[pve-devel] [PATCH 2/6] qemu_drive_mirror : handle multiple jobs

Alexandre DERUMIER aderumier at odiso.com
Wed Dec 21 11:05:23 CET 2016


I have seen a lot of improvement with nbd in qemu 2.8, maybe the connection hang has been fixed.

I'll try to do tests.


----- Mail original -----
De: "aderumier" <aderumier at odiso.com>
À: "Wolfgang Bumiller" <w.bumiller at proxmox.com>
Cc: "pve-devel" <pve-devel at pve.proxmox.com>
Envoyé: Mercredi 21 Décembre 2016 10:51:55
Objet: Re: [pve-devel] [PATCH 2/6] qemu_drive_mirror : handle multiple jobs

>>IIRC that was the only blocker. 
>> 
>>Basically the patchset has to work *without* tcp_ping() since it is an 
>>unreliable check, and then we still have to catch failing connections 
>>_correctly_. (There's no point in knowing that "some time in the past 
>>you were able to connect to something which may or may not have been a 
>>qemu nbd server", we need to know whether the drive-mirror job itself 
>>was able to connect.) 

For me, the mirror job auto abort if connection is failing during the migration. Do you see another behaviour ? 

the tcp_ping was just before launching the drive mirror command, because it was hanging in this case. 


----- Mail original ----- 
De: "Wolfgang Bumiller" <w.bumiller at proxmox.com> 
À: "aderumier" <aderumier at odiso.com> 
Cc: "dietmar" <dietmar at proxmox.com>, "pve-devel" <pve-devel at pve.proxmox.com> 
Envoyé: Mercredi 21 Décembre 2016 10:47:48 
Objet: Re: [pve-devel] [PATCH 2/6] qemu_drive_mirror : handle multiple jobs 

On Mon, Dec 19, 2016 at 07:05:58AM +0100, Alexandre DERUMIER wrote: 
> >>AFAIR the patch looks already quite good. We currently prepare for 
> >>the 4.4 release, but when that is done we can start adding new features 
> >>like local disk live migration. 
> 
> Hi, 
> Now that 4.4 has been release, could it be possible to apply live storage migration patches ? 
> 
> I'll have some time until end of the year to polish them. 

So far they've been working, but iirc there's still the issue where 
breaking/failing connections _during_ the operation and/or the block 
jobs failing to establish the connection in the first place go unnoticed 
where we end up with possibly deleted disks instead of properly failing. 
(Which is why I said I didn't like the tcp_ping check.) 

IIRC that was the only blocker. 

Basically the patchset has to work *without* tcp_ping() since it is an 
unreliable check, and then we still have to catch failing connections 
_correctly_. (There's no point in knowing that "some time in the past 
you were able to connect to something which may or may not have been a 
qemu nbd server", we need to know whether the drive-mirror job itself 
was able to connect.) 

> Also, I would like to improve them later, to add another feature, live migration to remote external proxmox cluster. (on different proxmox && storage). 
> Vmware has this feature in entreprise premium, and I'll need it for some customers next year where I can't have downtime. 

Sounds like a piece of work ;-) 

_______________________________________________ 
pve-devel mailing list 
pve-devel at pve.proxmox.com 
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel 




More information about the pve-devel mailing list