Sparse VMs from Templates - Storage issues
Hi all, I have a setup as detailed below - iSCSI Storage Domain - Template with Thin QCOW2 disk - Multiple VMs from Template with Thin disk oVirt Node 4.4.4 When the VMs boots up it downloads some data to it and that leads to increase in volume size. I see that every few seconds the VM gets paused with "VM X has been paused due to no Storage space error." and then after few seconds "VM X has recovered from paused back to up" Sometimes after a many pause and recovery the VM dies with "VM X is down with error. Exit message: Lost connection with qemu process." and I have to restart the VMs. My questions. 1. How to work around this dying VM? 2. Is there a way to use sparse disks without VM being paused again and again? Thanks in advance. Shantur
On Tue, Aug 10, 2021 at 4:24 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi all,
I have a setup as detailed below
- iSCSI Storage Domain - Template with Thin QCOW2 disk - Multiple VMs from Template with Thin disk
Note that a single template disk used by many vms can become a performance bottleneck, and is a single point of failure. Cloning the template when creating vms avoids such issues.
oVirt Node 4.4.4
4.4.4 is old, you should upgrade to 4.4.7.
When the VMs boots up it downloads some data to it and that leads to increase in volume size. I see that every few seconds the VM gets paused with
"VM X has been paused due to no Storage space error."
and then after few seconds
"VM X has recovered from paused back to up"
This is normal operation when a vm writes too quickly and oVirt cannot extend the disk quick enough. To mitigate this, you can increase the volume chunk size. Created this configuration drop in file: # cat /etc/vdsm/vdsm.conf.d/99-local.conf [irs] volume_utilization_percent = 25 volume_utilization_chunk_mb = 2048 And restart vdsm. With this setting, when free space in a disk is 1.5g, the disk will be extended by 2g. With the default setting, when free space is 0.5g the disk was extended by 1g. If this does not eliminate the pauses, try a larger chunk size like 4096.
Sometimes after a many pause and recovery the VM dies with
"VM X is down with error. Exit message: Lost connection with qemu process."
This means qemu has crashed. You can find more info in the vm log at: /var/log/libvirt/qemu/vm-name.log We know about bugs in qemu that cause such crashes when vm disk is extended. I think the latest bug was fixed in 4.4.6, so upgrading to 4.4.7 will fix this issue. Even with these settings, if you have a very bursty io in the vm, it may become paused. The only way to completely avoid these pauses is to use a preallocated disk, or use file storage (e.g. NFS). Preallocated disk can be thin provisioned on the server side so it does not mean you need more storage, but you will not be able to use shared templates in the way you use them now. You can create vm from template, but the template is cloned to the new vm. Another option with (still tech preview) is Managed Block Storage (Cinder based storage). If your storage server is supported by Cinder, we can managed it using cinderlib. In this setup every disk is a LUN, which may be thin provisioned on the storage server. This can also offload storage operations to the server, like cloning disks, which may be much faster and more efficient. Nir
Thanks for the detailed response Nir. In my use case, we keep creating VMs from templates and deleting them so we need the VMs to be created quickly and cloning it will use a lot of time and storage. I will try to add the config and try again tomorrow. Also I like the Managed Block storage idea, I had read about it in the past and used it with Ceph. Just to understand it better, is this issue only on iSCSI based storage? Thanks again. Regards Shantur On Tue, Aug 10, 2021 at 9:26 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Tue, Aug 10, 2021 at 4:24 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi all,
I have a setup as detailed below
- iSCSI Storage Domain - Template with Thin QCOW2 disk - Multiple VMs from Template with Thin disk
Note that a single template disk used by many vms can become a performance bottleneck, and is a single point of failure. Cloning the template when creating vms avoids such issues.
oVirt Node 4.4.4
4.4.4 is old, you should upgrade to 4.4.7.
When the VMs boots up it downloads some data to it and that leads to increase in volume size. I see that every few seconds the VM gets paused with
"VM X has been paused due to no Storage space error."
and then after few seconds
"VM X has recovered from paused back to up"
This is normal operation when a vm writes too quickly and oVirt cannot extend the disk quick enough. To mitigate this, you can increase the volume chunk size.
Created this configuration drop in file:
# cat /etc/vdsm/vdsm.conf.d/99-local.conf [irs] volume_utilization_percent = 25 volume_utilization_chunk_mb = 2048
And restart vdsm.
With this setting, when free space in a disk is 1.5g, the disk will be extended by 2g. With the default setting, when free space is 0.5g the disk was extended by 1g.
If this does not eliminate the pauses, try a larger chunk size like 4096.
Sometimes after a many pause and recovery the VM dies with
"VM X is down with error. Exit message: Lost connection with qemu process."
This means qemu has crashed. You can find more info in the vm log at: /var/log/libvirt/qemu/vm-name.log
We know about bugs in qemu that cause such crashes when vm disk is extended. I think the latest bug was fixed in 4.4.6, so upgrading to 4.4.7 will fix this issue.
Even with these settings, if you have a very bursty io in the vm, it may become paused. The only way to completely avoid these pauses is to use a preallocated disk, or use file storage (e.g. NFS). Preallocated disk can be thin provisioned on the server side so it does not mean you need more storage, but you will not be able to use shared templates in the way you use them now. You can create vm from template, but the template is cloned to the new vm.
Another option with (still tech preview) is Managed Block Storage (Cinder based storage). If your storage server is supported by Cinder, we can managed it using cinderlib. In this setup every disk is a LUN, which may be thin provisioned on the storage server. This can also offload storage operations to the server, like cloning disks, which may be much faster and more efficient.
Nir
On Wed, Aug 11, 2021 at 12:43 AM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Thanks for the detailed response Nir.
In my use case, we keep creating VMs from templates and deleting them so we need the VMs to be created quickly and cloning it will use a lot of time and storage.
That's a good reason to use a template. If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm. Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I will try to add the config and try again tomorrow. Also I like the Managed Block storage idea, I had read about it in the past and used it with Ceph.
Just to understand it better, is this issue only on iSCSI based storage?
Yes, on file based storage a snapshot is a file, and it grows as needed. On block based storage, a snapshot is a logical volume, and oVirt needs to extend it when needed. Nir
Thanks again.
Regards Shantur
On Tue, Aug 10, 2021 at 9:26 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Tue, Aug 10, 2021 at 4:24 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi all,
I have a setup as detailed below
- iSCSI Storage Domain - Template with Thin QCOW2 disk - Multiple VMs from Template with Thin disk
Note that a single template disk used by many vms can become a performance bottleneck, and is a single point of failure. Cloning the template when creating vms avoids such issues.
oVirt Node 4.4.4
4.4.4 is old, you should upgrade to 4.4.7.
When the VMs boots up it downloads some data to it and that leads to increase in volume size. I see that every few seconds the VM gets paused with
"VM X has been paused due to no Storage space error."
and then after few seconds
"VM X has recovered from paused back to up"
This is normal operation when a vm writes too quickly and oVirt cannot extend the disk quick enough. To mitigate this, you can increase the volume chunk size.
Created this configuration drop in file:
# cat /etc/vdsm/vdsm.conf.d/99-local.conf [irs] volume_utilization_percent = 25 volume_utilization_chunk_mb = 2048
And restart vdsm.
With this setting, when free space in a disk is 1.5g, the disk will be extended by 2g. With the default setting, when free space is 0.5g the disk was extended by 1g.
If this does not eliminate the pauses, try a larger chunk size like 4096.
Sometimes after a many pause and recovery the VM dies with
"VM X is down with error. Exit message: Lost connection with qemu process."
This means qemu has crashed. You can find more info in the vm log at: /var/log/libvirt/qemu/vm-name.log
We know about bugs in qemu that cause such crashes when vm disk is extended. I think the latest bug was fixed in 4.4.6, so upgrading to 4.4.7 will fix this issue.
Even with these settings, if you have a very bursty io in the vm, it may become paused. The only way to completely avoid these pauses is to use a preallocated disk, or use file storage (e.g. NFS). Preallocated disk can be thin provisioned on the server side so it does not mean you need more storage, but you will not be able to use shared templates in the way you use them now. You can create vm from template, but the template is cloned to the new vm.
Another option with (still tech preview) is Managed Block Storage (Cinder based storage). If your storage server is supported by Cinder, we can managed it using cinderlib. In this setup every disk is a LUN, which may be thin provisioned on the storage server. This can also offload storage operations to the server, like cloning disks, which may be much faster and more efficient.
Nir
If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm.
Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I think that's how stateless VMs work
On Wed, Aug 11, 2021 at 2:56 PM Benny Zlotnik <bzlotnik@redhat.com> wrote:
If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm.
Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I think that's how stateless VMs work
+1 It doesn't work exactly like Nir wrote above - stateless VMs that are thin-provisioned would have a qcow volume on top of each template's volume and when they starts, their active volume would be a qcow volume on top of the aforementioned qcow volume and that active volume will be removed when the VM goes down But yeah, stateless VMs are intended for such use case
On Wed, Aug 11, 2021 at 4:24 PM Arik Hadas <ahadas@redhat.com> wrote:
On Wed, Aug 11, 2021 at 2:56 PM Benny Zlotnik <bzlotnik@redhat.com> wrote:
If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm.
Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I think that's how stateless VMs work
+1 It doesn't work exactly like Nir wrote above - stateless VMs that are thin-provisioned would have a qcow volume on top of each template's volume and when they starts, their active volume would be a qcow volume on top of the aforementioned qcow volume and that active volume will be removed when the VM goes down But yeah, stateless VMs are intended for such use case
I was referring to transient disks - created in vdsm: https://github.com/oVirt/vdsm/blob/45903d01e142047093bf844628b5d90df12b6ffb/... This creates a *local* temporary file using qcow2 format, using the disk on shared storage as a backing file. Maybe this is not used by engine?
Hi Nir, Just to report. As suggested, I created a Posix compliant storage domain with CephFS and copied my templates to CephFS. Now I created VMs from CephFS templates and the storage error happens again. As I understand, the storage growth issue is only on iSCSI. Am I doing something wrong? Kind regards, Shantur On Wed, Aug 11, 2021 at 2:42 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Wed, Aug 11, 2021 at 4:24 PM Arik Hadas <ahadas@redhat.com> wrote:
On Wed, Aug 11, 2021 at 2:56 PM Benny Zlotnik <bzlotnik@redhat.com> wrote:
If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm.
Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I think that's how stateless VMs work
+1 It doesn't work exactly like Nir wrote above - stateless VMs that are thin-provisioned would have a qcow volume on top of each template's volume and when they starts, their active volume would be a qcow volume on top of the aforementioned qcow volume and that active volume will be removed when the VM goes down But yeah, stateless VMs are intended for such use case
I was referring to transient disks - created in vdsm: https://github.com/oVirt/vdsm/blob/45903d01e142047093bf844628b5d90df12b6ffb/...
This creates a *local* temporary file using qcow2 format, using the disk on shared storage as a backing file.
Maybe this is not used by engine?
I have actually tried many types of storage now and all have this issue. I am not of ideas what to do On Wed, Sep 22, 2021 at 4:39 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi Nir,
Just to report. As suggested, I created a Posix compliant storage domain with CephFS and copied my templates to CephFS. Now I created VMs from CephFS templates and the storage error happens again. As I understand, the storage growth issue is only on iSCSI.
Am I doing something wrong?
Kind regards, Shantur
On Wed, Aug 11, 2021 at 2:42 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Wed, Aug 11, 2021 at 4:24 PM Arik Hadas <ahadas@redhat.com> wrote:
On Wed, Aug 11, 2021 at 2:56 PM Benny Zlotnik <bzlotnik@redhat.com> wrote:
If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm.
Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I think that's how stateless VMs work
+1 It doesn't work exactly like Nir wrote above - stateless VMs that are thin-provisioned would have a qcow volume on top of each template's volume and when they starts, their active volume would be a qcow volume on top of the aforementioned qcow volume and that active volume will be removed when the VM goes down But yeah, stateless VMs are intended for such use case
I was referring to transient disks - created in vdsm: https://github.com/oVirt/vdsm/blob/45903d01e142047093bf844628b5d90df12b6ffb/...
This creates a *local* temporary file using qcow2 format, using the disk on shared storage as a backing file.
Maybe this is not used by engine?
On Wednesday, 22 September 2021 18:09:28 CEST Shantur Rathore wrote:
I have actually tried many types of storage now and all have this issue.
This is weird. Could you please use file-based storage (e.g. NFS) and post here whole exceptions from vdsm log (/var/log/vdsm/vdsm.log) and qemu log (/var/ log/libvirt/qemu/vm_name.log) from the host which runs the VM? Hopefully this will give us some hint what is the real issue there. Thanks Vojta
I am not of ideas what to do
On Wed, Sep 22, 2021 at 4:39 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi Nir,
Just to report. As suggested, I created a Posix compliant storage domain with CephFS and copied my templates to CephFS. Now I created VMs from CephFS templates and the storage error happens again.
As I understand, the storage growth issue is only on iSCSI.
Am I doing something wrong?
Kind regards, Shantur
On Wed, Aug 11, 2021 at 2:42 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Wed, Aug 11, 2021 at 4:24 PM Arik Hadas <ahadas@redhat.com> wrote:
On Wed, Aug 11, 2021 at 2:56 PM Benny Zlotnik <bzlotnik@redhat.com> wrote:
If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm.
Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I think that's how stateless VMs work
+1 It doesn't work exactly like Nir wrote above - stateless VMs that are thin-provisioned would have a qcow volume on top of each template's volume and when they starts, their active volume would be a qcow volume on top of the aforementioned qcow volume and that active volume will be removed when the VM goes down
But yeah, stateless VMs
are intended for such use case
I was referring to transient disks - created in vdsm: https://github.com/oVirt/vdsm/blob/45903d01e142047093bf844628b5d90df12b6 ffb/lib/vdsm/virt/vm.py#L3789
This creates a *local* temporary file using qcow2 format, using the disk on shared storage as a backing file.
Maybe this is not used by engine?
_______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/3UEXYH2IGNDWW YEHEHKLAREJS74LMXUI/
Yes, on file based storage a snapshot is a file, and it grows as needed. On block based storage, a snapshot is a logical volume, and oVirt needs to extend it when needed.
Forgive my ignorance, coming from vSphere background where a filesystem was created on iSCSI LUN. I take that this isn't the case in case of a iSCSI Storage Domain in oVirt. On Wed, Aug 11, 2021 at 12:26 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Wed, Aug 11, 2021 at 12:43 AM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Thanks for the detailed response Nir.
In my use case, we keep creating VMs from templates and deleting them so
we need the VMs to be created quickly and cloning it will use a lot of time and storage.
That's a good reason to use a template.
If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm.
Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I will try to add the config and try again tomorrow. Also I like the Managed Block storage idea, I had read about it in the past and used it with Ceph.
Just to understand it better, is this issue only on iSCSI based storage?
Yes, on file based storage a snapshot is a file, and it grows as needed. On block based storage, a snapshot is a logical volume, and oVirt needs to extend it when needed.
Nir
Thanks again.
Regards Shantur
On Tue, Aug 10, 2021 at 9:26 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Tue, Aug 10, 2021 at 4:24 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi all,
I have a setup as detailed below
- iSCSI Storage Domain - Template with Thin QCOW2 disk - Multiple VMs from Template with Thin disk
Note that a single template disk used by many vms can become a
bottleneck, and is a single point of failure. Cloning the template when creating vms avoids such issues.
oVirt Node 4.4.4
4.4.4 is old, you should upgrade to 4.4.7.
When the VMs boots up it downloads some data to it and that leads to increase in volume size. I see that every few seconds the VM gets paused with
"VM X has been paused due to no Storage space error."
and then after few seconds
"VM X has recovered from paused back to up"
This is normal operation when a vm writes too quickly and oVirt cannot extend the disk quick enough. To mitigate this, you can increase the volume chunk size.
Created this configuration drop in file:
# cat /etc/vdsm/vdsm.conf.d/99-local.conf [irs] volume_utilization_percent = 25 volume_utilization_chunk_mb = 2048
And restart vdsm.
With this setting, when free space in a disk is 1.5g, the disk will be extended by 2g. With the default setting, when free space is 0.5g the disk was extended by 1g.
If this does not eliminate the pauses, try a larger chunk size like 4096.
Sometimes after a many pause and recovery the VM dies with
"VM X is down with error. Exit message: Lost connection with qemu
performance process."
This means qemu has crashed. You can find more info in the vm log at: /var/log/libvirt/qemu/vm-name.log
We know about bugs in qemu that cause such crashes when vm disk is extended. I think the latest bug was fixed in 4.4.6, so upgrading to
4.4.7
will fix this issue.
Even with these settings, if you have a very bursty io in the vm, it may become paused. The only way to completely avoid these pauses is to use a preallocated disk, or use file storage (e.g. NFS). Preallocated disk can be thin provisioned on the server side so it does not mean you need more storage, but you will not be able to use shared templates in the way you use them now. You can create vm from template, but the template is cloned to the new vm.
Another option with (still tech preview) is Managed Block Storage (Cinder based storage). If your storage server is supported by Cinder, we can managed it using cinderlib. In this setup every disk is a LUN, which may be thin provisioned on the storage server. This can also offload storage operations to the server, like cloning disks, which may be much faster and more efficient.
Nir
On Wed, Aug 11, 2021 at 3:13 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Yes, on file based storage a snapshot is a file, and it grows as needed. On block based storage, a snapshot is a logical volume, and oVirt needs to extend it when needed.
Forgive my ignorance, coming from vSphere background where a filesystem was created on iSCSI LUN. I take that this isn't the case in case of a iSCSI Storage Domain in oVirt.
Yes, for block storage, we create a LVM volume group with one or more LUNs to create a storage domain. Disks are created using LVM logical volume on this VG. When you create a vm from template on block storage we create a new 1g logical volume for the vm disk, and create a qcow2 image on this logical volume with the backing file using the template logical volume. The logical volume needs to be extended when free space is low. This is done automatically on the host running the VM, but since oVirt is not in the data path, the VM may write data too fast and pause trying to write after the end of the logical volume. In this case the VM will be resumed when oVirt finish to extend the volume.
On Wed, Aug 11, 2021 at 12:26 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Wed, Aug 11, 2021 at 12:43 AM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Thanks for the detailed response Nir.
In my use case, we keep creating VMs from templates and deleting them so we need the VMs to be created quickly and cloning it will use a lot of time and storage.
That's a good reason to use a template.
If your vm is temporary and you like to drop the data written while the vm is running, you could use a temporary disk based on the template. This is called a "transient disk" in vdsm.
Arik, maybe you remember how transient disks are used in engine? Do we have an API to run a VM once, dropping the changes to the disk done while the VM was running?
I will try to add the config and try again tomorrow. Also I like the Managed Block storage idea, I had read about it in the past and used it with Ceph.
Just to understand it better, is this issue only on iSCSI based storage?
Yes, on file based storage a snapshot is a file, and it grows as needed. On block based storage, a snapshot is a logical volume, and oVirt needs to extend it when needed.
Nir
Thanks again.
Regards Shantur
On Tue, Aug 10, 2021 at 9:26 PM Nir Soffer <nsoffer@redhat.com> wrote:
On Tue, Aug 10, 2021 at 4:24 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi all,
I have a setup as detailed below
- iSCSI Storage Domain - Template with Thin QCOW2 disk - Multiple VMs from Template with Thin disk
Note that a single template disk used by many vms can become a performance bottleneck, and is a single point of failure. Cloning the template when creating vms avoids such issues.
oVirt Node 4.4.4
4.4.4 is old, you should upgrade to 4.4.7.
When the VMs boots up it downloads some data to it and that leads to increase in volume size. I see that every few seconds the VM gets paused with
"VM X has been paused due to no Storage space error."
and then after few seconds
"VM X has recovered from paused back to up"
This is normal operation when a vm writes too quickly and oVirt cannot extend the disk quick enough. To mitigate this, you can increase the volume chunk size.
Created this configuration drop in file:
# cat /etc/vdsm/vdsm.conf.d/99-local.conf [irs] volume_utilization_percent = 25 volume_utilization_chunk_mb = 2048
And restart vdsm.
With this setting, when free space in a disk is 1.5g, the disk will be extended by 2g. With the default setting, when free space is 0.5g the disk was extended by 1g.
If this does not eliminate the pauses, try a larger chunk size like 4096.
Sometimes after a many pause and recovery the VM dies with
"VM X is down with error. Exit message: Lost connection with qemu process."
This means qemu has crashed. You can find more info in the vm log at: /var/log/libvirt/qemu/vm-name.log
We know about bugs in qemu that cause such crashes when vm disk is extended. I think the latest bug was fixed in 4.4.6, so upgrading to 4.4.7 will fix this issue.
Even with these settings, if you have a very bursty io in the vm, it may become paused. The only way to completely avoid these pauses is to use a preallocated disk, or use file storage (e.g. NFS). Preallocated disk can be thin provisioned on the server side so it does not mean you need more storage, but you will not be able to use shared templates in the way you use them now. You can create vm from template, but the template is cloned to the new vm.
Another option with (still tech preview) is Managed Block Storage (Cinder based storage). If your storage server is supported by Cinder, we can managed it using cinderlib. In this setup every disk is a LUN, which may be thin provisioned on the storage server. This can also offload storage operations to the server, like cloning disks, which may be much faster and more efficient.
Nir
Another option with (still tech preview) is Managed Block Storage (Cinder based storage). It still tech preview in 4.4 ?? -----Original Message----- From: users-bounces@ovirt.org <users-bounces@ovirt.org> On Behalf Of Nir Soffer Sent: Wednesday, August 11, 2021 4:26 AM To: Shantur Rathore <shantur.rathore@gmail.com> Cc: users <users@ovirt.org>; Roman Bednar <rbednar@redhat.com> Subject: [ovirt-users] Re: Sparse VMs from Templates - Storage issues On Tue, Aug 10, 2021 at 4:24 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi all,
I have a setup as detailed below
- iSCSI Storage Domain - Template with Thin QCOW2 disk - Multiple VMs from Template with Thin disk
Note that a single template disk used by many vms can become a performance bottleneck, and is a single point of failure. Cloning the template when creating vms avoids such issues.
oVirt Node 4.4.4
4.4.4 is old, you should upgrade to 4.4.7.
When the VMs boots up it downloads some data to it and that leads to increase in volume size. I see that every few seconds the VM gets paused with
"VM X has been paused due to no Storage space error."
and then after few seconds
"VM X has recovered from paused back to up"
This is normal operation when a vm writes too quickly and oVirt cannot extend the disk quick enough. To mitigate this, you can increase the volume chunk size. Created this configuration drop in file: # cat /etc/vdsm/vdsm.conf.d/99-local.conf [irs] volume_utilization_percent = 25 volume_utilization_chunk_mb = 2048 And restart vdsm. With this setting, when free space in a disk is 1.5g, the disk will be extended by 2g. With the default setting, when free space is 0.5g the disk was extended by 1g. If this does not eliminate the pauses, try a larger chunk size like 4096.
Sometimes after a many pause and recovery the VM dies with
"VM X is down with error. Exit message: Lost connection with qemu process."
This means qemu has crashed. You can find more info in the vm log at: /var/log/libvirt/qemu/vm-name.log We know about bugs in qemu that cause such crashes when vm disk is extended. I think the latest bug was fixed in 4.4.6, so upgrading to 4.4.7 will fix this issue. Even with these settings, if you have a very bursty io in the vm, it may become paused. The only way to completely avoid these pauses is to use a preallocated disk, or use file storage (e.g. NFS). Preallocated disk can be thin provisioned on the server side so it does not mean you need more storage, but you will not be able to use shared templates in the way you use them now. You can create vm from template, but the template is cloned to the new vm. Another option with (still tech preview) is Managed Block Storage (Cinder based storage). If your storage server is supported by Cinder, we can managed it using cinderlib. In this setup every disk is a LUN, which may be thin provisioned on the storage server. This can also offload storage operations to the server, like cloning disks, which may be much faster and more efficient. Nir _______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/W653KLDZMLUNMK...
So, I did more digging and now I know how to reproduce it. I created a VM and added a disk on local ssd using scratchpad hook, formatted and mounted this scratchdisk. Now, when I try to do heavy IO on this scratchdisk on local ssd, like, dd if=/dev/zero of=/mnt/scratchdisk/test bs=1M count=10000, qemu pauses VM. Debug logs in libvirt shows 2021-09-23 11:04:32.765+0000: 463319: debug : virThreadJobSet:94 : Thread 463319 (rpc-worker) is now running job remoteDispatchNodeGetFreePages 2021-09-23 11:04:32.765+0000: 463319: debug : virNodeGetFreePages:1614 : conn=0x7f8620018ba0, npages=3, pages=0x7f8670009960, startCell=4294967295, cellCount=1, counts=0x7f8670007db0, flags=0x0 2021-09-23 11:04:32.765+0000: 463319: debug : virThreadJobClear:119 : Thread 463319 (rpc-worker) finished job remoteDispatchNodeGetFreePages with ret=0 2021-09-23 11:04:34.235+0000: 488774: debug : qemuMonitorJSONIOProcessLine:220 : Line [{"timestamp": {"seconds": 1632395074, "microseconds": 235454}, "event": "BLOCK_IO_ERROR", "data": {"device": "", "nospace": false, "node-name": "libvirt-3-format", "reason": "Input/output error", "operation": "write", "action": "stop"}}] 2021-09-23 11:04:34.235+0000: 488774: info : qemuMonitorJSONIOProcessLine:235 : QEMU_MONITOR_RECV_EVENT: mon=0x7f860c14b700 event={"timestamp": {"seconds": 1632395074, "microseconds": 235454}, "event": "BLOCK_IO_ERROR", "data": {"device": "", "nospace": false, "node-name": "libvirt-3-format", "reason": "Input/output error", "operation": "write", "action": "stop"}} 2021-09-23 11:04:34.235+0000: 488774: debug : qemuMonitorJSONIOProcessEvent:181 : mon=0x7f860c14b700 obj=0x7f860c0e7450 2021-09-23 11:04:34.235+0000: 488774: debug : qemuMonitorEmitEvent:1166 : mon=0x7f860c14b700 event=BLOCK_IO_ERROR 2021-09-23 11:04:34.235+0000: 488774: debug : qemuProcessHandleEvent:581 : vm=0x7f86201d6df0 2021-09-23 11:04:34.235+0000: 488774: debug : virObjectEventNew:624 : obj=0x7f860c0d82f0 2021-09-23 11:04:34.235+0000: 488774: debug : qemuMonitorJSONIOProcessEvent:206 : handle BLOCK_IO_ERROR handler=0x7f8639c77a90 data=0x7f860c0661c0 To confirm the local ssd is fine, have enough space where scratch disk is located and I could run dd in host without any issues. This happens on other storages as well. So this seems like an issue with qemu when heavy IO is happening on a disk. On Thu, Sep 23, 2021 at 7:19 AM Tommy Sway <sz_cuitao@163.com> wrote:
Another option with (still tech preview) is Managed Block Storage (Cinder based storage).
It still tech preview in 4.4 ??
-----Original Message----- From: users-bounces@ovirt.org <users-bounces@ovirt.org> On Behalf Of Nir Soffer Sent: Wednesday, August 11, 2021 4:26 AM To: Shantur Rathore <shantur.rathore@gmail.com> Cc: users <users@ovirt.org>; Roman Bednar <rbednar@redhat.com> Subject: [ovirt-users] Re: Sparse VMs from Templates - Storage issues
On Tue, Aug 10, 2021 at 4:24 PM Shantur Rathore <shantur.rathore@gmail.com> wrote:
Hi all,
I have a setup as detailed below
- iSCSI Storage Domain - Template with Thin QCOW2 disk - Multiple VMs from Template with Thin disk
Note that a single template disk used by many vms can become a performance bottleneck, and is a single point of failure. Cloning the template when creating vms avoids such issues.
oVirt Node 4.4.4
4.4.4 is old, you should upgrade to 4.4.7.
When the VMs boots up it downloads some data to it and that leads to increase in volume size. I see that every few seconds the VM gets paused with
"VM X has been paused due to no Storage space error."
and then after few seconds
"VM X has recovered from paused back to up"
This is normal operation when a vm writes too quickly and oVirt cannot extend the disk quick enough. To mitigate this, you can increase the volume chunk size.
Created this configuration drop in file:
# cat /etc/vdsm/vdsm.conf.d/99-local.conf [irs] volume_utilization_percent = 25 volume_utilization_chunk_mb = 2048
And restart vdsm.
With this setting, when free space in a disk is 1.5g, the disk will be extended by 2g. With the default setting, when free space is 0.5g the disk was extended by 1g.
If this does not eliminate the pauses, try a larger chunk size like 4096.
Sometimes after a many pause and recovery the VM dies with
"VM X is down with error. Exit message: Lost connection with qemu process."
This means qemu has crashed. You can find more info in the vm log at: /var/log/libvirt/qemu/vm-name.log
We know about bugs in qemu that cause such crashes when vm disk is extended. I think the latest bug was fixed in 4.4.6, so upgrading to 4.4.7 will fix this issue.
Even with these settings, if you have a very bursty io in the vm, it may become paused. The only way to completely avoid these pauses is to use a preallocated disk, or use file storage (e.g. NFS). Preallocated disk can be thin provisioned on the server side so it does not mean you need more storage, but you will not be able to use shared templates in the way you use them now. You can create vm from template, but the template is cloned to the new vm.
Another option with (still tech preview) is Managed Block Storage (Cinder based storage). If your storage server is supported by Cinder, we can managed it using cinderlib. In this setup every disk is a LUN, which may be thin provisioned on the storage server. This can also offload storage operations to the server, like cloning disks, which may be much faster and more efficient.
Nir _______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/W653KLDZMLUNMK...
participants (6)
-
Arik Hadas -
Benny Zlotnik -
Nir Soffer -
Shantur Rathore -
Tommy Sway -
Vojtech Juranek