Amazon's cloudy database service has donned its cap and graduated into general availability, after three and a half years of dealing with data workloads. The graduation of the Amazon Web Services Relational Database Service (AWS RDS) was announced by Amazon on Thursday, and along with the standard round of back-patting and …
And does that exclude scheduled downtime ? i.e. "we're going to restart your DB in the next 2 weeks no matter if you like it or not -- you can restart the db in advance so we don't force you, but if you don't restart it we will"
22 minutes is quite a long downtime, I suppose that is enough time for amazon to do those scheduled restarts(not common but they do occur) without having to pay customers anything.
I haven't used RDS (MySQL) in almost a year now but it was absolutely terrible back then. Most annoyingly customers did not have root access to the DB, so when MySQL replication failed (not if, when) the only way to recover was to rebuild the slave from scratch. There are several types of MySQL replication errors that are often safe to skip (in our case the use if in memory tables by the application makes those more common), but skipping was not allowed because we did not get that level of access.
RDS (at least at the time, assuming no significant changes since) was just as wasteful and inefficient as EC2 was/is. Want to take a snapshot of a 100GB database? Please wait for an hour or two while we read out 100GB of data and write it to another location. Want to create a snapshot of a slave? Sorry, no can do (maybe that is fixed now, wouldn't count on it though).
Want to see real I/O statistics? Sorry no can do. The I/O stats in the cloudwatch/rds web console thing was WILDLY inaccurate. I was on the phone with Amazon sales engineer(or some sort of engineer) who said "LOOK! THAT IS AWESOME YOU ARE GETTING 3,500 IOPS!" -- yeah and that 3,500 IOPS gave me 250 kilobytes of data.. That comes out to about 0.7 kilobytes per I/O. With numbers like that, you might as well ask one of my cats how the database is performing because you'll get a better answer.
Screen shot from back then
Take note of that write latency - nearly 200 milliseconds for that spike - for 250 kilobytes of data.
Database performance sucked even though it was a tiny DB in that case.
I really do feel sorry for any talented techies out there who are "forced" to use amazon, I feel for you.
Re: define refund?
"users can ask Amazon for a refund for the downtime"
i.e. you'll get $10 back for the few minutes outage.. but forget about compensation for loss of business.
Its all b*ll*cks.. Mr Bezos, if you really want to tempt me, put your money were you SLA is, for each minute of downtime you give me 1 day of free hosting.. so for a 22 minutes outage, I get 22 days free.
At the level we're talking here
an SLA is practically meaningless.
It gives a credit proportional to the amount you pay which bears no resemblance to the loss you can suffer when a service beyond your control fails in an unexpected way and your data is inaccessible.
For simple web hosting and low value, high volume consumer apps (flipboard is a great example) these services are a very sensible choice. For anything enterprise grade in which data unavailability or loss causes meaningful business pain each layer of abstraction away from a hands-on service presents a material risk that amounts to putting fingers in ears and hoping for the best. Colocation is the least bad outsourcing option, then managed hosting, Iaas and, worst of all, proprietary service in a box type solutions which give you no insight into how they work or when or how they may go wrong.
Of course, lots of web infrastructure is really not that important, so there's a big growth market for Amazon in web 2.0 type apps which are trendy but can go wrong sometimes without the world ending, but if corporates are really putting their critical data there then horror stories are inevitable.
- Updated Hidden network packet sniffer in MILLIONS of iPhones, iPads – expert
- Students hack Tesla Model S, make all its doors pop open IN MOTION
- BBC goes offline in MASSIVE COCKUP: Stephen Fry partly muzzled
- PROOF the Apple iPhone 6 rumor mill hype-gasm has reached its logical conclusion
- US judge: YES, cops or feds so can slurp an ENTIRE Gmail account