<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
</head>
<body bgcolor="#ffffff" text="#000000">
<small><font face="Helvetica, Arial, sans-serif"><small><font size="-1"><small><small><br>
Hello,<br>
<br>
I have 3 servers setup in following way:<br>
</small></small></font></small></font></small>
<ul>
<li><small><font face="Helvetica, Arial, sans-serif"><small>Two
servers (server1 and server2) are connected using DRBD, which is used
as volume for shared storage. This shared storage is formated using
OCFS2. Server1 has RAID10, while server2 has RAID1 under DRBD.<br>
</small></font></small></li>
<li><small><font face="Helvetica, Arial, sans-serif"><small>All three
servers are accessing shared storage using iSCSI target on server1.
Even when DRBD is in Primary/Primary mode, server1 and server2 must use
iSCSI, and not access DRBD directly, because they don't see server3 as
member of cluster, when it access it trough iSCSI. Must be because
iSCSI doesn't see changes happening directly on device below, which is
DRBD.<br>
</small></font></small></li>
<li><small><font face="Helvetica, Arial, sans-serif"><small>Shared
storage has about 1.7 milion files of about 250GB total size. Everyday
backup is made to remote server using rsync. This takes a lot of time
(few hours) because if many files and conflict with running web
application that is reading/write to shared volume while backup is in
progress.</small></font></small></li>
</ul>
<small><font face="Helvetica, Arial, sans-serif"><small> Hard drives
on server2 are mostly idle. They write block of data from time to
time.It would be more efficient for backup if it runs from OCFS2 copy
on DRBD of server2. I would mount it read-only. It would be possible to
create snapshot of LVM volume below DRBD, and to mount it read-only.<br>
<br>
Here finally comes my question. Do you think it's safe to mount OCFS2
in this way, since changes will happen from other nodes which will not
be in cluster with backup process? Read-only mount, if possible, would
prevent changes to go from backup process to live nodes. However, I
assume file system might become corrupted for backup side? Is there
flush method of some kind that I can initiate on all three servers or
iSCSI target, before making snapshot on server2 and mounting OCFS2 in
read-only and/or local mode?<br>
<br>
Does anyone have suggestion on how to make efficient backups of many
files? DRBD is nice backup feature, but corruption in file system is
distributed to both drives. Having separate file system copy is safe,
even when it's not real-time.<br>
<br>
Did anyone try using lscynd with OCFS2? Lsync2 monitors changes in
file system using inotify and transfers only changed folder to backup
server using rsync. I was reading that inotify works with delay with
OCFS2 when changes happen on remote node. However, few seconds delay is
not problem for me.<br>
<br>
Thanks,<br>
Nikola</small> </font></small><br>
<font size="-1"><font face="Helvetica, Arial, sans-serif"></font></font>
</body>
</html>