Re: [MooseFS-users] about moosefs 2.0

[Davies L. <dav...@gm...>]

On Thu, Nov 6, 2014 at 9:14 AM, Jakub Kruszona-Zawadzki
<ac...@mo...> wrote:
>
> On 06 Nov 2014, at 13:36, Adam K. Dean <ad...@dm...> wrote:
>
>> Hi,
>>
>>> It's BIG downgrade. LizardFS 2.5.2 is similar to MooseFS 1.6.27. MooseFS 2.0.X is far better, so changing system from Moose 2.0 to Lizard is like changing Moose 2.0 to Moose 1.6. Rather bad idea.
>>
>> With all due respect, the benefits of using LizardFS over MooseFS outweigh any perceived negatives, and yours would hardly be an objective viewpoint.
>
> What benefits? Sorry but can't see any.
>
>>
>> You have changed your business model, and while I cannot fault you for doing this as it is what you wanted to do, and it is your right to do so, it means that you have alienated people that otherwise would have stayed with MooseFS, myself included.
>>
>> Simpler Master-Master failover alone is a good enough reason to go to LizardFS.
>
> What master failover? Did you really test it? They just duct-taped mfsmetalogger, mfsmetarestore and mfsmaster together and named it "master-backup". Even in their CGI interface you see master-backups connected as "metaloggers". After master failure you need to manually log into master-backup, change it's "personality" and reload it. You also need to change your IP number, to tell your chunkservers and clients to reconnect. You can do the same in MooseFS CE using mfsmetalogger. After master failure you just have to stop metalogger and run master: "./mfsmetalogger ; ./mfsmaster -a", and as in Lizard of course change IP number. No difference in functionality - only in name of process.
>
> In MooseFS PRO we have master-ha - this is absolutely different thing. After master-leader failure one of master-followers automatically became leader. All clients and chunkservers reconnects also automatically to new master. Usually whole process takes about 5-10 seconds and is fully automatic. This is master-ha. And this is the feature we have only in MooseFS PRO version.

I had a patch to do similar things (master ha), which uses Raft
between active masters, and do automatic failure over in client and
chunk server, but I did not figure out how to publish it. The fork
approach is not good idea (it will make the community splitted even
more) , also I had not enough time to maintain it.

MooseFS is great (better than HDFS/GlusterFS/Ceph), but it also has
some weakness, such as small and not so active community.

>>
>> Regards,
>> Adam
>>
>> ----- Original Message -----
>> From: "Jakub Kruszona-Zawadzki" <jak...@ge...>
>> To: "NH. Nam Neddy" <na...@nd...>
>> Cc: "Andreas Hirczy" <ah...@it...>, moo...@li...
>> Sent: Thursday, 6 November, 2014 5:35:46 AM
>> Subject: Re: [MooseFS-users] about moosefs 2.0
>>
>
> Regards,
> Jakub Kruszona-Zawadzki.
> ------------------------------------------------------------------------------
> _______________________________________________
> moosefs-users mailing list
> moo...@li...
> https://lists.sourceforge.net/lists/listinfo/moosefs-users



-- 
 - Davies