[Thinlinc-technical] HA++

[Peter Astrand]

Hi, comments below:

>    Currently, we have a centralized FreeNX scenario with 2 frontend servers 
> and 29 backend servers, supporting +500 users, all running in the LAN, on old 
> server hardware.
>    This scenario has been running for 4 years. We created message queuing 
> and remote tasks to alleviate our stupid day to day sysadmin jobs, ...
>     More info, on this 2 year old presentation (sorry, portuguese)
>
> http://www.slideshare.net/rui_lapa/desktop-linux-na-tranquilidade-portolinux-presentation?qid=a331b942-8a98-4ae9-aea0-c02698e698b4&v=default&b=&from_search=1
>
>    At the moment, we are evolving the solution to "crazy", so we want to 
> provide a desktop to external "entities" and shops without internal network 
> access, while still providing internal linux desktops.
>    The crazy is, that the external desktops servers will run on a cloud 
> provider.
>    In the future, we plan to ONLY use cloud servers, while keeping the 2 
> vsmservers on the DMZ.

I agree that this is crazy :-) As I mentioned, the VSM servers should be 
located where the VSM agents are. There are several reasons for this, 
security being one of them.


>     At this time we are doing our final test, where we have 2 DMZ 
> vsmservers, 3 LAN agents and 2 Cloud agents.
>
>    The datacenter has 4 dedicated internet circuits, with 1 IP each, in a 
> in/output load balancing scenario. (F5 BigIp).
>
>    The cloud agents are getting 4 (tcp/904) pollings from each external ip 
> of each circuit.
>    Due to this, we configured cloud agents "/vsmagent/allowed_clients" 
> mapped to the 4 external NAT IP's.
>    Unfortunately, even though, we get the tcpdump from each IP to TCP/904 
> and see the replies, the vsmservers webadmin status load page keeps 
> considering these servers one cycle up, the next up/down, randomly.
>    The internal vsmserver have the "/vsmserver/terminalservers", with the 
> external cloud ip's and the internal lan ips.
>
>    But, is there anything more we need to do?

The web interface communicate with the VSM Server, so apparently the VSM 
server(s) loses contact with the cloud agents now and then. It's difficult 
to tell why. You could start with one single VSM server, ie disable HA, 
until the agent communication works reliably. Also, check out 
vsmserver.log. tcpdump/Wireshark can probably also tell you what's going 
on.

Regards, 
---
Peter Astrand		ThinLinc Chief Developer
Cendio AB		http://cendio.com
Teknikringen 8		http://twitter.com/ThinLinc
583 30 Linkoping	http://facebook.com/ThinLinc
Phone: +46-13-214600	http://google.com/+CendioThinLinc