<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<div class="moz-cite-prefix">On 01/01/2018 10:10 AM, Yaniv Kaul
wrote:<br>
</div>
<blockquote type="cite"
cite="mid:CAJgorsY0bkoyp3fM2e0noMJQ-6W=_v1K-CP6b+SfrZr8o=qYhQ@mail.gmail.com">
<div dir="ltr"><br>
<div class="gmail_extra">
<div class="gmail_quote">On Mon, Jan 1, 2018 at 12:50 AM,
Andrei V <span dir="ltr"><<a
href="mailto:andreil1@starlett.lv" target="_blank"
moz-do-not-send="true">andreil1@starlett.lv</a>></span>
wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">Hi !<br>
<br>
I'm installing 2-node failover cluster (2 x Xeon servers
with local RAID<br>
5 / ext4 for oVirt storage domains).<br>
Now I have a dilemma - use either GlusterFS replica 2 or
stick with NFS?<br>
</blockquote>
<div><br>
</div>
<div>Replica 2 is not good enough, as it can leave you with
split brain. It's been discussed in the mailing list
several times.</div>
<div>How do you plan to achieve HA with NFS? With drbd?</div>
</div>
</div>
</div>
</blockquote>
Hi, Yaniv,<br>
Thanks a lot for detailed explanation!<br>
<br>
I know Replica 2 is not optimal solution. <br>
Right now I have only 2 servers with internal RAIDs for nodes, and
till end of this week system had to be running in whatever
condition.<br>
May be its better to use local storage domain on each node, set
export domain on backup node, and backup VMs to 2nd backup node in
timed interval? <br>
Its not highly-available yet workable solution.<br>
<br>
<blockquote type="cite"
cite="mid:CAJgorsY0bkoyp3fM2e0noMJQ-6W=_v1K-CP6b+SfrZr8o=qYhQ@mail.gmail.com">
<div dir="ltr">
<div class="gmail_extra">
<div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
4.2 Engine is running on separate hardware.<br>
</blockquote>
<div><br>
</div>
<div>Is the Engine also highly available?</div>
</div>
</div>
</div>
</blockquote>
<br>
Its KVM appliance, could be launched on 2 SuSE servers.<br>
<br>
<blockquote type="cite"
cite="mid:CAJgorsY0bkoyp3fM2e0noMJQ-6W=_v1K-CP6b+SfrZr8o=qYhQ@mail.gmail.com">
<div dir="ltr">
<div class="gmail_extra">
<div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
Each node have its own storage domain (on internal RAID).<br>
</blockquote>
<div><br>
</div>
<div>So some sort of replica 1 with geo-replication between
them?</div>
</div>
</div>
</div>
</blockquote>
<br>
Could it be the following?<br>
1) Local storage domain on each node<br>
2) GlusterFS geo-replication or over these directories? Not sure
this will work.<br>
<br>
<blockquote type="cite"
cite="mid:CAJgorsY0bkoyp3fM2e0noMJQ-6W=_v1K-CP6b+SfrZr8o=qYhQ@mail.gmail.com">
<div dir="ltr">
<div class="gmail_extra">
<div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
All VMs must be highly available.<br>
</blockquote>
<div><br>
</div>
<div>Without shared storage, it may be tricky.</div>
</div>
</div>
</div>
</blockquote>
<br>
Seems to be timely VM backup to 2nd node is enough for this time.<br>
With current hardware anything above is too cumbersome to setup.<br>
<br>
<blockquote type="cite"
cite="mid:CAJgorsY0bkoyp3fM2e0noMJQ-6W=_v1K-CP6b+SfrZr8o=qYhQ@mail.gmail.com">
<div dir="ltr">
<div class="gmail_extra">
<div class="gmail_quote">
<div><br>
</div>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
One of the VMs is an accounting/stock control system with
FireBird SQL<br>
server on CentOS is speed-critical.<br>
</blockquote>
<div><br>
</div>
<div>But is IO the bottleneck? Are you using SSDs / NVMe
drives? </div>
<div>I'm not familiar enough with FireBird SQL server - does
it have an application layer replication you might opt to
use?</div>
<div>In such case, you could pass-through a NVM disk and
have the application layer perform the replication between
the nodes.</div>
<div> </div>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
No load balancing between nodes necessary. 2nd is just for
backup if 1st<br>
for whatever reason goes up in smoke. All VM disks must be
replicated to<br>
backup node in near real-time or in worst case each 1 - 2
hour.<br>
GlusterFS solves this issue yet at high performance
penalty.<br>
</blockquote>
<div><br>
</div>
<div>The problem with a passive backup is that you never
know it'll really work when needed. This is why
active-active is many time preferred.</div>
<div>It's also more cost effective usually - instead of some
HW lying around.</div>
<div> </div>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
>From what I read here<br>
<a
href="http://lists.ovirt.org/pipermail/users/2017-July/083144.html"
rel="noreferrer" target="_blank" moz-do-not-send="true">http://lists.ovirt.org/<wbr>pipermail/users/2017-July/<wbr>083144.html</a><br>
GlusterFS performance with oVirt is not very good right
now because QEMU<br>
uses FUSE instead of libgfapi.<br>
<br>
What is optimal way to go on ?<br>
</blockquote>
<div><br>
</div>
<div>It's hard to answer without additional details.</div>
<div>Y.</div>
<div> </div>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
Thanks in advance.<br>
Andrei<br>
<br>
______________________________<wbr>_________________<br>
Users mailing list<br>
<a href="mailto:Users@ovirt.org" moz-do-not-send="true">Users@ovirt.org</a><br>
<a href="http://lists.ovirt.org/mailman/listinfo/users"
rel="noreferrer" target="_blank" moz-do-not-send="true">http://lists.ovirt.org/<wbr>mailman/listinfo/users</a><br>
</blockquote>
</div>
<br>
</div>
</div>
</blockquote>
<p><br>
</p>
</body>
</html>