Tuesday, December 23, 2008

MySQL Cluster 6.4 - BETA

The Configurator and Sandbox (only the Custom part so far) has been updated to support MySQL Cluster 6.4!
MySQL Cluster 6.4 is a fantastic piece of software with new features such as:
  • Multi-threaded Support. A new configuration parameter, MaxNoOfExecutionThreads=[2,4,8] allows you to set how many cores the data node should utilize.
    MaxNoOfExecutionThreads=2 for two core machines
    MaxNoOfExecutionThreads=4 for quad core machines
    MaxNoOfExecutionThreads=8 for eight core machines
    With 8 you will get 4 query handling threads! Can you beat Jonas' 1M reads on a single box?
  • Online add node - allows you to scale online! I will publish a procedure how to do this very soon. Currently the Configurator and the Sandbox do not support this.
  • Windows Support (I have not tried this and not 100% of the quality, nor how to build it, but it should be there).
  • Snapshot backup - START BACKUP, [SNAPSHOTSTART | SNAPSHOTEND]
    SNAPSHOTEND - the backup contains the database at backup end time (default and same as behavior in previos releases)
    SNAPSHOTSTART - the backup contains the database at backup start time
And remember it is a beta, and be prepared for bugs and other oddities until it has stabilized.

8 comments:

Matthew Montgomery said...

Hi Johan,

http://dev.mysql.com/doc/refman/5.1/en/mysql-cluster-news-5-1-30-ndb-6-4-0.html states that "This is a new Alpha development release". I understood 6.4 BETA would not be ready for at least a few more months.

漢(と書いてオトコ) said...

Hi Johan,

Why don't you mention about ndb$info?

Jonas Oreland said...

ndb$info is not part of this release
but will be added later

hingo said...

Matt: There is some confusion here. Since Jon left early to spend Christmas in Australia, the manual page you link to has not been updated (for other things either). So in this case you shouldn't trust it as a source, even if usually you should.

Early on I think it was talked about as an alpha, but most people (including the managers who should decide such things) are now calling it beta, so beta it is. (It's not many months to GA anyway, so beta is perfectly appropriate.)

Muthukumaran said...

Hi Johan,

What is the impact of data-distribution (based on hash-on-key) to partitions due to dynamic addition of data-nodes?

Jon Stephens said...

Muthukumaran,

New data (that is, data added after the new nodes are added) is distributed using all nodes. To redistribute data already stored in the cluster before the new nodes were added, run ALTER TABLE table-name REORGANIZE PARTITION on each NDBCLUSTER table.

See Adding MySQL Cluster Data Nodes Online for details.

Muthukumaran said...

Hi John,

Just a related but more internal question regarding distribution/re-distribution. Does NDB use "Consistent Hashing" of keys (as in memcached or Amazon Dynamo or Scalaris type of key-value stores)to distribute data among the NDB nodes or it uses any other strategy?

anthor said...

quick question about 6.4 - compiling it on Mac OSX 10.5 results in a linker error on the ndb_mgmd (see below).
Is ndb_mgmd just broken now? or are there compiler flags I can use to help out?

thx.
still erroring:on compiling -o ndb_mgmd
ld: duplicate symbol Vector...::size() const in ConfigManager.o and MgmtSrvr.o