[linux-lvm] Mirroring implementation

[linux-lvm] Mirroring implementation

[brem belguebli]

[-- Attachment #1: Type: text/plain, Size: 939 bytes --]

Hi,
I have a few questions concerning LVM mirror implementation, and I couldn't
find any suitable design doc

I need to setup a dual site ( a few miles between the 2 sites, around 1 ms
RTD) mirror based cluster system.

-  what is the default behaviour of the mirroring, writing in parallel to
both mirror legs or serialized ? what about read requests ?

- for write requests, is the ack returned to the application (thru VFS) when
the 2 legs are updated or only when the first is updated ?
- Is there a way to control it ?

- what happens if the log device is lost ?

I saw a post about LVM mirror vs mdadm where it is about the location of the
mirror log (
https://www.redhat.com/archives/linux-lvm/2009-July/msg00012.html) someone
gave the answer that when creating the LV's with --alloc anywhere it will
auto locate the log on any of the PV's.

According to what I could read, the --alloc doesn't intend for that, am I
right ?

Brem ,

[-- Attachment #2: Type: text/html, Size: 2457 bytes --]

Re: [linux-lvm] Mirroring implementation

[malahal]

brem belguebli [brem.belguebli@gmail.com] wrote:
>    Hi,
>    I have a few questions concerning LVM mirror implementation, and I
>    couldn't find any suitable design doc
>    I need to setup a dual site ( a few miles between the 2 sites, around 1 ms
>    RTD) mirror based cluster system.

You need asynchronous mirroring (remote replication module). This is not
in mainline (not sure if included in the latest RHEL release). The
following comments apply to dm-mirror module that is already in the
mainline.

>    - *what is the default behaviour of the mirroring, writing in parallel to
>    both mirror legs or serialized ? what about read requests ?

It writes in parallel. Reads from a single device (switches on a device
failure only).

>    - for write requests, is the ack returned to the application (thru VFS)
>    when the 2 legs are updated or only when the first is updated ?

When two are updated. (you need 'remote replication' module for the
latter behaviour)

>    - Is there a way to control it ?

     I don't think so.

>    - what happens if the log device is lost ?

The mirror stops and a daemon re-allocates the log and the mirror should
be back to business (this depends on your lvm configuration though).

>    I saw a post about LVM mirror vs mdadm where it is about the location of
>    the mirror log
>    ([1]https://www.redhat.com/archives/linux-lvm/2009-July/msg00012.html)
>    someone gave the answer that when creating the LV's with --alloc anywhere
>    it will auto locate the log on any of the PV's.
>    According to what I could read, the --alloc doesn't intend for that, am I
>    right ?

Replace your "auto locate" to "allocate the log on any of the PV's".

What is the intention of the "--alloc anywhere" from your reading?

--Malahal.

Re: [linux-lvm] Mirroring implementation

[Lars Ellenberg]

On Mon, Jul 27, 2009 at 06:56:28AM -0700, malahal@us.ibm.com wrote:
> brem belguebli [brem.belguebli@gmail.com] wrote:
> >    Hi,
> >    I have a few questions concerning LVM mirror implementation, and I
> >    couldn't find any suitable design doc
> >    I need to setup a dual site ( a few miles between the 2 sites, around 1 ms
> >    RTD) mirror based cluster system.
> 
> You need asynchronous mirroring (remote replication module). This is not
> in mainline (not sure if included in the latest RHEL release). The
> following comments apply to dm-mirror module that is already in the
> mainline.

There is also MD RAID1, see man mdadm, the --write-mostly and --write-behind options.

And of course, DRBD http://www.drbd.org
which is not RAID1 turned into replication,
but a dedicated replication solution.

-- 
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com

Re: [linux-lvm] Mirroring implementation

[brem belguebli]

[-- Attachment #1: Type: text/plain, Size: 3097 bytes --]

Hi Malahal,


2009/7/27, malahal@us.ibm.com <malahal@us.ibm.com>:
>
> brem belguebli [brem.belguebli@gmail.com] wrote:
> >    Hi,
> >    I have a few questions concerning LVM mirror implementation, and I
> >    couldn't find any suitable design doc
> >    I need to setup a dual site ( a few miles between the 2 sites, around
> 1 ms
> >    RTD) mirror based cluster system.
>
> You need asynchronous mirroring (remote replication module). This is not
> in mainline (not sure if included in the latest RHEL release). The
> following comments apply to dm-mirror module that is already in the
> mainline.


Do you mean a Disk array based replication mechanism (SRDF, CA, etc...)? or
is it system based , and if so do you have any technical document link to
provide me ?


>    - *what is the default behaviour of the mirroring, writing in parallel
> to
> >    both mirror legs or serialized ? what about read requests ?
>
> It writes in parallel. Reads from a single device (switches on a device
> failure only).
>
> >    - for write requests, is the ack returned to the application (thru
> VFS)
> >    when the 2 legs are updated or only when the first is updated ?
>
> When two are updated. (you need 'remote replication' module for the
> latter behaviour)

As I said, I couldn't find suitable doc about this, and in fact that's the
behaviour I'm expecting (acking only when the 2 legs are up to date) . Can
you point me to the good doc ?


>    - Is there a way to control it ?
>
>     I don't think so.
>
> >    - what happens if the log device is lost ?
>
> The mirror stops and a daemon re-allocates the log and the mirror should
> be back to business (this depends on your lvm configuration though).


As the mirrorlog is in my case bound to a given SAN device (let's say
/dev/sdx) and this device belongs to the SAN array located on the datacenter
that experienced a disaster, my mirror won't be able to be brought back
online as long as /dev/sdx is down .

>    I saw a post about LVM mirror vs mdadm where it is about the location
> of
> >    the mirror log
> >    ([1]https://www.redhat.com/archives/linux-lvm/2009-July/msg00012.html
> )
> >    someone gave the answer that when creating the LV's with --alloc
> anywhere
> >    it will auto locate the log on any of the PV's.
> >    According to what I could read, the --alloc doesn't intend for that,
> am I
> >    right ?
>
> Replace your "auto locate" to "allocate the log on any of the PV's".
>
> What is the intention of the "--alloc anywhere" from your reading?


According to the lvm man page, --alloc anywhere, in crontrary to --alloc
strict (default) would allocate "data" LE to any pv forming the VG,
resulting in having striped or mirrored LE on the same PV (sic!).

It doesn't seem to concern log device which is de facto a 3rd device not
necessary belonging to the VG, or the man page is not up2date.


--Malahal.
>
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@redhat.com
> https://www.redhat.com/mailman/listinfo/linux-lvm
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
>

[-- Attachment #2: Type: text/html, Size: 4560 bytes --]

Re: [linux-lvm] Mirroring implementation

[Roy Sigurd Karlsbakk]

On 27. juli. 2009, at 15.56, malahal@us.ibm.com wrote:

> It writes in parallel. Reads from a single device (switches on a  
> device
> failure only).


Wouldn't it be better to read in parallel, from different places? I've  
heard some controllers can do this, so that the reading will be  
somehow like reading from a RAID-0 stripe.

roy
--
Roy Sigurd Karlsbakk
(+47) 97542685
roy@karlsbakk.net
http://blogg.karlsbakk.net/
--
I all pedagogikk er det essensielt at pensum presenteres  
intelligibelt. Det er et element�rt imperativ for alle pedagoger �  
unng� eksessiv anvendelse av idiomer med fremmed opprinnelse. I de  
fleste tilfeller eksisterer adekvate og relevante synonymer p� norsk.

Re: [linux-lvm] Mirroring implementation

[brem belguebli]

[-- Attachment #1: Type: text/plain, Size: 1407 bytes --]

Hi Roy,

I would be better indeed, but practically, even if it is supposed to request
both disks (san arrays) in //, the closest array  will almost always be the
first to answer unless it is overloaded.

What I'm looking for is the behaviour for writes, as I don't want to deal
with failure scenarios in which I could find outdated data on my last
remaining mirror leg.




2009/7/28, Roy Sigurd Karlsbakk <roy@karlsbakk.net>:
>
> On 27. juli. 2009, at 15.56, malahal@us.ibm.com wrote:
>
> It writes in parallel. Reads from a single device (switches on a device
>> failure only).
>>
>
>
> Wouldn't it be better to read in parallel, from different places? I've
> heard some controllers can do this, so that the reading will be somehow like
> reading from a RAID-0 stripe.
>
> roy
> --
> Roy Sigurd Karlsbakk
> (+47) 97542685
> roy@karlsbakk.net
> http://blogg.karlsbakk.net/
> --
> I all pedagogikk er det essensielt at pensum presenteres intelligibelt. Det
> er et elementært imperativ for alle pedagoger å unngå eksessiv anvendelse av
> idiomer med fremmed opprinnelse. I de fleste tilfeller eksisterer adekvate
> og relevante synonymer på norsk.
>
>
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@redhat.com
> https://www.redhat.com/mailman/listinfo/linux-lvm
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
>

[-- Attachment #2: Type: text/html, Size: 2658 bytes --]