View Issue Details

IDProjectCategoryView StatusLast Update
0017622CentOS-8xfsdumppublic2020-07-27 11:22
Reporterolaf 
PrioritynormalSeveritymajorReproducibilityalways
Status newResolutionopen 
Platformx86_64OSCentos 8OS Version8.1.1911
Product Version8.1.1911 
Target VersionFixed in Version 
Summary0017622: xfsdump is unable to complete when high disk I/O
DescriptionWhen there is a high input/output on the disks, the xfsdump is unable to finish.
This is because it calls "sync" and it takes ages to complete.
This is when I try to dump a filesystem on one device and the high I/O is on another.
The stack looks as below:
[root@vh3 ~]# cat /proc/3661/stack
[<0>] wait_on_page_bit+0x11a/0x1d0
[<0>] __filemap_fdatawait_range+0x89/0xf0
[<0>] filemap_fdatawait_keep_errors+0x1a/0x40
[<0>] iterate_bdevs+0xad/0x142
[<0>] ksys_sync+0x7c/0xb0
[<0>] __do_sys_sync+0xa/0x10
[<0>] do_syscall_64+0x5b/0x1e0
[<0>] entry_SYSCALL_64_after_hwframe+0x44/0xa9

This is a huge problem, as it makes backups impossible.
Steps To ReproduceThe setup I have is:
two mirrors, total 4 drives:
first mdadm mirror, on this LUKS, then LVM and xfs filesystems on logical volumes, 600 GB total, on SAS controller
second mdadm mirror, on this LUKS, the deice size is 10 TB on SATA controller
Start a lot of writes to the second mirror "cat /dev/zero | pv > /dev/mdX"
After some time try to xfsdump an xfs filesystem on the first mirror
The xfsdump is stuck as it calls "sync" syscall.
As the 10TB write takes about 15 hours - the sync doesn't complete unless all writes are finished.
TagsNo tags attached.

Activities

There are no notes attached to this issue.

Issue History

Date Modified Username Field Change
2020-07-27 11:22 olaf New Issue