3-0-21pre1 and ha-samba

[David]

Hi all,

Just wondering if the "rewrite of CIFS oplock implementation" in the recent
3.0.21pre1 release of samba means that samba can now truly be made
"highly-available", with respect to the state troubles I am used to with
current ha-samba setups (ie. having to restart a file transaction when samba
server fails during a read/write)? Or is this still not yet possible?

This link:

http://tr.samba.org/samba/docs/man/Samba-HOWTO-Collection/SambaHA.html#id261
8730

Is what I am going on; explains simply that samba cannot handle "sharing
state" (if I understand correctly) and thus cannot "soft fail" like nfs can;

however this guy:

http://lists.linux-ha.org/pipermail/linux-ha/2000-February/000938.html

claims that he has been able to achieve this (that is, a server failure
during a file transaction with client resume) simply by making the two samba
servers (the active and passive) take the same netbios name.

Does anybody have a working solution like this? I have a 2-node
active-passive "heartbeat" cluster from linux-ha which works in as much as
any current transactions on a samba client fail when the samba server fails
over, and can be restarted immediately after the other samba server takes
over. But I am trying to find a way to prevent them from failing. Can this
be done?

Regards,

David.