>- see footer for list info -<
update ---

The cluster nodes can now see each other. This was achieved by making the
LAN port the primary interface. To do this in Windows 2003: open Network
Connections (Start menu or control panel). Go to advanced menu > advanced
settings. You can reorder connections here. 
.  MS issue: I don't know why this is different from changing the interface
metrics. After changing the primary to the LAN interface, ipconfig still
lists the DMZ first!
.  Adobe issue: Why does Cold Fusion / JRUN clustering need to have its
interface primary? surely it should just use whichever interface is in the
right subnet? NLB / IIS doesn't have this requirement; NLB continued to
function after I'd made the DMZ interface secondary with no reconfiguration
whatsoever. Please get this fixed in the next versions of CF / JRUN.

Remaining problems - session replication
Sessions only seem to replicate when a cluster node is stopped. We have some
code which outputs a list of sessions currently running. We're using sticky
sessions on both NLB and CF clustering. 
.  A session appearing on one node will not appear on the other until we
stop the node to test failover.
.  Not all sessions failover immediately. There can be a 3-5 second gap
between each session appearing on the remaining node
.  Not all the sessions successfully failover - even with only a handful of
sessions
I conclude that session replication doesn't work properly. I think we'll go
back to wddx-ing our complex datatypes and then sticking them into client
variables held on a db....

Remaining problems - file not found errors
.  Occasionally we get a file not found error from CF for a file we know
exists (and the other node can see). We keep templates on a file server. The
CF instances run under a domain account that has permission to acces the
network share where the templates are held
.  Really weird - sometimes a 403.1 from IIS.
.  Recursion too deep; the stack overflowed - from IIS. This happened
occasionally until we followed the tips here: www.macromedia.com/go/950218dc
- this KB needs to be updated to state it applies to CF7 too.
.  We found that it is impossible to create new instances of CF if CF/Jrun
is running under a domain account, You have to have it running under local
system, create your instances (as windows services) and then change the
account they run under

Personally I think that under at least with JRUN providing the J2EE layer
for CF that clustering is too problematic. We would have been better off
with single instances of CF tied to IIS which was clustered using NLB, and
using client variables for session data that needs to be shared across
nodes. This has worked well for me since CF4.5, using both Cisco Local
Director and Windows NLB. 

Hopefully clustering will be improved with JRUN 5.

Alternatively, on the windows platform, Adobe should borrow a few ideas from
New Atlanta and run CF as a J# app on top of .NET, which would allow the use
of asp.net session replication.

-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of Aitor Ibarra
Sent: Sun 07 May 2006 11:22
To: [email protected]
Subject: [CF-Dev] Clustering problem - "network error" in Instance
Managerwith Microsoft NLB

>- see footer for list info -<
I've also posted this on the Adobe forums:
http://www.adobe.com/cfusion/webforums/forum/messageview.cfm?catid=143&threa
did=1150566&enterthread=y#4119603

Our 2 node cluster isn't working properly. The instances can't see each
other properly (they show "Network Error" in CF Admin> Enterprise> Instance
Manager). The upshot of this is that sessions are not being shared / failed
over.

Background...

- each server has two network interfaces. One of these is used for NLB, the
other for all other network traffic (including the CF cluster). The two NICs
are connected to different IP subnets (NLB is 10.0.0.x , LAN is
192.168.2.x). These are also seperate ethernets. I think the 'primary'
interface is the NLB one; i.e. it is the one listed first in ipconfig.
- CF 7 Enterprise (multi instance) installed on both servers (which are
running Windows 2003 Standard).
- New instance created on each server and bound to all IIS sites
- Instances registered in CF Admin, cluster created
- Windows NLB set up. We had some problems with this that were probably down
to our router. Now working fine in both Unicast and Multicast modes, but we
have kept on Unicast.
- Now the CF instances can't see each other.... so can't share session
variables
- the links to CF admin for each instance work, however, so it's not a basic
networking issue like not being able to ping/browse each other

I have tried adding the servers as unicast nodes in jrun.xml for each
instance as suggested here: http://www.adobe.com/go/26e04a0d 

I have also manually edited security.properties on both servers as the
symptoms are the same as described here: 
http://www.adobe.com/cfusion/knowledgebase/index.cfm?id=f23aa5ca

The only thing I haven't tried is swapping interfaces so that the LAN
interface is "primary". Does this really matter? Can this be done without
swapping cables by changing interface metrics in windows? Is it possible
that CF is trying to communicate through the NLB interface?


_______________________________________________

For details on ALL mailing lists and for joining or leaving lists, go to
http://list.cfdeveloper.co.uk/mailman/listinfo

--
CFDeveloper Sponsors:-
>- Hosting provided by www.cfmxhosting.co.uk -<
>- Forum provided by www.fusetalk.com -<
>- DHTML Menus provided by www.APYCOM.com -<
>- Lists hosted by www.Gradwell.com -<
>- CFdeveloper is run by Russ Michaels, feel free to volunteer your help -<


_______________________________________________

For details on ALL mailing lists and for joining or leaving lists, go to 
http://list.cfdeveloper.co.uk/mailman/listinfo

--
CFDeveloper Sponsors:-
>- Hosting provided by www.cfmxhosting.co.uk -<
>- Forum provided by www.fusetalk.com -<
>- DHTML Menus provided by www.APYCOM.com -<
>- Lists hosted by www.Gradwell.com -<
>- CFdeveloper is run by Russ Michaels, feel free to volunteer your help -<

Reply via email to