librbd: sync flush should re-use existing async flush logic #18403
Conversation
dillaman
changed the title
dillaman
force-pushed
the
wip-librbd-flush
branch
from
f52ef52
to
7f8b7a7
Compare
| } | ||
| r = ctx.wait(); | ||
|
|
||
| ictx->perfcounter->inc(l_librbd_flush); |
There was a problem hiding this comment.
@dillaman Are you going to leave "flush" counter unused?
There was a problem hiding this comment.
BTW, in "qa/workunits/rbd/test_admin_socket.sh" we have a test that checks "flush" perfcounter is updated after "flush" admin socket command.
There was a problem hiding this comment.
Yeah -- I thought it was already covered here [1] but it's for AIO flush. I'll combine them in the next push.
[1]
ceph/src/librbd/io/ImageRequest.cc
Line 753 in 9b18093
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
dillaman
force-pushed
the
wip-librbd-flush
branch
2 times, most recently
from
f2d1ac2
to
8ad2c3b
Compare
src/tools/rbd/action/Bench.cc
Outdated
| int r = image.flush(); | ||
| if (r < 0) { | ||
| std::cerr << "rbd: failed to flush: " << cpp_strerror(r) << std::endl; | ||
| return r; |
There was a problem hiding this comment.
Do we want this to be fatal? E.g. previously running bench read against exclusively mapped image would succeed.
There was a problem hiding this comment.
Tweaked it to be fatal for write/read-write tests and skipped on the final flush for read-only tests
dillaman
force-pushed
the
wip-librbd-flush
branch
from
8ad2c3b
to
e215a82
Compare
|
@dillaman |
For the new read-based bench tests, flushing prior to the start of the test will result in the exclusive lock being acquired and the object map being utilized. Signed-off-by: Jason Dillaman <dillaman@redhat.com>
dillaman
force-pushed
the
wip-librbd-flush
branch
from
e215a82
to
918df91
Compare
|
@trociny updated |
No description provided.