AWS Post-Mortem

http://aws.amazon.com/message/2329B7/

by zeit_geist • 15 years ago

145 65 15 years ago

aws.amazon.com

I'm thinking AWS needs to implement the Availability Zones: AZ-ChaosMonkey and AZ-ChaosApe. Having a dedicated playground for breaking things where they can start to observe how this complex system reacts to simple failures and gaps in assumptions.

harshaw• 15 years ago

Sure. Presumably Amazon has a test lab that replicates multiple zones :) Perhaps your point is that Amazon should make this test lab public so people can contribute to the QA effort?

IIRC many of these datacenter failures start with a utility company power outage followed by a failure of the secondary power systems (I'm thinking of some past failures at softlayer and other providers). I wonder if it is prohibitively expensive to do a real life system test on a big data center (or prohibitively expensive once the data center is on line). For example, how often do they turn off one of the mains (unexpectedly) to see what happens with the backup system?

gwern• 15 years ago

It's hard to believe that they wouldn't test the mains by doing just that.

I visited the NY LaGuardia TRACON recently - built perhaps 40 or 50 years ago? - and saw the generator & battery room, where they turn off the utility power every few months just to see whether things are working. So it's not exactly a new idea or an idea that other life-&-death-mission-critical operations don't dare use.

byoung2• 15 years ago

IIRC many of these datacenter failures start with a utility company power outage followed by a failure of the secondary power systems

That happened at Rackspace a few years back: http://techcrunch.com/2009/06/30/what-went-down-at-rackspace...

I have an account with GoGrid, and they do a regular testing of their backup generators. I'm not sure if they throw the switch on the mains, though.

cperciva5• 15 years ago

I wonder if it is prohibitively expensive to do a real life system test on a big data center

It's probably prohibitively dangerous. Backup power systems don't have many-nines of reliability; generators which are reliable enough for the once-a-decade event when a car crash knocks out your utility power aren't anywhere near the reliability needed to run your datacentre for an hour every month as a test.

spartango• 15 years ago

This is in the pipe.

mtkd• 15 years ago

It's a good communication from Amazon - maybe a little too long - could use a summary block at top.

The compensation looks generous too.

grourk• 15 years ago

There's a summary block at the bottom -- but it's not a summary.

saturn7• 15 years ago

As someone who has put a considerable amount of resources moving things into cloud computing - I wanted to believe. But I have changed my mind.

Cloud computing scales the efficiencies, yes. It also scales the problems. And because of this, AWS is by several orders of magnitude the worst of my current hosts.

I have dedicated servers. No downtime in past year. I have a couple of cloud servers with rackspace. No downtime (although i don't recommend them). I have some VPSes with local providers. No downtime.

AWS? More than 24hrs downtime in the last year. Seriously, for someone trying to run web sites reliably - screw that. I'm not using AWS any more.

And don't even get me started on the apologists. "EBS slow as treacle? Well you should have been running a multi zone raid-20 redundant array! Duh!". "EC2 instances dying at random? Well you should architect and implement a multi-master failover intelligent grid!"

I used to be under some kind of crazy delusional spell that the above was correct and it was somehow my fault that I wasn't correctly adapting to AWS's numerous failings. Well, no more. Now I realise that I should just stick with the super reliable service I know and love from traditional operators. You need to programmatically grow and shrink your app server flock? Great, use AWS. For the other 99.999% of us - stick with what you were using before.

quanticle• 15 years ago

Is there any reason in particular why you wouldn't recommend Rackspace?

saturn7• 15 years ago

Hm. Well, I don't like them. It's subjective, you might disagree. But off the top of my head:

1. Contracts. They want 1 years minimum contracts for any dedicated servers. For truly gargantuan orders I could understand this but for one puny server? Never.

2. Their definition of "cloud" is different from mine. To use their "cloud" services your servers need to be public facing, ie on public IPs. Want them on your own VPN? You can stil get the cloud prices but not the API. you create and cancel servers via tickets. This is different from VPS how?

3. Sloooow provisioning - even if you are able to use their "public cloud" API to provision a server - prepare to wait hours for it to be done, leading me to suspect it does nothing more than email a tech to provision a VPS and hook it up somehow. Oh, you can't pause them to save money either, again making me think these "cloud" servers are nothing more than slicehosts with an extra layer of abstraction

Is that enough? I could go on.

tlack• 15 years ago

For what its worth, I found Rackspace's support to not be worth the extremely high cost. Softlayer is just as good.

tybris• 15 years ago

"As part of our efforts to continually improve our Rackspace Cloud offerings, we will be performing maintenance on our Cloud Servers environment. The maintenance window has been rescheduled and will now occur on Friday November 12th, 2010 from 10:00 pm US-CDT (3:00 am GMT) and end Saturday November 13th, 2010 at 10:00am US-CDT (3:00 pm GMT). This maintenance is required to update our billing systems. These changes will not affect nor change any of your current billing fees."

Ok, scheduled down-time is better than unscheduled down-time, but it's still down-time.

cperciva5• 15 years ago

AWS? More than 24hrs downtime in the last year.

You must have incredible luck. Sure, AWS has a lot of outages; but most of them only affect a small proportion of their users.

saturn7• 15 years ago

Huh? North Virginia was down for something like 30 hours in April. I had some instances in there. Hardly "incredible luck". And it was not just me: http://www.google.com.au/search?q=AWS+Virginia+outage

karambahh• 15 years ago

I think you nailed it. To each his own, and a good infrastructure probably is a mix of cloud, virtual, physical servers. I currently have a similar setup as yours (but I suspect mine is smaller, about 40 machines total), but my main concern is with my primary DC which seems to periodically loose utility power, loose A/C, cut all fibers at once, etc....

rapind2• 15 years ago

I'm the same. I fully embraced S3 and EC2 when they came out (even played with SQS) and enthusiastically told everyone I could that this was the future, it's the new electricity, etc.

While I still think that eventually it will end up as a utility I'm opting out of the cloud for anything production for the time being. I'll keep an eye on it of course. Does anyone know if Heroku has spread their services across HA zones?

I'll be keeping my Linodes though. They've been great.

aaronbrethorst• 15 years ago

    Does anyone know if Heroku has spread their services across HA zones?

I don't know for a fact, but the multiple Heroku sites of mine that went down concurrently with this problem would indicate 'no.'

dangrossman2• 15 years ago

Same story.

A few years back I had moved the majority of my sites onto Amazon's web services, using more and more of them as they were released. W3Counter.com was, just a year ago, using multiple EC2 instances of various sizes, Elastic Load Balancers, Amazon RDS instances with Multi-AZ failover, RAID arrays of EBS disks... and during those years, its reliability and performance degraded significantly compared to when it was run on physical servers. I was hit by the big Virginia outage as well.

Two months ago, I took the time to rearchitect everything and move every site back to various dedicated servers at SoftLayer. I feel like I'm in control again... no more worrying about things like the network latency of my hard disks. Despite prepaying significantly for reserved instances on both EC2 and RDS, the costs of dedicated hardware are still significantly less, too.

It was painful to waste thousands of dollars in reserved resources I'll now never use, but you know what they say about sunk costs.

rdl4• 15 years ago

The thing I wonder about is wtf they didn't manually switch to generator when their automatic controls failed. They had presumably ~5 minutes of UPS; it took them 40 minutes to do this. This probably isn't directly Amazon's fault, but whatever contract datacenter they are using in Europe (probably a PTT, or possibly an international carrier; really curious what facility)

I'm wary of using >1 generators to back up loads, thus requiring sync on generators for backup anyway -- much more comfortable with splitting the load up by room and having one generator per, with some kind of switch to allow for pulling generators out for maintenance. This pretty much limits you to 2-3MW per room (the largest economical diesel gensets), but that's not horrible.

Really high reliability sites actually run onsite generation as PRIMARY (since it's less reliable to start), and then utility as backup. With the right onsite generation equipment, it can be cheaper/more efficient than the grid, too (by using combined cycle; use heat output to run cooling directly).

Still, the 365 Main power outages take the cake; they used rotational UPSes (generators with huge flywheels) which had software bugs such that if input power got turned off and on several times (a common utility failure mode), the unit shut itself off entirely. Doh.

jwatte3• 15 years ago

They explained that a ground fault prevented generators from delivering power. Manual start doesn't help in that case.

rdl4• 15 years ago

From what I read, they said ground fault confused their PLCs (synchro gear for paralleling multiple generators). This shouldn't affect the generator (engine, generator) outputting power.

Electronics are much more sensitive to ground faults, etc. than mechanical and electrical devices.

A big manual transfer switch (as backup), which is presumably what they ended up using, is fairly bulletproof.

jwatte3• 15 years ago

I thought best practice backup power was to use large flywheels for re-generation, and spin up diesel engines to power the wheel in the event of a loss. That way, there is no phase synchronization issue, just a mechanical clutch. Seems like this outage could have been prevented with better gear?

o1iver• 15 years ago

There seems to be a pretty simple solution to these problems: diversification. Like most things in life, putting all your eggs in one basket it not the right choice.

The people who use only AWS or only RackSpace or only 1&1 are equally wrong.

What you have to do it diversify. Run a production site ghost on some other platform (software/hardware bugs, ...), run by some other provider (bankruptcy, theft, ...) in another country (power cuts, earthquake, ...). As soon as the primary goes down you switch on the secondary. The probability of a total blackout is then squared: 10^-3*10^-3=10^-6.

The great thing with these "cloud" platforms is that your secondary system can even "go to sleep" saving you money and then spin up instances as soon as the primary goes down. This is by the way how banks, airport-systems and probably the NSA do it!

marcamillion• 15 years ago

It seems to me that Amazon Web Services will never truly be VERY stable.

Not because I am being cynical, but just based on the nature of what they are doing.

They are the biggest provider of large scale cloud-based computing services. They are pushing the boundaries. They are bound to always come upon problems that no one has ever seen before (including themselves) just based on the very nature of their business.

So if you are looking for 'rock-solid reliability', maybe it is better to wait for another big company (Google, Apple, etc.) to come behind and fix all the mistakes that Amazon made the first time.

That being said, I use AWS and I love it. Granted, I don't use EBS (not directly, via Heroku) and yes I have encountered downtime recently, it's not that big of a deal. I know they aren't messing around, and they are in uncharted territory.

I can't reasonably expect them to have the best uptime for a platform that no one has ever built or done before, on the first time around the block. That's very unreasonable.

That being said, I will continue using them from now until I outgrow them or the economics becomes painful, because the value I get with paying for what I use far outweighs 24 - 48 hours of downtime per year.

gfodor• 15 years ago

For all those complaining about AWS I think it's important to not fall into the trap of throwing all of Amazon's services into the same bucket. EBS (and hence, RDS) have shown time and time again to be the most complex offerings and more prone to failure.

Generally speaking, at least for now, the parts of your system built on top of EBS should be carefully architected to survive in the face of erratic EBS latency, data corruption, or even downtime. (All of which are part of the standard AWS contract, but happen much more often in practice than if you are used to the mean failure time of a hard disk sitting in a cage.)

This pattern leads me to believe that services such as VoltDB that do not directly rely upon attached storage will prove to be the paradigm necessary to get reliable cloud computing, at least in the AWS ecosystem. On-demand provisioning of disk is an extraordinary hard problem, and a world where local ephemeral storage provides durability through redundancy across nodes and AZ's is probably where we are headed.

robryan• 15 years ago

This seems to share a lot of parallels with the last big outage in terms of the API request overload and the EBS replication. Seems like the system need to be able to tell a bit better between a node going down and require a remirror and most of an availability zone going down.

AWS Post-Mortem

Related Stories