Comments on: Finding your MySQL High-Availability solution – The definitions

By: peter

peter — Sat, 17 Oct 2009 03:57:34 +0000

Yves,

Thanks for great post. The downtime is indeed tricky measurement. The planned vs Unplanned downtime and online upgrades of software or schema changes is one issue. The measurement interval is another issue – there is the difference between couple of min downtime every day vs going down for a day once in a several years. Basically 99.9% uptime every day of the year is very different from 99.9% measured across the full year.

Finally I wish it always would be black and white with systems up or down – in a lot of cases we can see system being degraded, acting funny and being unable to perform some of the functions with required performance but not totally down.

By: Finding your MySQL High-Availability solution â€“ The questions | MySQL Performance Blog

Fri, 16 Oct 2009 21:02:57 +0000

[...] having reviewed the definition my the previous post (The definitions), the next step is to respond to some [...]

By: Travis

Travis — Wed, 14 Oct 2009 00:28:10 +0000

@yves
I was referring to mysql not being easy to monitor via snmp, at least not for us less skilled folks. MRTG makes nice work out of the box for cpu, memory, etc. Making changes via my.cnf is definately a trial by wait process. I modified my innodb log file size today from 25M to 128M and my disk IO decreased very nicely. I obtained that from your other blog
Mucho thanks!

By: yves

yves — Tue, 13 Oct 2009 13:46:03 +0000

@Morgan, the availability should count the online upgrade operations. That’s one of the great thing about NDB Cluster, all the system can be upgraded without down time.

@Giulio, it is mainly a question of being comfortable to work a given company. Oracle is a very successful company but it also has values that I am not ready to be associated with.

@Bill, They are coming… I have a few days off after being on site for three weeks, I should publish the next one this week.

@Travis, I am not sure I am following you. Monitoring servers is fairly easy with Cacti or even through SNMP and custom scripts (Maybe I should blog about this…) If the slave has a consistent database image why could it not be able to continue to operate? I mean, the write traffic will of course needed to be redirected to it.

@Imran, You are completely right, failback can be a real pain and I should highlight that point more.

By: imran

imran — Tue, 13 Oct 2009 08:52:27 +0000

One other term that warrants much love is that of failback – which depending on the setup can either go smoothly or be incredibly painful.

By: Travis Sullivan

Travis Sullivan — Mon, 12 Oct 2009 19:41:25 +0000

With a lack of snmp support in mysql, monitoring performance of clustered servers is quite a task. Even if we have several mysql replication servers, when the master db fails, I don’t understand how the slave servers continue to operate.

By: Bill

Bill — Mon, 12 Oct 2009 16:56:07 +0000

Looking forward to more in depth articles on the subject. It will be nice to read from without the Sun/Mysql(Soon to be Oracle) marketing fluff in the way. All of their case studies and webinars, while emphasizing the benefits of common Mysql HA Setups fail to address the drawbacks.

By: Giulio

Giulio — Sat, 10 Oct 2009 09:48:58 +0000

Can I ask why you leave SUN? And what do you think about the future of mysql at sun/oracle and outside?

By: Darius Jahandarie

Darius Jahandarie — Sat, 10 Oct 2009 05:32:49 +0000

I’ve always found the “Level of Availability” a fairly useless measurement. As more “9″s are tacked onto the end, it just makes it feel even less quantitative and more marketesque. An actual “7 days per year” sound much more analytical and believable.

By: Morgan Tocker

Morgan Tocker — Sat, 10 Oct 2009 01:16:51 +0000

@Yves – I like the way that you pointed out how level of availability is always subject to debate, I couldn’t agree more.

I wonder if you could comment on unscheduled versus scheduled maintenance? I’ve seen some technology providers claim high “nines” support but they neglect the ability to easily do online upgrades.