Red Hat Bugzilla – Bug 1251639
[PowerKVM]The md5sum value changed when file transfered after enable/disable multi queues repeatedly
Last modified: 2015-08-18 14:10:19 EDT
Description of problem: In the guest,enable and disable multi queues in a loop,ping guest from another external host during the loop,then transfer 6 files from guest to host,some md5sum value changed. Version-Release number of selected component (if applicable): qemu-kvm-rhev-2.3.0-13.el7.ppc64le kernel-3.10.0-302.el7.ppc64le How reproducible: 3/3 Steps to Reproduce: 1.Boot the guest with multiple queues and vhost=on: /usr/libexec/qemu-kvm...-netdev tap,id=hostnet1,script=/etc/qemu-ifup,vhost=on,queues=8 -device virtio-net-pci,netdev=hostnet1,id=net1,mac=00:54:5c:5b:5f:5a,disable-legacy=off,disable-modern=on,vectors=18,mq=on -smp 8 2.Transfer 6 files from guest to host,check the md5sum value of the files before and after transfered 3.Repeatedly change the working queues inside guest: # for i in $(seq 5000); do for j in 1 2 1 3 1 4 1 5 1 6 1 7 1 8; do ethtool -L eth0 combined $j ; ethtool -l eth0; done; done 4.ping guest from another external host, until the 5000 times finished. # ping $guest_ip 5.Transfer another 6 files from guest to host,and check the md5sum value of the files before and after transfered Actual results: After step2,all the md5sum value are same before and after file transfered from guest to host After step4,one time no packet lost,two times 1% packet lost After step5,some md5sum value changed after file transfered from guest to host,I test this issue 3 times ,one time 2 value changed,two times 1 value changed. Expected results: The md5sum value should alaways be same before and after file transfered from guest to host. Additional info:
Are you saying that you have been unable to reproduce with current kernel and qemu? If so, I think we will have to close as INSUFFICIENT_DATA. It's very surprising that the md5sum would change across the transfer with scp, since scp has internal checksumming. A file mismatch in this case would suggest not a network problem, but a problem either reading the disk on the source, or writing the disk on the destination. Unless the mismatch was due simply to a truncation of the files (interrupted transfer).
(In reply to David Gibson from comment #6) > Are you saying that you have been unable to reproduce with current kernel > and qemu? If so, I think we will have to close as INSUFFICIENT_DATA. > Yes，I cannot reproduce this issue with the current kernel and qemu. > It's very surprising that the md5sum would change across the transfer with > scp, since scp has internal checksumming. A file mismatch in this case > would suggest not a network problem, but a problem either reading the disk > on the source, or writing the disk on the destination. Unless the mismatch > was due simply to a truncation of the files (interrupted transfer). During the file transfer without any interrupt action.
Ok, since it can't be reproduced with the current, packages, I'm closing the bug.