Terracotta Discussion Forums (LEGACY READ-ONLY ARCHIVE)

Hi.

Is it possible to configure TC server listening to different network segment
according to usage like DSO or system management ?

I could find how to define port number for each purpose but not IP
address or host name.

Becouse of security policy of our customer, the network segment
for system management is separated from the segment for service.

The server we use has 2 or more NICs on it and each interface has
different IP address for different network segment. so that the server
faces to multiple network segments those are separated for purpose.

If it's NOT possible, is there any plan to let assgning different IP address
to each DSO, jmx-port and l2-group-port ?

Thank you.

See the "bind" attribute under /tc:tc-config/servers/server

in the following document:
Configuration Guide

Cheers,

-Dave

Hi,

Thanks for response, but I think the bind attribute in <sever> element
represents listening address of the server by which TC server receives
all communication including DSO, JMX and other TC server in the array
(l2-group).

Perhaps I could not mention what I exactly want. What I want is something
like below.

1. Configure TC server listening 192.168.0.1/24 for DSO communication from TC client.

2. In the same configuration, set IP address 192.168.1.1/24 for JMX communication.

3. And also, set IP address 192.168.2.1/24 for l2 communication (inter-TC servers communication).

I want to do all 3 things in one same configuration.
So, Is it possible or going to be possible ?

Thank you.

Currently there is no way to specify a different address for each port, only to specify ports.

The funny thing is, they way to do what you want to accomplish would be to have TC bind to all interfaces and then just firewall off the ports to the specific interfaces you want, this would have to be done in the OS or on the network of course.

I can see why you would want to do what you are trying to accomplish even though it sounds very messy to configure from a network standpoint, if you really want this feature is suggest you open a JIRA.

Cheers,

-Dave

Thanks for made it clear that current version dose not have the feature.
I just thought TC may have that kind of feature because some other server product like Apache HTTPD or Tomcat have.

I guess we can have a same effect by the way you mentioned and I'll try it.

Just one thing I wonder is that you thought it's very messy from network standpoint. It's relatively common approach around me to divide network segment according to the purpose and I think the issue what I post is one of a typical concern　related to it.
Dose it sounds strange ?

Thank you.

Well, Apache and Tomcat will let you bind to different IPs for virtual hosting or for different protocols, not for failover. Tomcat clustering is multicast and therefore on a different IP than normal traffic but it's a completely different model. (I could never get their unicast clustering to work)

Regardless, the Terracotta server is not an app server or a web server so I don't think you can really compare the two.

When I said that it was messy, I wasn't talking about the fact that you wanted to separate functions on different subnets. I was thinking about a setup when you were using Network Active passive and you wanted to have the machines fail over in a fully redundant setup. Instead of configuring one network segment to be redundant on your NICs, you would now need 6 NICs all set to fail over, as well as the associated redundant switch setup and you would have to keep track of which segment the L1s were using when they need to failover to a passive TC server. Additionally, the heartbeating that the active and passives do might wind up being on a different network segment than the DSO or JMX traffic and you could get into a situation where one of the 3 segments fails and if it wasn't the heartbeating segment, the L2 would probably not fail over.

Just sounds messy to me. Doesn't mean it is not technically possible.

Cheers,

-Dave

Agree with dmangot. This sounds awesome on first blush but it is dangerous, I think.

As Dave points out you could have your Terracotta cluster refusing to fail over while your apps can't reach us simply because you are using completely separate networking infrastructure for the different data. One path can fail while the other is still up.

I think multiple NICs is a good idea, but bonded or through dual redundant switches with switch-heartbeating and automatic failover.

I think our HA configuration guide explains all this actually.

--Ari

dmangot , ari thanks for response.

At first. I agree what dmangot wrote about the comparison with app server or web server. What I wrote was too rough for correct understanding.

I still have a feeling that I couldn't explain the network I imagine exactly, but I guess I could grasp the point.

Because L2s should know both the status of network to L1s and other L2s, the DSO traffic and health check traffic should go in the same network.

I'll review my network diagram and read HA configuration guide again.
The complexity of the network dmangot mentioned must be also the problem but I re-start from the basics and think about it later.

Thank you.