os/bluestore: print aio in batch #18873
Conversation
|
This PR looks more clean than #18803 . |
tchaikov
force-pushed
the
wip-bluestore-aio-log
branch
from
02049a5
to
f2d526e
Compare
src/os/bluestore/aio.cc
Outdated
| unsigned i = 0; | ||
| os << "aio: "; | ||
| for (auto& iov : aio.iov) { | ||
| os << "\n [" << i++ << "] " << iov.iov_base << " " << iov.iov_len; |
There was a problem hiding this comment.
Do you think if it makes sense to print iov_base/iov_len in hex?
There was a problem hiding this comment.
iov_base is a pointer so it's printed in hex already. but iov_len is a length, i am not sure if we should print it in hex. why shall we do so? because hex digits can be printed aligned and look better? but so do the dec digits..
There was a problem hiding this comment.
everything else in bluestore is in hex (offsets and lengths). it's way easier to see block alignment etc. we should do hex here imo
liewegas
changed the title
tchaikov
force-pushed
the
wip-bluestore-aio-log
branch
from
f2d526e
to
a06360b
Compare
src/os/bluestore/aio.cc
Outdated
| os << "aio: "; | ||
| for (auto& iov : aio.iov) { | ||
| os << "\n [" << i++ << "] " << std::hex | ||
| << iov.iov_base << " " << iov.iov_len; |
There was a problem hiding this comment.
please reset to decimal on completion to avoid potential side effects..
There was a problem hiding this comment.
@ifed01 fixed and repushed
KernelDevice::aio_{submit,write,read}() are critical paths. calling
cct->_conf->subsys.should_gather() multi-times is not optimal. the
downside of this issue is that if the aio is printed, the size of
buffer in PrebufferedStreambuf could be large if the number of iov is
large, that could be a heavy load to the memory subsystem.
Signed-off-by: Kefu Chai <kchai@redhat.com>
tchaikov
force-pushed
the
wip-bluestore-aio-log
branch
from
a06360b
to
67c6f76
Compare
KernelDevice::aio_{submit,write,read}() are critical paths. calling
cct->_conf->subsys.should_gather() multi-times is not optimal. the
downside of this issue is that if the aio is printed, the size of
buffer in PrebufferedStreambuf could be large if the number of iov is
large, that could be a heavy load to the memory subsystem.
Signed-off-by: Kefu Chai kchai@redhat.com