clustering with tomcat and

[ro...@gmx.de]

Dear all, 

We are quite far in evaluating ofbiz for a heavy load system on the ecommerce (cluster with at least two application nodes and a HA-Postegres database ) 
This is working fine in most cases and when there is little load on the systems. 

But there are two things we didn't succeed to solve yet: 

- session replication tomcat to tomcat:  although we succeeded to replicate session with an standard Tomcat we do not succeed to replicate sessions between two embedded ofbiz-tomcats  - is the implementation of the embedded tomcat different of the standalone version ?


- minerva XA-Pool (?)  - during our tests we found out that the connection pooler works not work properly if we test heavy load on two cluster nodes and one database (Postgres 8.1) - the database gets locked by one of the instances( could not commit...), most often it happens on the sequenceID table and then all the connections from the pool get used.  Bad thing is the even if we stop the locking instance the other node does not recover.
(we tested also with the minerva from the hotwax media-repository) 


How we test: 
- we run some end-to-end tests with a testing tool( browse catalog, register users, login/out, add to basket, send orders, run services in the backend)
- put some additional io-load on the database node 

...and wait few seconds/minutes to see it crashing  
If there is no additional load on the database node it locks afte two days latest ...


So I'll be really happy to hear that there is some experience in clustering   ofbiz for load and availability reasons. 

Thanks
Robert
-- 
GMX FreeMail: 1 GB Postfach, 5 E-Mail-Adressen, 10 Free SMS.
Alle Infos und kostenlose Anmeldung: http://www.gmx.net/de/go/freemail

Re: clustering with tomcat and

[David E Jones <jo...@hotwaxmedia.com>]

You should try the latest OFBiz from SVN. There are various recent  
improvements in this area:

1. now using Apache DBCP instead of Minerva
2. problem with transactions, synchronization, and locking in EE  
sequencer
3. problem with concurrency and locking in async and scheduled  
services (this was just fixed a couple of days ago)

-David


On Dec 3, 2007, at 3:35 AM, robert.ruprecht@gmx.de wrote:

> Dear all,
>
> We are quite far in evaluating ofbiz for a heavy load system on the  
> ecommerce (cluster with at least two application nodes and a HA- 
> Postegres database )
> This is working fine in most cases and when there is little load on  
> the systems.
>
> But there are two things we didn't succeed to solve yet:
>
> - session replication tomcat to tomcat:  although we succeeded to  
> replicate session with an standard Tomcat we do not succeed to  
> replicate sessions between two embedded ofbiz-tomcats  - is the  
> implementation of the embedded tomcat different of the standalone  
> version ?
>
>
> - minerva XA-Pool (?)  - during our tests we found out that the  
> connection pooler works not work properly if we test heavy load on  
> two cluster nodes and one database (Postgres 8.1) - the database  
> gets locked by one of the instances( could not commit...), most  
> often it happens on the sequenceID table and then all the  
> connections from the pool get used.  Bad thing is the even if we  
> stop the locking instance the other node does not recover.
> (we tested also with the minerva from the hotwax media-repository)
>
>
> How we test:
> - we run some end-to-end tests with a testing tool( browse catalog,  
> register users, login/out, add to basket, send orders, run services  
> in the backend)
> - put some additional io-load on the database node
>
> ...and wait few seconds/minutes to see it crashing
> If there is no additional load on the database node it locks afte  
> two days latest ...
>
>
> So I'll be really happy to hear that there is some experience in  
> clustering   ofbiz for load and availability reasons.
>
> Thanks
> Robert
> -- 
> GMX FreeMail: 1 GB Postfach, 5 E-Mail-Adressen, 10 Free SMS.
> Alle Infos und kostenlose Anmeldung: http://www.gmx.net/de/go/freemail