Ocfs2 users - Apr 2020 - [Ocfs2-devel] mount existent ocfs2 volume in the local mode without the extra DLM, pacemaker/corosync layers

Hello Gang,

On 4/10/20 4:36 PM, Gang He wrote:
> Hello Guys,
> 
> Usually, we create a ocfs2 cluster on the top DLM(pacemaker/corosync) with
a shared volume.
> but the customers complained they could not mount existent ocfs2 volume in
the local mode without DLM/cluster envirnment in backup/restore scenario.
> in this case, the customers want to mount this ocfs2 volume from one node
without cluster aware way, sometimes the customers have stoped(uninstalled) the
whole cluster before the backup/restore, they just want to access the data from
the ocfs2 volume.
> Then, I'd like to add a mount option, e.g. "nocluster" (I
feel using "local" option is a little confused for the users, since we
have a local mount volume concept).
> If the users use "nocluster" option to mount a ocfs2 volume, the
whole mount will not depend on the DLM/cluster related services. they
> will mount the existent ocfs2 volume directly (like local mount) for
avoiding setup the cluster stacks.
> The code change is small with my preliminary investigation, the main
affected places are mount.ocfs2/mounted.ocfs2 tools and mount process in the
ocfs2 kernel modules.
Some quick questions:

1.Will the 'nocluster' ocfs2 volume be read-only?
2.How to achieve filesystem consistency if the whole cluster or some 
members of it crashed before this 'nocluster' mount.
3.Is there any mechanism guard protecting no other node do mount after 
this 'nocluster' mount.
4.Why do we have potential work for mounted.ocfs2?
5.I feel debugfs.ocfs2 can dump specified file or entire volume, is that 
satisfied for you?

Thanks,
Changwei

> 
> Any comments/suggestions?
> 
> 
> Thanks a lot
> Gang
> 
> _______________________________________________
> Ocfs2-devel mailing list
> Ocfs2-devel at oss.oracle.com
> https://oss.oracle.com/mailman/listinfo/ocfs2-devel
>

Hello Changwei,

See my comments inline.

________________________________________
From: Changwei Ge <chge at linux.alibaba.com>
Sent: Friday, April 10, 2020 5:08 PM
To: Gang He; ocfs2-devel at oss.oracle.com; ocfs2-users at oss.oracle.com
Subject: Re: [Ocfs2-devel] mount existent ocfs2 volume in the local mode without
the extra DLM, pacemaker/corosync layers

Hello Gang,

On 4/10/20 4:36 PM, Gang He wrote:
> Hello Guys,
>
> Usually, we create a ocfs2 cluster on the top DLM(pacemaker/corosync) with
a shared volume.
> but the customers complained they could not mount existent ocfs2 volume in
the local mode without DLM/cluster envirnment in backup/restore scenario.
> in this case, the customers want to mount this ocfs2 volume from one node
without cluster aware way, sometimes the customers have stoped(uninstalled) the
whole cluster before the backup/restore, they just want to access the data from
the ocfs2 volume.
> Then, I'd like to add a mount option, e.g. "nocluster" (I
feel using "local" option is a little confused for the users, since we
have a local mount volume concept).
> If the users use "nocluster" option to mount a ocfs2 volume, the
whole mount will not depend on the DLM/cluster related services. they
> will mount the existent ocfs2 volume directly (like local mount) for
avoiding setup the cluster stacks.
> The code change is small with my preliminary investigation, the main
affected places are mount.ocfs2/mounted.ocfs2 tools and mount process in the
ocfs2 kernel modules.
Some quick questions:

1.Will the 'nocluster' ocfs2 volume be read-only?
The user can read or write this mount (like local mount), but the user needs to
make sure only one node can mount
the volume like this way.

2.How to achieve filesystem consistency if the whole cluster or some
members of it crashed before this 'nocluster' mount.
The mount process will do the recovery for each node's journal, but the
mount process will not do DLM related things.

3.Is there any mechanism guard protecting no other node do mount after
this 'nocluster' mount.
Not, we can give a warning during the mount period.
If the customer need to understand this risk. 

4.Why do we have potential work for mounted.ocfs2?
Maybe we need to add some code, let mounted.ocfs2 aware this mount option.
In short, I am not 100% sure if there is something to do.

5.I feel debugfs.ocfs2 can dump specified file or entire volume, is that
satisfied for you?
For the developer, we can do like that way.
But for the customer, it is too hard.
and that, the customer wants to access lots of files for one time.

Thanks
Gang

Thanks,
Changwei

>
> Any comments/suggestions?
>
>
> Thanks a lot
> Gang
>
> _______________________________________________
> Ocfs2-devel mailing list
> Ocfs2-devel at oss.oracle.com
> https://oss.oracle.com/mailman/listinfo/ocfs2-devel
>

At least for class o2cb stack, I suppose if cluster was once set up 
successfully, it won't be hard to mount from a single node.
Or do you lose your cluster configuration somehow?

On 4/10/20 5:28 PM, Gang He wrote:
> Hello Changwei,
> 
> See my comments inline.
> 
> ________________________________________
> From: Changwei Ge <chge at linux.alibaba.com>
> Sent: Friday, April 10, 2020 5:08 PM
> To: Gang He; ocfs2-devel at oss.oracle.com; ocfs2-users at oss.oracle.com
> Subject: Re: [Ocfs2-devel] mount existent ocfs2 volume in the local mode
without the extra DLM, pacemaker/corosync layers
> 
> Hello Gang,
> 
> On 4/10/20 4:36 PM, Gang He wrote:
>> Hello Guys,
>>
>> Usually, we create a ocfs2 cluster on the top DLM(pacemaker/corosync)
with a shared volume.
>> but the customers complained they could not mount existent ocfs2 volume
in the local mode without DLM/cluster envirnment in backup/restore scenario.
>> in this case, the customers want to mount this ocfs2 volume from one
node without cluster aware way, sometimes the customers have stoped(uninstalled)
the whole cluster before the backup/restore, they just want to access the data
from the ocfs2 volume.
>> Then, I'd like to add a mount option, e.g. "nocluster" (I
feel using "local" option is a little confused for the users, since we
have a local mount volume concept).
>> If the users use "nocluster" option to mount a ocfs2 volume,
the whole mount will not depend on the DLM/cluster related services. they
>> will mount the existent ocfs2 volume directly (like local mount) for
avoiding setup the cluster stacks.
>> The code change is small with my preliminary investigation, the main
affected places are mount.ocfs2/mounted.ocfs2 tools and mount process in the
ocfs2 kernel modules.
> 
> Some quick questions:
> 
> 1.Will the 'nocluster' ocfs2 volume be read-only?
> The user can read or write this mount (like local mount), but the user
needs to make sure only one node can mount
> the volume like this way.
> 
> 2.How to achieve filesystem consistency if the whole cluster or some
> members of it crashed before this 'nocluster' mount.
> The mount process will do the recovery for each node's journal, but the
mount process will not do DLM related things.
> 
> 3.Is there any mechanism guard protecting no other node do mount after
> this 'nocluster' mount.
> Not, we can give a warning during the mount period.
> If the customer need to understand this risk.
> 
> 4.Why do we have potential work for mounted.ocfs2?
> Maybe we need to add some code, let mounted.ocfs2 aware this mount option.
> In short, I am not 100% sure if there is something to do.
> 
> 5.I feel debugfs.ocfs2 can dump specified file or entire volume, is that
> satisfied for you?
> For the developer, we can do like that way.
> But for the customer, it is too hard.
> and that, the customer wants to access lots of files for one time.
> 
> Thanks
> Gang
> 
> Thanks,
> Changwei
> 
> 
>>
>> Any comments/suggestions?
>>
>>
>> Thanks a lot
>> Gang
>>
>> _______________________________________________
>> Ocfs2-devel mailing list
>> Ocfs2-devel at oss.oracle.com
>> https://oss.oracle.com/mailman/listinfo/ocfs2-devel
>>