Yesterday I’ve attended MySQL Customers Conference in London. This event is much smaller size than Users Conference (one day and about 170 people attending) and surely less geeky – there were no one from MySQL Development Support or Consulting teams and Sales Engineers were as close as you could get. Though Anders Karlsson and Ivan Zoratti are one of the best of the kind being more technical than quite a few of the teams mentioned.
It started with Robins presentation about MySQL features and Roadmap. Not what it was something new but it was good to hear about release plans for MySQL 5.1 (Q1 2008) and MySQL 6.0 (Q4 2008) – which sounds pretty ambitious to me but we shall see how many resources 5.1 will take after it is released to get all the issues resolved. Falcon will be default transactional storage engine in MySQL 6.0 – I did not truly understand if it will be recommended transactional storage engine or just default storage engine will switch from MyISAM to Falcon.
It was rather interesting to hear a lot about features of new storage engines such as Nitro and InfoBright but there was nothing told about Innodb features, even ones such as page compression which were publicly announced before.
Somewhere in 6.x series proper performance monitoring instrumentation and semi-synchronous replication is promised together with a lot of other good stuff.
Robin also showed some optimized improvements for subqueries in MySQL 6.0 which are pretty cool and which we still do not have a time to take a closer view at.
There is also not much told about Maria (aka MyISAM++) in product road map presentation, so I guess this is also kept low profile, while later in the conference Anders and David barely mentioned it.
Performance Tuning Presentations – I’ve visited two of these, one by Anders and one by Ivan. There was not much of new stuff to learn (I would not be doing much job if there would be). Though it was fun and mostly correct – we had interesting discussions with Anders. Ivan had number of benchmarks presented in his presentations which however were hard to understand as table schema as well as queries were often missing not even to say about settings.
Customer Presentations – There was one by Swedish Police and one by Net-A-Porter. I was on the second one and I kept thinking how these guys happened to be chosen ? Well It was perfect presentation from Marketing side perhaps the speaker kept quoting benefits of MySQL Enterprise from MySQL Marketing Materials . There is not much of technical insight otherwise – pretty simple site, master slave replication, plans to use DRBD, nothing which I would call fancy or exciting. Though this is may be what London audience is looking for – how does “Average Joe” benefits from MySQL Enterprise offering and MySQL Services. I would prefer to hear about some project which cooler from the technical standpoint and I bet it would not be that hard to find.
Partner Presentations Couple of sponsors had a time to talk. Dolphin Interconnect was perhaps most interesting for guys dealing with High Performance. Though I’d really like them to see to have some work done outside of MySQL Cluster – for example quite possible their solution could help a lot to shorten latency dealing with MemCache servers etc.
MySQL High Availability Solutions This was very good talk by Anders – it felt like he likes this topic a lot and spend plenty of time preparing this presentation and Anders also had a lot of practice in the area. It was mostly focused on DRDB which I’m not the big fan of for various reasons.
The last part was Storage Engine Panel which was very unusual to see without Brian Aker which normally takes part in all Storage Related activities. There were some Q&A about various storage engines – mostly basic stuff.
As a final note I should say this was much more than British conference – were were a lot of people from Sweden, Norway, Denmark, Netherlands and other countries. I guess they are much better of with English speaking conference than one in Paris for example 🙂
About the Author
Peter ZaitsevPeter managed the High Performance Group within MySQL until 2006, when he founded Percona. Peter has a Master's Degree in Computer Science and is an expert in database kernels, computer hardware, and application scaling.
Peter,
I work for the company that develops and maintains DRBD. As such, I always like to learn about the reasons why people _don’t_ want to use DRBD, or are reluctant to. Can you elaborate on the “various reasons” for which you are no fan of DRBD?
Always happy to hear your thoughts.
Cheers,
Florian
Florian,
Thanks for your interest.
Generally I have nothing against DRBD – it is great product which I’ve been using since about 2000 (there were number of bugs with ext3 at that time etc… long time ago)
I just do not think shared or replicated storage is best solution for most MySQL High Availability needs.
The reasons are:
– Bad memory or MySQL Bugs on Primary can corrupt storage… which will be replicated by DRBD
– Long switch time even for Innodb tables – on large system tuned for performance filesystem recovery from journal and Innodb recovery can take 20+ minutes. You can shorten it sometimes (not always) but will pay performance.
– Pretty much unusable for MyISAM tables which may take infinity to check and repair
– Waste of hardware as Secondary node sits and doing nothing (there are ways around but it is not as simple as with slave)
As you can see the reasons have Nothing to do with DRBD – SAN based would have all these reasons plus more – as SAN is expensive piece of single point of failure.
I think for most shops you can do good enough with replication based fail over. Hopefully this will get even better with Integration of Google Synchronous Replication Patches.
There is another way I have in mind but never had time to script it – to use DRBD for binary logs only rather than for database. In this case secondary node (which is slave at the same time) can use these logs to fully catch up before switch over. In this case you will be able to get best of all worlds – consistence up to last transaction, typically fast switch and possibility to use the slave.
Peter, appreciate your sharing your thoughts. About the points you mentioned, I’ve addressed some of those in a blog post a few months ago (http://blogs.linbit.com/florian/2007/06/18/drbd-limitations-or-are-they/) — take a look if you’re interested. Doesn’t really rebut your statements, just presents my view on some of these commonly-voiced concerns.
The idea you state in your final paragraph (keeping just the binlog on DRBD) sounds quite intriguing. Always a good thought to cut down on recovery times on failover. 🙂
Thanks Florian,
I will not get into much discussion of these. In many cases it is not as hard but neither as simple as you put it, ie MySQL Cluster is far from feasible solution for most applications.
I was going to write comparison of few High Availability Techniques some time soon.
Oh Yes and one important point I forgot about is of course “live” ALTER TABLE which one can do in replicated MySQL setup – running alter table on passive node in no-log mode switching and repeating the process.
I attended this conference in London and I have to say I was very disappointed with nearly all of it. Robin’s Roadmap talk was the most interesting, but I sense a bit of ‘wishful thinking’ mixed in with those release dates.
Anders Performance Tuning talk was useless and poorly planned. If you hadn’t corrected his slides for him it would have been a total waste of time. Nothing useful was covered – yet listening to the questions coming from the audience there were people there who really wanted to get some “insider” tips on MySQL performance, and who walked away with nothing. I know he wasn’t supposed to give the talk and had to prepare it in a hurry, but boy it showed.
Ivans talk wasn’t much use either, although far more rehearsed and better presented. Basically half an hour of meaningless graphs with very little (if any!) context.
The customer presentation I saw was (from a MySQL point of view) totally pointless. I have to admit by this point I felt I was just wasting my time being there and left. I didn’t bother staying for Anders second talk, fearing it’d be just as useless as his first (although on reading your post above I regret this now). I also didn’t think the Panel discussion would be worth it as some of the key MySQL people were missing.
All-in-all the venue itself was nice and well organised, but the price of the event compared to the return from it was terrible – and there is no way I shall attend one again in a hurry. I wasn’t expecting a MySQL “training” event, but I doubt there was anyone in the room who didn’t know most of the stuff that was being told to us. We needed it more complex – far more complex. Something like the level of quality we get from this blog.
Thanks Rich,
For real “meat” you should go to the Users Conference in US. This is called “Customers Conference” for the good reason – at large extend it is sales event, which you need to pay for to attend 🙂
They really shouldn’t have included those talks in the line-up then 🙂 If it was purely a sales pitch it should have been sold as one – you could tell from a number of the people there (myself included) that they were already sold on using MySQL, but were there for the technical side of things + future plans. I’d have rather seen them spend half an hour showing off MySQL cluster in operation, or the MySQL monitor application, than hear about how I shouldn’t put a function on the left side of the = sign.
Sadly attending events in the US isn’t an option.