FileStore: use pwritev instead of lseek+writev #7349
Conversation
|
update on #6718 @jack-changtao plz review |
|
@yuyuyu101 could you help @jack-changtao to squash his two commits into a single one? |
| bytes = 0; | ||
|
|
||
|
|
||
| int r = do_writev(fd, iov, offest, iovlen, bytes); |
Signed-off-by: Tao Chang <changtao@hihuron.com>
yuyuyu101
force-pushed
the
add-pwritev
branch
from
0a32b63
to
1cb030d
Compare
| @@ -123,6 +123,9 @@ | |||
| /* Define to 1 if you have the `syncfs' function. */ | |||
| #cmakedefine HAVE_SYNCFS 1 | |||
|
|
|||
| /* Define to 1 if you have the `pwritev' function. */ | |||
| #cmakedefine HAVE_PWRITEV 1 | |||
There was a problem hiding this comment.
do you want to add
CHECK_FUNCTION_EXISTS(pwritev HAVE_PWRITEV)
in CMakeLists.txt also?
yuyuyu101
force-pushed
the
add-pwritev
branch
from
1cb030d
to
be6dbcd
Compare
|
@yuyuyu101 could you reconsider #7349 (comment) ? w.r.t |
yuyuyu101
force-pushed
the
add-pwritev
branch
from
be6dbcd
to
39c05ab
Compare
|
@tchaikov thanks! I missed it |
|
The write_fd changes break the crush map check on pool creation ('ceph osd pool create foo 12' fails). Probably because we are doing write_fd to a pipe to the child process? |
|
@liewegas updated, revert orignal write_fd |
|
@yuyuyu101 do you mind squashing this down so that write_fd isn't changed and then changed back? |
yuyuyu101
force-pushed
the
add-pwritev
branch
from
92c81a8
to
0891703
Compare
|
@liewegas done |
|
unittest_bufferlist (and some other make check tests) fail |
Signed-off-by: Haomai Wang <haomai@xsky.com>
Signed-off-by: Haomai Wang <haomai@xsky.com>
yuyuyu101
force-pushed
the
add-pwritev
branch
from
0891703
to
aa7b505
Compare
|
@liewegas eh, pass now... |
Signed-off-by: Haomai Wang <haomai@xsky.com>
|
retest this please |
osd: FileStore: use pwritev instead of lseek+writev Reviewed-by: Kefu Chai <kchai@redhat.com> Reviewed-by: Sage Weil <sage@redhat.com>
No description provided.