* performance degradation due to extra fuse_update_attributes() call in fuse_file_aio_write()
@ 2019-11-05 9:25 Vasily Averin
0 siblings, 0 replies; only message in thread
From: Vasily Averin @ 2019-11-05 9:25 UTC (permalink / raw)
To: Miklos Szeredi, linux-fsdevel; +Cc: Alexey Kuznetsov
Dear Miklos,
could you please elaborate the reason of fuse_update_attributes() call in fuse_file_aio_write() ?
After rebase of openVz7 kernel to RHEL7.7 we have found significant performance degradation:
from 136 kiops to 75 kiops during first 10 second and down to 25 kiops later.
The reason of this was an extra fuse_update_attributes() call in fuse_file_aio_write().
Our old code was an equivalent of v6 patch version submitted by Maxim Patlasov
https://lkml.org/lkml/2013/10/10/281
fuse_file_aio_write():
if (get_fuse_conn(inode)->writeback_cache)
return generic_file_aio_write(iocb, iov, nr_segs, pos);
However in final commit 4d99ff8f12eb "fuse: Turn writeback cache on" you have included an additional fuse_update_attributes() call
fuse_file_aio_write():
if (get_fuse_conn(inode)->writeback_cache) {
/* Update size (EOF optimization) and mode (SUID clearing) */
err = fuse_update_attributes(mapping->host, NULL, file, NULL);
if (err)
return err;
return generic_file_aio_write(iocb, iov, nr_segs, pos);
}
Unfortunately we did not found any description related to this change.
Could you please elaborate which kind of the problem you have noticed in v6 patch version
and explain how fuse_update_attributes() call fixes it?
Thank you,
Vasily Averin
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2019-11-05 9:25 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2019-11-05 9:25 performance degradation due to extra fuse_update_attributes() call in fuse_file_aio_write() Vasily Averin
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).