Hi all! My scenario: We currently use oVirt and are studying the possibility of migrating to OpenStack in the future. Today we use Storages Enterprise 3PAR and Dell PowerVault. I've even tried to integrate 3PAR, but it's too complex for me for now... Stupid question... If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ? Cheers! -- Att, Jorge Visentini +55 55 98432-9868
Hi Jorge, With the LVM driver: 1. Live migration should be supported although nova team can confirm 2. I don't know what you mean by HA instances (maybe it's a nova specific thing which again they can confirm) 3. Volume Snapshot is supported with the LVM driver (and it is one of the core features required for a driver to be included in the cinder codebase) 4. For a list of driver specific features, you can refer to the Cinder support matrix[1]. Hope that helps. [1] https://docs.openstack.org/cinder/latest/reference/support-matrix.html Thanks Rajat Dhasmana On Fri, Dec 22, 2023 at 8:19 PM Jorge Visentini <jorgevisentini@gmail.com> wrote:
Hi all!
My scenario:
We currently use oVirt and are studying the possibility of migrating to OpenStack in the future. Today we use Storages Enterprise 3PAR and Dell PowerVault. I've even tried to integrate 3PAR, but it's too complex for me for now...
Stupid question... If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ?
Cheers! -- Att, Jorge Visentini +55 55 98432-9868
I meant by HA, it's HA of instances... if a host goes down, it will automatically rise to another host. Sorry my english kkkkk Em sex., 22 de dez. de 2023 às 13:42, Rajat Dhasmana <rdhasman@redhat.com> escreveu:
Hi Jorge,
With the LVM driver: 1. Live migration should be supported although nova team can confirm 2. I don't know what you mean by HA instances (maybe it's a nova specific thing which again they can confirm) 3. Volume Snapshot is supported with the LVM driver (and it is one of the core features required for a driver to be included in the cinder codebase) 4. For a list of driver specific features, you can refer to the Cinder support matrix[1].
Hope that helps.
[1] https://docs.openstack.org/cinder/latest/reference/support-matrix.html
Thanks Rajat Dhasmana
On Fri, Dec 22, 2023 at 8:19 PM Jorge Visentini <jorgevisentini@gmail.com> wrote:
Hi all!
My scenario:
We currently use oVirt and are studying the possibility of migrating to OpenStack in the future. Today we use Storages Enterprise 3PAR and Dell PowerVault. I've even tried to integrate 3PAR, but it's too complex for me for now...
Stupid question... If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ?
Cheers! -- Att, Jorge Visentini +55 55 98432-9868
-- Att, Jorge Visentini +55 55 98432-9868
The link you provided helps me a lot. Thank you very much for the prompt response! Em sex., 22 de dez. de 2023 às 14:04, Jorge Visentini < jorgevisentini@gmail.com> escreveu:
I meant by HA, it's HA of instances... if a host goes down, it will automatically rise to another host.
Sorry my english kkkkk
Em sex., 22 de dez. de 2023 às 13:42, Rajat Dhasmana <rdhasman@redhat.com> escreveu:
Hi Jorge,
With the LVM driver: 1. Live migration should be supported although nova team can confirm 2. I don't know what you mean by HA instances (maybe it's a nova specific thing which again they can confirm) 3. Volume Snapshot is supported with the LVM driver (and it is one of the core features required for a driver to be included in the cinder codebase) 4. For a list of driver specific features, you can refer to the Cinder support matrix[1].
Hope that helps.
[1] https://docs.openstack.org/cinder/latest/reference/support-matrix.html
Thanks Rajat Dhasmana
On Fri, Dec 22, 2023 at 8:19 PM Jorge Visentini <jorgevisentini@gmail.com> wrote:
Hi all!
My scenario:
We currently use oVirt and are studying the possibility of migrating to OpenStack in the future. Today we use Storages Enterprise 3PAR and Dell PowerVault. I've even tried to integrate 3PAR, but it's too complex for me for now...
Stupid question... If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ?
Cheers! -- Att, Jorge Visentini +55 55 98432-9868
-- Att, Jorge Visentini +55 55 98432-9868
-- Att, Jorge Visentini +55 55 98432-9868
On Fri, 2023-12-22 at 14:05 -0300, Jorge Visentini wrote:
The link you provided helps me a lot.
Thank you very much for the prompt response! repsonces inline
Em sex., 22 de dez. de 2023 às 14:04, Jorge Visentini < jorgevisentini@gmail.com> escreveu:
I meant by HA, it's HA of instances... if a host goes down, it will automatically rise to another host.
Sorry my english kkkkk
Em sex., 22 de dez. de 2023 às 13:42, Rajat Dhasmana <rdhasman@redhat.com> escreveu:
Hi Jorge,
With the LVM driver: 1. Live migration should be supported although nova team can confirm yes it is we lvm it in our default live migration job 2. I don't know what you mean by HA instances (maybe it's a nova specific thing which again they can confirm) no this is an ovirt or enterprise virt feature which tripleo also supprot that was reject by the nova project.
instance ha uses pacemeaker similar external monitoring solution to automaticaly evacuate or migrate instnace when the hyperisior hosting the vm has a detectable issue like a failure of a disk in a raid array ro power supply. with the discontinuation of tripleo using a service masakari https://docs.openstack.org/masakari/latest/ is generally what i recommend for people not using redhat openstack.(redhat openstack does not productise masakari) if you are using upstream openstack or a distobution where its aviabel however it is worth evaulating for this usecase.
3. Volume Snapshot is supported with the LVM driver (and it is one of the core features required for a driver to be included in the cinder codebase) 4. For a list of driver specific features, you can refer to the Cinder support matrix[1].
Hope that helps.
[1] https://docs.openstack.org/cinder/latest/reference/support-matrix.html
Thanks Rajat Dhasmana
On Fri, Dec 22, 2023 at 8:19 PM Jorge Visentini <jorgevisentini@gmail.com> wrote:
Hi all!
My scenario:
We currently use oVirt and are studying the possibility of migrating to OpenStack in the future. Today we use Storages Enterprise 3PAR and Dell PowerVault. I've even tried to integrate 3PAR, but it's too complex for me for now...
Stupid question... If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ? if i understand correctly what you are descibign is host mounting a san lun on each of your sotrage nodes and then using the LVM driver to partion that lun and rexport it form the cidner volume storage nodes?
that could work but i suspect the performance would suffer unless you have the a very high bandwith connetion to the san or lun. one option to consider is isntead of using cinder volumes with lvm driver is to just confiugre nova to use local stoarge and mount the nova state dir on a lun form the san i.e. on all host map /mnt/remote_stroage to the same lun. then configure nova state dir to /mnt/remote_storage/$hostname/nova instead of /var/lib/nova that will mean that the local storage is treated as shared storage (you can do the same thing with an nfs share instead of a lun) so nova will know it does not need to copy the vm data when live migarting or cold migrating it also means on an evacuate the vm data will not be lost. normally if you use local stoarge with nova and you evecaute the root disk is recreatred form the glance image on the new host. if you have dell powervalut or 3par you woudl likely be best configuring cinder to use there dedicated drivers instead of using lvm as the lvm driver does not currently supprot managing volumes in a ha manner. i.e. if the cinder volume process that managed the lvm volume is offlien its not possibel to deploy a second instnace in an acitve/passive ro active active manner to in yoru config as the storage would be form a san it would be possibel to have active/passive or active acitve lvm but that would requrie code development in the lvm driver that is not currently planned. basicaly usign somethign like tooz to implement a distibuted lock managere/leader election. if you use the vendor specific storage driver hwoever they should already allow active/active or active passive toplogies wehre you can have multiple isntance of cinder volume managing the same storage array.
Cheers! -- Att, Jorge Visentini +55 55 98432-9868
-- Att, Jorge Visentini +55 55 98432-9868
Excellent explanation, smooney I understand perfectly. I don't really like using local storage, especially when it is mounted in a directory... I have terrible experience with cluster file systems, such as OCFS lol So, of course, the best scenario is to use the vendor driver... Well... I'll see what I do in my tests here. Thank you very much for the explanation!! Em sex., 22 de dez. de 2023 às 14:34, <smooney@redhat.com> escreveu:
The link you provided helps me a lot.
Thank you very much for the prompt response! repsonces inline
Em sex., 22 de dez. de 2023 às 14:04, Jorge Visentini < jorgevisentini@gmail.com> escreveu:
I meant by HA, it's HA of instances... if a host goes down, it will automatically rise to another host.
Sorry my english kkkkk
Em sex., 22 de dez. de 2023 às 13:42, Rajat Dhasmana < rdhasman@redhat.com> escreveu:
Hi Jorge,
With the LVM driver: 1. Live migration should be supported although nova team can confirm yes it is we lvm it in our default live migration job 2. I don't know what you mean by HA instances (maybe it's a nova specific thing which again they can confirm) no this is an ovirt or enterprise virt feature which tripleo also supprot
On Fri, 2023-12-22 at 14:05 -0300, Jorge Visentini wrote: that was reject by the nova project.
instance ha uses pacemeaker similar external monitoring solution to automaticaly evacuate or migrate instnace when the hyperisior hosting the vm has a detectable issue like a failure of a disk in a raid array ro power supply.
with the discontinuation of tripleo using a service masakari https://docs.openstack.org/masakari/latest/ is generally what i recommend for people not using redhat openstack.(redhat openstack does not productise masakari) if you are using upstream openstack or a distobution where its aviabel however it is worth evaulating for this usecase.
3. Volume Snapshot is supported with the LVM driver (and it is one of the core features required for a driver to be included in the cinder codebase) 4. For a list of driver specific features, you can refer to the Cinder support matrix[1].
Hope that helps.
[1]
https://docs.openstack.org/cinder/latest/reference/support-matrix.html
Thanks Rajat Dhasmana
On Fri, Dec 22, 2023 at 8:19 PM Jorge Visentini <
jorgevisentini@gmail.com>
wrote:
Hi all!
My scenario:
We currently use oVirt and are studying the possibility of migrating to OpenStack in the future. Today we use Storages Enterprise 3PAR and Dell PowerVault. I've even tried to integrate 3PAR, but it's too complex for me for now...
Stupid question... If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ? if i understand correctly what you are descibign is host mounting a san lun on each of your sotrage nodes and then using the LVM driver to partion that lun and rexport it form the cidner volume storage nodes?
that could work but i suspect the performance would suffer unless you have the a very high bandwith connetion to the san or lun.
one option to consider is isntead of using cinder volumes with lvm driver is to just confiugre nova to use local stoarge and mount the nova state dir on a lun form the san
i.e. on all host map /mnt/remote_stroage to the same lun. then configure nova state dir to /mnt/remote_storage/$hostname/nova instead of /var/lib/nova
that will mean that the local storage is treated as shared storage (you can do the same thing with an nfs share instead of a lun) so nova will know it does not need to copy the vm data when live migarting or cold migrating it also means on an evacuate the vm data will not be lost.
normally if you use local stoarge with nova and you evecaute the root disk is recreatred form the glance image on the new host.
if you have dell powervalut or 3par you woudl likely be best configuring cinder to use there dedicated drivers instead of using lvm as the lvm driver does not currently supprot managing volumes in a ha manner. i.e. if the cinder volume process that managed the lvm volume is offlien its not possibel to deploy a second instnace in an acitve/passive ro active active manner to
in yoru config as the storage would be form a san it would be possibel to have active/passive or active acitve lvm but that would requrie code development in the lvm driver that is not currently planned. basicaly usign somethign like tooz to implement a distibuted lock managere/leader election.
if you use the vendor specific storage driver hwoever they should already allow active/active or active passive toplogies wehre you can have multiple isntance of cinder volume managing the same storage array.
Cheers! -- Att, Jorge Visentini +55 55 98432-9868
-- Att, Jorge Visentini +55 55 98432-9868
-- Att, Jorge Visentini +55 55 98432-9868
On Fri, 22 Dec 2023 15:09:32 -0300 Jorge Visentini <jorgevisentini@gmail.com> wrote:
So, of course, the best scenario is to use the vendor driver...
Note that your guests exchange the data with 3PAR without any "vendor driver". You can export volume from the array with iscsi, or over FC. The reason why Cinder uses a "driver" is to manage your LUNs without LVM. When a request comes to allocate a volume, Cinder then invokes a driver (such as 3PAR) in order to talk to the array and obtain the LUN, which can be attached to your guest by Nova. With that in mind, this looks like a terrible idea: " If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ?" How do you ensure that your hypervisors have the LVM information consistent inside the LUNs you sliced up? My suggestion is to use one LUN per one OpenStack volume and not try to subdivide them with LVM. Your guests can use LVM on volumes that are attached to them, of course. But that has nothing to do with Cinder driver. -- Pete
Oi Jorge! I've worked on the integration between cinder and 3par in the past, and mostly the issues are from the 3par side/reconf (chap) Can you initialize a separate thread so we can also help you with that option? I believe it'll be the best for you in the long run. What OSP version are you using? And what deployment strategy? Cheers! --- Alvaro Soto. Note: My work hours may not be your work hours. Please do not feel the need to respond during a time that is not convenient for you. ---------------------------------------------------------- Great people talk about ideas, ordinary people talk about things, small people talk... about other people. On Sat, Dec 23, 2023, 1:31 AM Pete Zaitcev <zaitcev@redhat.com> wrote:
On Fri, 22 Dec 2023 15:09:32 -0300 Jorge Visentini <jorgevisentini@gmail.com> wrote:
So, of course, the best scenario is to use the vendor driver...
Note that your guests exchange the data with 3PAR without any "vendor driver". You can export volume from the array with iscsi, or over FC. The reason why Cinder uses a "driver" is to manage your LUNs without LVM. When a request comes to allocate a volume, Cinder then invokes a driver (such as 3PAR) in order to talk to the array and obtain the LUN, which can be attached to your guest by Nova.
With that in mind, this looks like a terrible idea:
" If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ?"
How do you ensure that your hypervisors have the LVM information consistent inside the LUNs you sliced up?
My suggestion is to use one LUN per one OpenStack volume and not try to subdivide them with LVM.
Your guests can use LVM on volumes that are attached to them, of course. But that has nothing to do with Cinder driver.
-- Pete
Hi Alvaro! So, my tests were very basic... I just tested LUN creation, delivery, and removal. Yes, I've read about some integration problems. I have now started testing with version 2023.2 of OSP. As soon as I create a multi node environment, I will open a new thread for you to help me. Thank you very much for the tips! Em sáb., 23 de dez. de 2023 às 07:20, Alvaro Soto <alsotoes@gmail.com> escreveu:
Oi Jorge! I've worked on the integration between cinder and 3par in the past, and mostly the issues are from the 3par side/reconf (chap)
Can you initialize a separate thread so we can also help you with that option? I believe it'll be the best for you in the long run.
What OSP version are you using? And what deployment strategy?
Cheers!
--- Alvaro Soto.
Note: My work hours may not be your work hours. Please do not feel the need to respond during a time that is not convenient for you. ---------------------------------------------------------- Great people talk about ideas, ordinary people talk about things, small people talk... about other people.
On Sat, Dec 23, 2023, 1:31 AM Pete Zaitcev <zaitcev@redhat.com> wrote:
On Fri, 22 Dec 2023 15:09:32 -0300 Jorge Visentini <jorgevisentini@gmail.com> wrote:
So, of course, the best scenario is to use the vendor driver...
Note that your guests exchange the data with 3PAR without any "vendor driver". You can export volume from the array with iscsi, or over FC. The reason why Cinder uses a "driver" is to manage your LUNs without LVM. When a request comes to allocate a volume, Cinder then invokes a driver (such as 3PAR) in order to talk to the array and obtain the LUN, which can be attached to your guest by Nova.
With that in mind, this looks like a terrible idea:
" If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ?"
How do you ensure that your hypervisors have the LVM information consistent inside the LUNs you sliced up?
My suggestion is to use one LUN per one OpenStack volume and not try to subdivide them with LVM.
Your guests can use LVM on volumes that are attached to them, of course. But that has nothing to do with Cinder driver.
-- Pete
-- Att, Jorge Visentini +55 55 98432-9868
Hi Pete! Note that your guests exchange the data with 3PAR without any "vendor driver". You can export volume from the array with iscsi, or over FC. The reason why Cinder uses a "driver" is to manage your LUNs without LVM. When a request comes to allocate a volume, Cinder then invokes a driver (such as 3PAR) in order to talk to the array and obtain the LUN, which can be attached to your guest by Nova. *- I didn't know that it is possible to work with 3PAR without the "vendor driver". My only test was with the driver, doing exactly what you said: Manage your LUNs without LVM.* How do you ensure that your hypervisors have the LVM information consistent inside the LUNs you sliced up? *- Yeah... I'm used to oVirt, so oVirt has a system that controls the consistency of LVM information. I couldn't say about OpenStack. But it's probably not the right way to work. I mean, it wouldn't be a good practice.* My suggestion is to use one LUN per one OpenStack volume and not try to subdivide them with LVM. Your guests can use LVM on volumes that are attached to them, of course. But that has nothing to do with Cinder driver. *- Yes, I understood. That's exactly what I want to do.* *But I keep thinking... one LUN per volume... if an instance has four volumes, then I will have four LUNs... I don't know if it's a bit exaggerated, talking about, for example, 300 volumes haha. But I understand your logic.It really clarified me a lot!* Many thanks, Pete! Em sáb., 23 de dez. de 2023 às 04:24, Pete Zaitcev <zaitcev@redhat.com> escreveu:
On Fri, 22 Dec 2023 15:09:32 -0300 Jorge Visentini <jorgevisentini@gmail.com> wrote:
So, of course, the best scenario is to use the vendor driver...
Note that your guests exchange the data with 3PAR without any "vendor driver". You can export volume from the array with iscsi, or over FC. The reason why Cinder uses a "driver" is to manage your LUNs without LVM. When a request comes to allocate a volume, Cinder then invokes a driver (such as 3PAR) in order to talk to the array and obtain the LUN, which can be attached to your guest by Nova.
With that in mind, this looks like a terrible idea:
" If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ?"
How do you ensure that your hypervisors have the LVM information consistent inside the LUNs you sliced up?
My suggestion is to use one LUN per one OpenStack volume and not try to subdivide them with LVM.
Your guests can use LVM on volumes that are attached to them, of course. But that has nothing to do with Cinder driver.
-- Pete
-- Att, Jorge Visentini +55 55 98432-9868
On 22/12, Jorge Visentini wrote:
Hi all!
My scenario:
We currently use oVirt and are studying the possibility of migrating to OpenStack in the future. Today we use Storages Enterprise 3PAR and Dell PowerVault. I've even tried to integrate 3PAR, but it's too complex for me for now...
Hi, Using the Cinder drivers would be the best idea when working with OpenStack. I don't know what deployment tool you are using, some deploy OpenStack services directly on the host and others in containers, but it should be possible to get both working even when external dependencies are needed. For example, for the 3PAR driver you'll need to have installed the "python-3parclient" Python package (inside the container image if you are running containerized), and then you'll only need to configure it following their documentation [1]. I've encountered some issues when running the current 3PAR driver code against a very old 3PAR array, so in my dev environment (that I use to write os-brick FC code) I usually run a modified version of the Cinder driver [2][3][4] with a modified version of the python-3parclient [5][6]. Regards, Gorka. [1]: https://docs.openstack.org/cinder/latest/configuration/block-storage/drivers... [2]: https://review.opendev.org/c/openstack/cinder/+/756709 [3]: https://review.opendev.org/c/openstack/cinder/+/756710/9 [4]: https://review.opendev.org/c/openstack/cinder/+/756711/13 [5]: https://github.com/hpe-storage/python-3parclient/pull/77 [6]: https://github.com/hpe-storage/python-3parclient/pull/79
Stupid question... If I work directly with LVM, that is, deliver a LUN to all my hosts, create an LVM and deliver it to Cinter, will I have the Live Migration, HA instances, snapshot and other features ?
Cheers! -- Att, Jorge Visentini +55 55 98432-9868
participants (6)
-
Alvaro Soto
-
Gorka Eguileor
-
Jorge Visentini
-
Pete Zaitcev
-
Rajat Dhasmana
-
smooney＠redhat.com