[libvirt] A question for external snapshots with flag“VIR_DOMAIN_SNAPSHOT_CREATE_QUIESCE”
Hello, I got a question here. When we create consistent active external snapshots with flag “VIR_DOMAIN_SNAPSHOT_CREATE_QUIESCE” , the qemuDomainSnapshotFSFreeze will be called firstly to freeze all filesystems in vm, and then create snapshots. For windows vm, freezing filesystems used by vss service. If IO pressure in vm is too big or internal error happened in vss , freezing filesytems will failed , and qemuDomainSnapshotFSFreeze returns 0 which meaning that no filesystems are frozen. In the function qemuDomainSnapshotCreateActiveExternal, libvirt creates external snapshots all the same in such a situation that qemuDomainSnapshotFSFreeze returns 0, but the created snapshots are not consistent snapshots in such a situation . So shouldn't we abandon creating snapshots and goto cleanup in the situation that qemuDomainSnapshotFSFreeze returns 0? The code below: static int qemuDomainSnapshotCreateActiveExternal(virConnectPtr conn, virQEMUDriverPtr driver, virDomainObjPtr vm, virDomainSnapshotObjPtr snap, unsigned int flags) { virObjectEventPtr event; bool resume = false; int ret = -1; qemuDomainObjPrivatePtr priv = vm->privateData; char *xml = NULL; bool memory = snap->def->memory == VIR_DOMAIN_SNAPSHOT_LOCATION_EXTERNAL; bool memory_unlink = false; bool atomic = !!(flags & VIR_DOMAIN_SNAPSHOT_CREATE_ATOMIC); bool transaction = virQEMUCapsGet(priv->qemuCaps, QEMU_CAPS_TRANSACTION); int thaw = 0; /* 1 if freeze succeeded, -1 if freeze failed */ bool pmsuspended = false; virQEMUDriverConfigPtr cfg = NULL; int compressed; char *compressedpath = NULL; /* If quiesce was requested, then issue a freeze command, and a * counterpart thaw command when it is actually sent to agent. * The command will fail if the guest is paused or the guest agent * is not running, or is already quiesced. */ if (flags & VIR_DOMAIN_SNAPSHOT_CREATE_QUIESCE) { int freeze = qemuDomainSnapshotFSFreeze(driver, vm, NULL, 0); if (freeze < 0) { /* the helper reported the error */ if (freeze == -2) thaw = -1; /* the command is sent but agent failed */ goto cleanup; } thaw = 1; } ...... ......
please don't include libvir-list-owner in your mails. Also, it would be nice if you could convince your mailer to wrap long lines. On 02/28/2017 06:37 AM, lizhengui wrote:
Hello, I got a question here. When we create consistent active external snapshots with flag “VIR_DOMAIN_SNAPSHOT_CREATE_QUIESCE” , the qemuDomainSnapshotFSFreeze will be called firstly to freeze all filesystems in vm, and then create snapshots. For windows vm, freezing filesystems used by vss service. If IO pressure in vm is too big or internal error happened in vss , freezing filesytems will failed , and qemuDomainSnapshotFSFreeze returns 0 which meaning that no filesystems are frozen.
That sounds like a bug in the qga implementation for freezing file systems on Windows - if the system is not frozen, qga should return failure, not 0. Since libvirt doesn't maintain qga, this should probably be reported to the qemu folks.
In the function qemuDomainSnapshotCreateActiveExternal, libvirt creates external snapshots all the same in such a situation that qemuDomainSnapshotFSFreeze returns 0, but the created snapshots are not consistent snapshots in such a situation . So shouldn't we abandon creating snapshots and goto cleanup in the situation that qemuDomainSnapshotFSFreeze returns 0?
Yes, even if qga behavior is patched, libvirt should probably figure out if it is feasible to work around existing broken behavior. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
On Tue, Feb 28, 2017 at 09:22:16AM -0600, Eric Blake wrote:
please don't include libvir-list-owner in your mails. Also, it would be nice if you could convince your mailer to wrap long lines.
Also sending the question multiple times from multiple e-mail addresses makes bad reputation and creates huge mess. I'm also not at all in favour of adding random people in Cc. You're basically just spamming their inboxes.
On 02/28/2017 06:37 AM, lizhengui wrote:
Hello, I got a question here. When we create consistent active external snapshots with flag “VIR_DOMAIN_SNAPSHOT_CREATE_QUIESCE” , the qemuDomainSnapshotFSFreeze will be called firstly to freeze all filesystems in vm, and then create snapshots. For windows vm, freezing filesystems used by vss service. If IO pressure in vm is too big or internal error happened in vss , freezing filesytems will failed , and qemuDomainSnapshotFSFreeze returns 0 which meaning that no filesystems are frozen.
That sounds like a bug in the qga implementation for freezing file systems on Windows - if the system is not frozen, qga should return failure, not 0. Since libvirt doesn't maintain qga, this should probably be reported to the qemu folks.
In the function qemuDomainSnapshotCreateActiveExternal, libvirt creates external snapshots all the same in such a situation that qemuDomainSnapshotFSFreeze returns 0, but the created snapshots are not consistent snapshots in such a situation . So shouldn't we abandon creating snapshots and goto cleanup in the situation that qemuDomainSnapshotFSFreeze returns 0?
Yes, even if qga behavior is patched, libvirt should probably figure out if it is feasible to work around existing broken behavior.
According to our comment which states that the return value is number of filesystems freezed, I expected it to be documented behaviour. If that is the case, we can fix it right now. See my response in the other thread.
-- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
-- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list
participants (3)
-
Eric Blake -
lizhengui -
Martin Kletzander