Blogs › seppo's blog ›

Watching the Paint to Dry

Submitted by seppo on January 27, 2009 - 15:07

The life cycle of nerd contains an inevitable empty phase after all the hectic crunching to get release out is over. Nerd-work, is actually not so different from carpenter, artisan - a violin builder for instance. When the instrument is ready, there is nothing much to do, but to watch it dry. He cannot play with those thick fingers and ears, which don't make difference between flats and sharps.

We cannot use a DB cluster, but we have built one.

But, what's next?

Open Source Projects

We decided to split our "wsrep mysql patch" open source project in two:

  • wsrep API project, to define just the API
  • mysql patch project, which contains integration of wsrep in MySQL engine

Then, we will start two other projects:

  • postgres patch project for integrating wsrep in postgreSQL
  • openrep, which is open source implementation for wsrep replication.

These should appear in launchpad some time soon, seems we have more violins to handcraft.

Fixing wsrep

In the MySQL wsrep integration, we have observed some issues,
which need attention. Luckily, we have received open source contributions, which helps our work a lot, and more than anything, encourages us to try harder (hmm.. to do what, watching the drying violin? Well, I think I am straightening up already while writing this)


  1. One evaluator tried to run mysql test suite against Galera and observed a lot of failures. We got open source contribution to fix this bootstrap issue, and with the fix, most tests are supposed to pass. However, there remains issues with replication tests due to empty binlog files.

  • The empty binlog files, has been a long known issue in the implementation, which we kind of "have learnt to live with". Now we are getting contribution to fix this. Not everybody likes to watch empty files, I see.

  • Auto increment management is not automatic and this is something that can spoil the otherwise divine pleasure of evaluating Galera. Currently, if you want to use autoinc columns, you should define auto_increment_increment and auto_increment_offset for each node, so that values from separate cluster nodes won't overlap. It is possible to automate this, and let Galera to take care of correct auto increment management. Alex, (who is automation freak, or lazy configurator, perhaps?) created a bug about this in launchpad. I think, we are going to fix it soon, and publish demo2 to return peace of mind experience for
    evaluating.

    • But, enough fiddling with the blog already, back to busy-ness