Re: [pve-devel] cache unsafe write for qemu-img convert, vma extract and vmstate snapshot

for the vmstate, on a ceph ssd cluster, I have now 10s vs 50s for 32gb vmstate. - Mail original - De: "aderumier" À: "pve-devel" Envoyé: Lundi 1 Août 2016 08:50:59 Objet: [pve-devel] cache unsafe write for qemu-img convert, vma extract and vmstate snapshot As reported by Eneko La

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

On 08/01/2016 08:51 AM, Alexandre Derumier wrote: Signed-off-by: Alexandre Derumier --- PVE/QemuServer.pm | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm index 7778fb8..2414fd8 100644 --- a/PVE/QemuServer.pm +++ b/PVE/QemuServer.pm @@ -5

Re: [pve-devel] [PATCH container 3/3] don't let lxc handle container reboots directly

On Thu, Jul 28, 2016 at 05:25:19PM +0200, Dietmar Maurer wrote: > Would be great to include references to bug reports here (if any).. Well there's this one: , where they state you can't reload the config, but it's not a bug, it's a design choice or missing f

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

> On August 1, 2016 at 9:26 AM Dominik Csapak wrote: > > > On 08/01/2016 08:51 AM, Alexandre Derumier wrote: > > Signed-off-by: Alexandre Derumier > > --- > > PVE/QemuServer.pm | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.p

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

> > is this really safe? > > > > this also impacts cloning and the "move disk" function. > > what if i clone a vm to an nfs share and immediately move the vm to > > another host, then start it? > > I guess qemu-img does a flush before closing (or flush is automatic on > close?)? Answering mysel

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

El 01/08/16 a las 09:26, Dominik Csapak escribió: On 08/01/2016 08:51 AM, Alexandre Derumier wrote: Signed-off-by: Alexandre Derumier --- PVE/QemuServer.pm | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm index 7778fb8..2414fd8 100644 --

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

>>Answering myself, 'close' does not issue flush/fsync. close send a flush block.c static void bdrv_close(BlockDriverState *bs) { BdrvAioNotifier *ban, *ban_next; assert(!bs->job); /* Disable I/O limits and drain all pending throttled requests */ if (bs->throttle_state) {

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

> Am 01.08.2016 um 09:26 schrieb Dominik Csapak : > > On 08/01/2016 08:51 AM, Alexandre Derumier wrote: >> Signed-off-by: Alexandre Derumier >> --- >> PVE/QemuServer.pm | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm >> index 7778

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

ok nevermind then On 08/01/2016 09:44 AM, Alexandre DERUMIER wrote: Answering myself, 'close' does not issue flush/fsync. close send a flush block.c static void bdrv_close(BlockDriverState *bs) { BdrvAioNotifier *ban, *ban_next; assert(!bs->job); /* Disable I/O limits and drain

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

On Mon, Aug 01, 2016 at 09:48:17AM +0200, Dominik Csapak wrote: > On 08/01/2016 09:44 AM, Alexandre DERUMIER wrote: > >>>Answering myself, 'close' does not issue flush/fsync. > > > >close send a flush > > > > ok nevermind then Even if it didn't, if we're already patching it, adding a flush-on-clos

Re: [pve-devel] cache unsafe write for qemu-img convert, vma extract and vmstate snapshot

El 01/08/16 a las 08:50, Alexandre Derumier escribió: As reported by Eneko Lacunza, ceph/rbd by default use writethrough even if writeback is defined, until a flush is detected. We currently use writeback without sendind any flush for 3 things: qemu-img convert : (default is unsafe, but we fo

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

Applied. qemu-img does a blk_unref() which at the end on the device which should eventually end up in bdrv_close() doing a flush(). So this should be safe in theory. On Mon, Aug 01, 2016 at 08:51:00AM +0200, Alexandre Derumier wrote: > Signed-off-by: Alexandre Derumier > --- > PVE/QemuServer.pm

Re: [pve-devel] [PATCH 1/2] rbd : disable cache_writethtrough_until_flush with cache=unsafe

applied, this should only affect "unsafe" settings anyway as it only applies to BDRV_OO_NO_FLUSH. On Mon, Aug 01, 2016 at 08:51:01AM +0200, Alexandre Derumier wrote: > Signed-off-by: Alexandre Derumier > --- > ...-rbd_cache_writethrough_until_flush-with-.patch | 29 > ++ > d

Re: [pve-devel] [PATCH 2/2] enable cache=unsafe for vma extract_content and qmp_savevm_start

Applied. Would be nice to have an option there, but it makes sense for our use cases considering qemu shoulf always flush on close in these cases. On Mon, Aug 01, 2016 at 08:51:02AM +0200, Alexandre Derumier wrote: > We don't send any flush here, so we need to open with cache=unsafe. > > Signed-o

Re: [pve-devel] [PATCH kvm 1/2] fix various CVEs

applied both patches On Tue, Jul 26, 2016 at 11:51:37AM +0200, Thomas Lamprecht wrote: > For upstream commits 926cde5f3e4d2504ed161ed0 and > cc96677469388bad3d664793 is no CVE number assigned yet. > > Signed-off-by: Thomas Lamprecht > --- > Readded CVE CVE-2016-2391 and CVE-2016-5126 > > patch

[pve-devel] [PATCH manager] When basic capabilities are missing, don't display the corresponding actions in the tree menu

Note that we *hide* the corresponding actions, instead of disabling by greying out the menu command. Disabling here does not make sense, since a low privilege user has no way to re enabling the menu commands by himself. This is mainly a usability improvement, as the real capabilities test is done

[pve-devel] [PATCH kvm] Fix CVE-2016-6490: virtio: check vring descriptor buffer length

--- Note: already applied ...rtio-check-vring-descriptor-buffer-length.patch | 37 ++ debian/patches/series | 1 + 2 files changed, 38 insertions(+) create mode 100644 debian/patches/extra/CVE-2016-6490-virtio-check-vring-descriptor-buffer-lengt

Re: [pve-devel] [PATCH manager] When basic capabilities are missing, don't display the corresponding actions in the tree menu

How refreshing to see a 4 year old issue get taken care of! ;) Thanks El 01/08/16 a las 14:20, Emmanuel Kasper escribió: Note that we *hide* the corresponding actions, instead of disabling by greying out the menu command. Disabling here does not make sense, since a low privilege user has no wa

Re: [pve-devel] [PATCH] qemu-img convert : use default cache=unsafe instead writeback

> Am 01.08.2016 um 09:44 schrieb Alexandre DERUMIER : > >>> Answering myself, 'close' does not issue flush/fsync. > > close send a flush > > It's was introduce by this commit > > [Qemu-devel] [PATCH v3] qemu-img: let 'qemu-img convert' flush data > https://lists.nongnu.org/archive/html/qemu-d

[pve-devel] [PATCH 2/2] add package for QDevice Net Daemon

The QDevice Net Daemon is the daemon running outside of the cluster with the purpose of providing a vote to the corosync-qdevice model "net". Create an own package for it as it's not related to corosync directly (i.e. it has no dependency on (lib)corosync) and may run on a non Proxmox VE server.

[pve-devel] [PATCH 0/2] Add initial QDevice support

Include QDevice support in our corosync package and add a new package for the QDevice Net Daemon. The command corosync-qdevice-net-certutil -h gives a good overview for setting up QDevice for a cluster. NOTE: I strongly discourage using QDevices in cluster setup with an uneven node count,

[pve-devel] [PATCH 1/2] compile corosync with QDevice support

Compile qdevices with corosync and add a systemd service file for the qdevice daemon. Disable this daemon per default. QDevices need certutil for generating and signing certificates, needed for a secure communication over TLS with the QDevice Net Daemon (qnetd). certutil is included in libnss3-too

Re: [pve-devel] [PATCH 0/2] Add initial QDevice support

On 08/01/2016 03:54 PM, Thomas Lamprecht wrote: Include QDevice support in our corosync package and add a new package for the QDevice Net Daemon. The command corosync-qdevice-net-certutil -h gives a good overview for setting up QDevice for a cluster. NOTE: I strongly discourage using QDevic

[pve-devel] [PATCH manager v2] closes #195 : hide disallowed commands in tree

When basic capabilities are missing, don't display the corresponding actions in the tree menu. Note that we *hide* the corresponding actions, instead of disabling by greying out the menu command. Disabling here does not make sense, since a low privilege user has no way to reenable the menu command

Re: [pve-devel] [PATCH manager v2] closes #195 : hide disallowed commands in tree

applied, thanks. ___ pve-devel mailing list pve-devel@pve.proxmox.com http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel