Re: [Xen-devel] [PATCH v10 5/5] libxl: network buffering cmdline switch

[Shriram Rajagopalan <rshriram@xxxxxxxxx>]

On Fri, Jun 6, 2014 at 1:34 AM, Hongyang Yang <yanghy@xxxxxxxxxxxxxx> wrote:

On 06/06/2014 01:30 AM, Ian Jackson wrote:

Yang Hongyang writes ("[PATCH v10 5/5] libxl: network buffering cmdline switch"):

Command line switch to 'xl remus' command, to enable network buffering.
Pass on this flag to libxl so that it can act accordingly.

You provide a global option to control the script, but no per-domain
config option. ÂWhy ?

There is a per-domain option to provide custom netbuffer scripts. "-N"

Â

A similar question arises about the network buffering boolean.

Wouldn't it be better if these were options on the devices, in the
domain configuration ?

Do you mean we make "-n -N" options into domain configuration?
I think these options are only related to remus and may not be used that
often because we provided a default network script which would be suitable
for most cases. these options are sort of second choices for users, may not
worth to be set in the domain configuration.

Feel free to tell me I'm wrong and it is better this way, if that's
true - just explain it.

+ Â Â * TODO: Split-Brain check.

What are your plans for the split brain check ?

For the moment, a DRBD backed VM will have much less chance of having a split-brain

than non-drbd cases. DRBD has built in split-brain resolution and is capable of interfacing

with a wide variety of external subsystems like corosync etc.

Split-Brain is a complicated issue. Heartbeats will work on a LAN and avoid spurious timeouts

based failover. ÂHowever, on a larger network where there are several elements in between the

primary and backup, you can't really know whether the primary died or the link died. You will then

need quorum, reachability based failover, etc. All of this is policy driven. So its best left to the user.

Â

My plan was to add another option that would allow the user to provide her own

scripts for checking the liveliness of the remote host. Depending on the script's

return value, remus checkpoints may continue/terminate & promote backup->primary, etc.

It's hard to do the split brain check under current implementation because
there's only one remus connection between the two domain. We may need to add
a heardbeat module to do this.

Thanks,
Ian.
.

--
Thanks,
Yang.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel