<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
<title></title>
</head>
<body bgcolor="#ffffff" text="#000000">
Jeff Fisher wrote:
<blockquote cite="mid43F5EF9E.6070902@lfchosting.com" type="cite">
<pre wrap="">Pierre Ancelot wrote:
</pre>
<blockquote type="cite">
<pre wrap="">I'm afraid i have to change my system, i rnu drbd/NFS/heartbeat on a
linux 2.6.
It HANGS if i ever do anything that overloads it a little and it keeps
on hanging bad, all my cluster becomes unavailable... (too bad for an
high availability cluster).
On worst cases i have to reboot all my nodes.
I even think about having an SMB share so bad it goes...
Any idea what's going on ? maybe a config issue of NFS
Maybe, as i have read recently, NFS on linux sucks.
Any alternative viable ? i can't let this cluster run on doubts.
</pre>
</blockquote>
<pre wrap=""><!---->
When I tested samba + drbd + keepalived, if I pulled the plug on the
primary and caused a failover to the secondary -- there was massive data
corruption when using samba.
Does anyone use drbd + samba and have it handle failovers nicely?</pre>
</blockquote>
Imo this can't be done because Samba is stateful. When you<br>
replicate the NFS administration (/var/lib/nfs) on a drbd'ed<br>
partition you obtain HA read&writes on your NFS clients.<br>
When you apply that knowledge to your Samba serving<br>
problem you might want to add a translation cluster in-front<br>
of your drbd HA-NFS fileservers (SMB->NFS). Although you<br>
still have a point of failure you can extend the amount of<br>
storage space by huge amounts while spreading the risk and<br>
reducing the Samba administration.<br>
<br>
The node which handles the Samba service needs to be very<br>
stable (hotswap mirrored root disk / hotswap PSU's).<br>
<br>
Just a thought,<br>
Leroy<br>
</body>
</html>