https://bugzilla.kernel.org/show_bug.cgi?id=203247 Theodore Tso (tytso@xxxxxxx) changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |RESOLVED Resolution|--- |INVALID --- Comment #11 from Theodore Tso (tytso@xxxxxxx) --- Looking at your iostat -x output, /dev/sda is completely utilized: Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s sda 0.00 1154.00 47.00 836.00 864.00 10076.00 avgrq-sz avgqu-sz await r_await w_await svctm %util 24.78 136.46 116.05 331.72 103.93 1.13 100.00 Note the 100% utilization and the average queue size of 136.46 I/O requests. So as I said, this sounds like I/O overload. Your workload is simply issuing more I/O than the disk can handle. Why is going to be a userspace issue. If you don't care about stale data (e.g., belonging to deleted files, which may have privacy issues if you care about GDPR, HIPPA, PCI, etc. mandates) you could try mounting with the mount option "data=writeback". It could be that updating to a newer version of the kernel (4.9.86 is quite antique, and you're probably missing all sorts of security and otehr bug fixes) and a newer userspace, and then setting up a new set of file systems formatted with ext4 will also help performance. Or there might be something going on with userspace writes to /dev/sda that shouldn't be happening or which could be optimized away. But I don't see anything here which is a kernel bug. -- You are receiving this mail because: You are watching the assignee of the bug.
