM.async, ganesha_stats
therefore showed NFS READ/WRITE in the millions with empty FSAL counters,
so async FSAL latency could not be measured.
Time submit→completion for nonblocking read/write: start on submit,
finish in the callback (or immediately when result != 0). Skip recording
on CLOSEFD resume to avoid double-counting. Reuse existing READ/WRITE
opcodes so DBus consumers see the same op names as sync.
Change-Id: I322f354383c8d18e1d3290673d58c3a6eb57458e
Signed-off-by: Edwin Rodriguez <edwin.rodriguez1@ibm.com>
---
M src/FSAL/FSAL_CEPH/ceph_stats.h
M src/FSAL/FSAL_CEPH/handle.c
2 files changed, 57 insertions(+), 2 deletions(-)
git pull ssh://review.gerrithub.io:29418/ffilz/nfs-ganesha refs/changes/49/1259549/1
To view, visit change 1259549. To unsubscribe, or for help writing mail filters, visit settings.