Jin,
I don't know which implementation of TPC-W you are using. But if all
nodes complain identically about the same key violation it is certainly
a bug in the application.
I'm testing TPCW on sequoia 2.10.2. For the test, I'm using one
controller + one backend server for debugging puprpose.. Originally, I
was testing with 4 backends.
There are two tomcat servers connected to sequoia.
I encouters some erros about duplicate key problems when request rates
are high. If the request rate is low, then it's okay. I guess the
problem is that most of the tables in the TPCW implementation doesn't
use auto incremented
primary key. So before inserting a new record, it reads the max id
value in the table and increases it by one for the new id. After
insertion opertaion, TPCW commits transaction. MySQL server works
correctly with the operation.
If the code does something like: BEGIN; SELECT max(id) FROM ... ; INSERT
INTO xxx VALUES (id) ... ; COMMIT
then this is certainly not thread safe and will lead to duplicate key
exceptions when load increases.
Both operation should be done atomically which could be achieved with a
subselect in the INSERT statement (I don't know if you version of MySQL
supports it).
But with Sequoia, it looks like the operation deosn't perform
automicly and the read operatation sometimes is stale.
For sure Sequoia will execute in parallel the different SELECT that will
come from different transactions. Only writes will get serialized (not
that the same could happen on a single MySQL instance and you are maybe
just lucky not to have seen the problem yet on a single MySQL instance).
Does anybody have the same problem? Anybody successed in testing TPCW
with RAID1b-1?
I did not run it recently but I used to run the implementation from
University of Wisconsin quite successfully.
Hope this helps,
Emmanuel
--
Emmanuel Cecchet
Chief Scientific Officer, Continuent
Blog: http://emanux.blogspot.com/
Open source: http://www.continuent.org
Corporate: http://www.continuent.com
Skype: emmanuel_cecchet
Cell: +33 687 342 685
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
