List:Cluster« Previous MessageNext Message »
From:Tomas Ulin Date:November 23 2004 11:42am
Subject:Re: LOAD Testing NDB mySQL 4.1.7
View as plain text  
5 storage nodes is not a valid configuration if you need redundancy, 
please use a multiple of 2, and set NoOfReplicas=2.  Now, if 1 node 
fails they all go down since there is no redundancy in the system...

You will have a hard getting performance out of a setup with only one 
mysqlserver, you will see storage nodes loaded about 10%.

I would recommend to have at least twice as many mysl servers as storage 
nodes, maybe as many as twice, depending on the type of queries your 
running.

In 4.1.8 we have made several select performance improvements with some 
tuning options...

Anyways, you shouldn't be getting the errors you're getting below.

I need:
config.ini
ndb_<id>_cluster.log
ndb_<id>_error.log                    (only from the node that dies first)
ndb_<id>_trace.log.<no>           (only from the node that dies first)


And if you can provide something that will reproduce your problems...  
that is the best...


T

Dathan Vance Pattishall wrote:

>Wrote a program to test my NDB cluster:
>
>
>
>1 NDB MGMNT Server
>
> 
>
>5 NDB Storage Nodes
>
> 
>
>1 mySQL client.
>
> 
>
> 
>
>The goal was to achieve >= speed to a myISAM table loaded into memory for a
>sustained period of a few hours.
>
> 
>
>Test setup.
>
> 
>
>A process that forks off database connections an runs typical quieries for
>the Friendster.com site.
>
> 
>
>45 simultaneous threads sustained doing pure select Non-cacheable qureiries.
>
>NDB got up to 1200 qps but then failed with these errors:
>
> 
>
>dathan@ndb4:~> /usr/local/mysql/bin/perror --ndb 4025
>
>Error code 4025:  Node failure caused abort of transaction: Temporary error:
>Node Recovery error
>
> 
>
>/usr/local/mysql/bin/perror --ndb 4010
>
>Error code 4010:  Node failure caused abort of transaction: Temporary error:
>Node Recovery error
>
> 
>
> 
>
>Then the slave SQL thread failed with 4010 as well.
>
> 
>
> 
>
>Then connecting to the management node I see that ALL storage nodes failed.
>
> 
>
>What information does the list need in an effort to track down the problem?
>
> 
>
>DVP
>
>----
>
>Dathan Vance Pattishall     http://www.friendster.com
>
> 
>
>
>  
>

Thread
LOAD Testing NDB mySQL 4.1.7Dathan Vance Pattishall19 Nov
  • Re: LOAD Testing NDB mySQL 4.1.7Tomas Ulin23 Nov
  • Re: LOAD Testing NDB mySQL 4.1.7Tomas Ulin23 Nov