On Fri, Feb 15, 2019 at 03:03:34AM -0000, Cheng Chen wrote:
> When the drive-backup is running, I manually crash the iscsi server(or
> interrupt network, eg. iptables -j DROP).
>
> Then after less than 1 minute:
> virsh qemu-monitor-command ${DOMAIN} --pretty '{ "execute": "query-block" }' will block and no any response, until timeout. This is still excusable.
> But, the disk(drive-virtio-disk0)will occur the same situation:in vm os, the disk will block and no any response.
>
> In other words, when qemu and iscsi-server lose contact, It will cause
> the vm unable.
I haven't tried to reproduce this but I guess QEMU reaches a
synchronization point where it waits for all outstanding requests to
complete. Since the iSCSI target is unresponsive QEMU gets stuck.
These issues can sometimes be fixed by avoiding the synchronization
point (a backtrace should reveal where the main loop thread is stuck)
but other times it really is necessary to wait for all requests and the
solution isn't as obvious.
On Fri, Feb 15, 2019 at 03:03:34AM -0000, Cheng Chen wrote: command ${DOMAIN} --pretty '{ "execute": "query-block" }' will block and no any response, until timeout. This is still excusable. virtio- disk0)will occur the same situation:in vm os, the disk will block and no any response.
> When the drive-backup is running, I manually crash the iscsi server(or
> interrupt network, eg. iptables -j DROP).
>
> Then after less than 1 minute:
> virsh qemu-monitor-
> But, the disk(drive-
>
> In other words, when qemu and iscsi-server lose contact, It will cause
> the vm unable.
I haven't tried to reproduce this but I guess QEMU reaches a
synchronization point where it waits for all outstanding requests to
complete. Since the iSCSI target is unresponsive QEMU gets stuck.
These issues can sometimes be fixed by avoiding the synchronization
point (a backtrace should reveal where the main loop thread is stuck)
but other times it really is necessary to wait for all requests and the
solution isn't as obvious.