Comments for HiveDB

Comment on Switching from SVN to Git by Alex Li

Alex Li — Fri, 30 May 2008 22:57:16 +0000

It is a wonderful move to stay away from SVN.

Unfortunately, Git seems does not handle file rename/move well either during merge time. This is terrible if you do refactoring a lot in Java. That is the main reason we are switching to bazaar instead. More to read here: http://www.markshuttleworth.com/archives/123

Comment on HiveDB Maven Repository by britt

britt — Sat, 19 Jan 2008 19:27:31 +0000

@MikeD

Hi MikeD,

I’m not sure that I understand your question correctly. So, if I’m answering the wrong question forgive me.

The primary difference between HiveDB’s partition by key and a load-balancing scenario like a reverse proxy is that HiveDB is designed to scale storage capacity rather than even distribute incoming load across a number of servers. So, they serve different purposes.

Comment on Preparing Your Data to be Horizontally Scaled with HiveDB by britt

britt — Sat, 19 Jan 2008 19:20:28 +0000

@Divya B

1. HiveDB doesn’t handle replication. In general we defer to MySQL native replication. However, HiveDB does support storing records in more than one location if you want to handle replication within your application.

2. Currently we don’t provide a facility for back-filling records onto new nodes so that records are evenly distributed. The default assignment algorithm will evenly distribute new records across nodes, but if you want to move existing records onto the new node you will have to write your own implementation of the Migrator interface and use it to balance the nodes.

We do have plans to implement a balancing and migration tool in the future, we just don’t have it yet.

Comment on Preparing Your Data to be Horizontally Scaled with HiveDB by Divya B

Divya B — Sat, 19 Jan 2008 17:53:13 +0000

Could you also give us some information about the following?

1. Is there any data replication happening in this setup?
2, How is it handling uniform distribution of data when a new box is added?

Comment on HiveDB Maven Repository by MikeD

MikeD — Thu, 17 Jan 2008 02:05:58 +0000

Hi - many thanks for making this technology available.

From a quick reading of the documentation, it appears that the HiveDB library assumes that a single Java app server will connect to multiple MySQL database servers.

Do you have any resources you could link to that discuss the pros and cons of this ‘partition routing’ being done in the app server .vs. earlier (such as an Apache reverse proxy with mod_rewrite)?

Comment on Preparing Your Data to be Horizontally Scaled with HiveDB by gio

gio — Tue, 06 Nov 2007 13:45:38 +0000

There’s a typo in the title “preparing you data” :-)
Interesting piece of software.

Best regards,

Comment on HiveDB Client Summit by Justin McCarthy

Justin McCarthy — Thu, 28 Jun 2007 22:49:33 +0000

Looking forward to it.

Comment on HiveDB Client Summit by Ronald Bradford

Ronald Bradford — Thu, 28 Jun 2007 20:08:46 +0000

Doh!, if only it was the next week when I was in SF.

Comment on Hive Connections are Transaction Scoped by justin

justin — Tue, 15 May 2007 05:36:42 +0000

I’m interested to hear some reader feedback on this design decision.

Rather than asking clients to favor short-lived connections, HiveDB could have opted to throw exceptions on all JDBC operations that were capable of generating writes (for instance, preparedStatement.executeUpdate(…)).

Doing so would have provided a guarantee that nodes marked “read only” will not receive new records, however at the expense of overloading the existing SQLException semantics.

Thoughts? How do readers feel about catching SQLException as a mechanism for detecting whether an instance of a JDBC Connection has been marked read only?