Friday, March 23, 2007

mysql-ha-erl trunk, rc-1 on monday

I've imported the new erlang tree into svn.
You can check it out at:

https://svn.sourceforge.net/svnroot/mysql-ha/trunk/mysql-ha-erl/

It's still at an early stage, but the slave routine is outlined and the skeleton works fine.
You can test it on just one node if you start two erlang shells with the proper -name parameters.

This tree uses an extended /etc/mysql-ha.conf file. I'll be documenting all this during the next few weeks.

In the meantime, I'll be releasing rc-1 on monday, with an updated Installation Guide.

Happy Spawning!

Saturday, March 10, 2007

an erlang fork and rc-1 comming soon

Hello all,

I'm back from my break, and while rc-1 isn't ready yet, it's really close now, I promise :)
I'm also working on a complete rewrite of the cluster in erlang. My original goal was to have a concrete project to do to learn the language, but now I've seen the light, and I honestly believe the erlang version will be better.

There's no svn trunk for this yet, I'll probably import it before the weekend is over so any interested party can take a peek. At the current rate, I expect a first release of beta quality by the end of march, with no need for ssh or ping, amongst other advantages of using erlang!

Back to the current almost stable tree, I expect to release rc-1 by thursday or friday, and after that, I'll work a bit again on the installer package, upgrade all the docs, etc.

As usual, any testing is welcome. We're now running on Fedora Core, CentOS (and that should mean RHEL, WhiteBox and Scientific) and we're almost there with Ubuntu and Debian.

After all the docs are up to date with rc-1, I'll just wait for any remaining bugs to show up and keep improving distro support (I have my eye on the BSDs too).

Regads,

Friday, February 09, 2007

summer break

The project is entering a slow month right now.
I recently had a small procedure done to my back, and once I'm fully recovered, I expect to take a couple
of weeks off.

However, I will continue to work on the cluster, with no new features, just heavy testing, hoping to
come back with a release candidate by the end of the month.

Wish me luck :)

Wednesday, January 17, 2007

beta-0.6.8 : thresholds up to microseconds!

beta-0.6.8 is out today.

This includes a fix for the startup procedure after a hard fail (the bug being that we couldn't start the
ssh-agent properly due to a conflict with the previous socket), and a major change in the thresholds system.

The cluster now supports thresholds in seconds (as up until now), milliseconds and microseconds (reliablilty depends on hardware, man usleep for more info).

I've changed all invocations to the sleep function for usleep, except in the main loops, and I'm giving admins the chance to express THRESHOLDS in seconds (no units), milliseconds (ms) and
microseconds (us) in /etc/mysql-ha.

The variables affected are:

MONITOR_CHK_THRESHOLD
MYSQL_KILL_WAIT
MYSQL_RESTART_WAIT
SIG_KILL_WAIT

This will be updated in the Installation Guide tomorrow.

I believe this gives administrators much more control over the cluster and takeover/recover time, though it needs proper documentation in the Guide, because if the values are too low for your environment (bad network, loaded server) you can have a false positive. Still, I've taken some good measures to prevent a false positive failover/takeover from happening.

Please go grab this, and happy clustering!

Wednesday, December 27, 2006

Celebrating new year with our first beta

I'm celebrating the new year with the first mysql-ha beta release.

I've split the releases in two packages:
  • mysql-ha, which continues with the release numbering we've used until now, and includes only the code needed in order to run the cluster. You must install and configure the cluster manually, following this guide. This package is now on beta.
  • installer, which is still alpha, and is numbered from 0.1.0 again.

I hope that this split will encourage more people to try out the cluster, by isolating the code that had more bugs and caused more troubles (the automatic installer) from the almost stable cluster code.