Closed steven-varga closed 2 years ago
Hit similar problem. Seems to be related to changes in the kernel where page size of 4kb is now used for file system buffersize e.g frbuf. This ignores the 4mb buffer used for actual data transfer. Seems to impact the statfs. Maybe logic needed in the kernel module to handle different size depending on type of object. E.g torvalds/linux@9f8fd53cd05596f6792f769c9fd5fd2b0d624507
Have used the fuse method instead to get around the problem.
Thanks for letting us know about this. We have tracked down the problem to an uninitialized variable in the kernel module. Luckily it was an easy fix. I don't think this change got into the release pull that just happened, but we will get it in ASAP and backport to earlier kernels. Again, thanks for letting us know.
This issue has been resolved and backported to the relevant kernel versions as of July 2021 (forgot to close the issue earlier, apologies for that).
I am to report a misbehaviour on ubuntu 18.04 system provided df disk free utility, that it reports the wrong size of mounted pvfs.
But from shell
df -h
where/homedir
is layered on top of a3.1TB
drive prints out3.1GB
as if the block size was wrong.I tried this with
v2.9.7
v2.9.8
and the master combined with two kernel version:v5.6.12
andv5.8.14
built in orangefs module. I copied massive data on theorangefs
filesystem and md5 checksum them each time. Out of10
checksum tests, no test failed, an unlikely event if something was wrong with the data stored on the server.interestingly
pydf
utility reports the correct information:Has anyone noticed similar behaviour?