TREX Name Server ?

[Former Member]

Hi *,

I have a small conceptual question implementing a clustered TREX (2 hosts only).

I have 1 server running all tasks as Master and a 2nd server running an Index Server and a Preprocessing.

To my knowledge, the coordination between the servers goes through the Name Server on machine 1. This will distribute the load (search requests) etc to the 2nd machine and a 3rd if there was one.

However, in this scanrio (simple distributed TREX), there is only 1 web server - on the Master.

The documentation of SAP says that in the CM configuration in the portal, you can enter a Name Server only and exceptionally if you're using more than one Web server. Thus I cannot connect the Name Server of my Master to CM, but will connect it rather as a single host.

Question: If I'm not announcing the Name Server of the Master in CM, how does the load distribution take place to the Slave ???

[Former Member]

I'm sure Karsten. It's just plenty of people do have distributed environments and quite a few questions about nameserver configuration etc.. all I need is a high level direction because to be honest right now i'm not sure anymore.

I see contradictions in the documentation which I can't explain.

It seems you only need a name server if you have multiple servers running the same services (Indexing and Queues for example), but you don't need it if you just split 1 server running Indexing and Queuing, and a second running IS/PP. Replication is probably needed to ensure the second server is in sync.

But then how are the results returned to the master, is it via tcpip/java or is an extra webserver necessary on the slave too ?

If yes, then we get 2 webservers, which requires a name server on the master.

It just sounds messy and I need some advice on how this really works to ensure i take the right decision for the environment i'm building now..

A thousand thanks for your support !

D.

[KarstenH]

Hi again,

>> I see contradictions in the documentation which I can't explain.

I caught your point and did also point BOTH responsible docu writers to this thread here.

Cheers,

Karsten

[Former Member]

Hi Dimitar,

It would be interesting to know in which document did you read the piece of information that you would need more than one Webserver? Was it a TREX documentation or a KM documentation? Furthermore, please tell us which version of the Portal and TREX are you talking about?

My guess is that we are dealing with an EP 6 SP2 Portal or newer (NetWeaver '04) and a TREX 6.1 version. In that case you only specifiy the TREX Name Server within the Portal. Access from the Portal first leads via the TREX Name Server, which "knows" all TREX services running within the landscape. So, the Portal does not care about the number of Webservers, Index Servers, etc. anymore.

If you have found some statement within the documentation that says that you would need a Webserver running on each TREX instance within your TREX landscape this would be wrong, except there is a certain context desribed in the documentation. Could it probably also be that your documentation refers to EP 5.0 scenarios but you are dealing with an EP 6.0 / NW'04 scenario?

I hope this helps for clarification.

Best regards,

Sascha.

[Former Member]

Hi Sascha, Karsten,

Thanks a LOT for the fast feedback, it is very much appriciated !

Sascha, I wish it was the landscape you describe. TREX 6.1 is already much nicer in terms of administration then 6.0. Unfortunately it's not the current case.

I have an EP5 SP5 and a TREX 6.0 SP1.

The last step of the configuration described for a simple distributed TREX scenario is

-

---

Only on the portal server, and only if implementing multiple Web servers for TREX:

• Register the name server with Content Management [Page 46]

• Restart the servlet engine

-

---

Here's a link to the guide :

https://www.sdn.sap.com/irj/servlet/prt/portal/prtroot/docs/library/uuid/079cca90-0201-0010-d1b3-db4...

That same statement can be seen throughout the other scaling guides as well, not only this one. And to my understanding it implies that you should use a name server "only if implementing multuple Web servers".

But if that is not the case, then is a name server needed at all, if the different services are not duplicated on different servers ?

Let's say I have 2 boxes as follows :

Master:

Full install, all services (NS,IS,PP,QS,WS..)

Slave:

IS,PP,QS (no webserver, no NS)

I would use the replication procedure to get the indexes of the manster unto the slave server. I will configure the slave to point out it IS a slave server.

I will configure the Master to point out it IS a master server.

Then in that scenario I need to connect the Name Server of Master into the EP5 Portal.

Is that correct ?

Or do I have to connect the portal to the master not via the name server definition, but each service (queue, indexing etc) server definitions (well actually the local_config.properties) ?

---

What happens if I setup 1 full master without it's QUEUE Server and Name Server, then install a second server which only runs the Queue Server. Then manually point each individual property of the portal to the 2 servers (no name server used at all) ? Index are replicated again using the replication procedure (zip files basically to my understanding, name server has no role in the replication to my understanding)

[Former Member]

Hi Dimitar,

In the case that you are using an EP 5.0 Portal, the documentation is correct. You do not need to specify a TREX Name Server in your EP if you are only working with one HTTP Server. Actually, it is not recommended to use it in such a case as it would slow down the communication process between KM/Portal and TREX a little bit. If there is a TREX Name Server specified it will always be asked for the appropriate Web Server. If there is only one Web Server this does not make sense. If there is no TREX Name Server, the specified Web Server will directly be used, i.e. you save one indirection. That means it should not harm the system if you specify the TREX Name Server in the Portal, but it would not really help. If there is no TREX Name Server specified then you also have to set the other services explicitly (Index Server and Queue Server.)

In order to clarify this point, you need to start a TREX Name Server process on the master TREX machine in any case if you want to work with replication.

With respect to the second scenario mentioned, I cannot give you a clear picture of what will happen. If it is not described by the documentation this example is not supported. Personally, I think that it will only work correctly with a TREX Name Server started. (For TREX 6.1 and newer, the TREX Name Server is mandatory for each TREX instance anyway.)

Best regards,

Sascha.

[Former Member]

so nobody has implemented a simple distributed environment and none knows what the server name is used for ?

[KarstenH]

Hi Dimitar,

1) I will point some people to your post who have...

2) As you write: "...so nobody has..."

Let me just point out once again that SDN has no answer guarantee...and definitely not within 24 hours...

Regards,

Karsten