[origami777]

I realize that some systems may need to have all of their servers located close together in a single AZ. But barring that, if this took you offline, you should really consider spreading your instances across AZs. It's so easy there's no excuse not to do it.

Another thing to look into is EC2 Auto Recovery [1]. I don't know if this would've kicked in with today's event, but it's worth setting up as an extra safety net.

[1] https://aws.amazon.com/blogs/aws/new-auto-recovery-for-amazo...

edit: I'm basing this off the status page which indicated that only one AZ was impacted.

[rnhu]

The site I manage is load balanced across both AZs ap-southeast-2a and ap-southeast-2b which did not save it. At the moment ec2 statuses are not being updated which is preventing ELBs from registering instances as healthy.

Both AZs are directly under the deluge and I don't believe only one AZ is affected for a second.

The size of the storm can be seen here http://www.bom.gov.au/products/IDR713.loop.shtml#skip

[origami777]

This is the most concerning thing to me. The Multi-AZ, redundant setup is worthless if the ELB can't do its job properly. I've seen some funky behavior from the ELBs when it comes to instance state. They really need to make this better.

[PebblesHD]

Sadly our use case (private data etc.) prevents us from leaving the local availability zone, meaning when it went down today we were left totally unavailable. The recovery itself is ongoing but our applications are resilient enough to detect the restored connections and automatically add themselves back into the cluster.

[NeutronBoy]

Availability zones are different from regions. You can be in multiple AZ's within the Sydney region still.

[origami777]

That's interesting. Is it an Australian regulation? Curious that they'd make it in such a way that the data had to reside in the same building/zone.

[PebblesHD]

Indeed it is, it was a massive struggle getting approval to move into a cloud service in the first place.

[inopinatus]

I almost hate to point this out, then, but you did consider that there's no guarantee that an AZ is a single DC, right?

[Twirrim]

It's pretty much guaranteed not to be the case.

[ageofwant]

Really ? Pointing the local gov department's officer to AWS's IRAP compliance cert was all that was needed to move quite a lot of their stuff unto AWS.

[inopinatus]

Yeah, and I'm curious about which sector or agency is the culprit here. Even APRA (the financial regulator) are cloud-friendly now, if you engage them at the start of an adoption process. My wild guess is health insurance, being a sector where IT is notoriously hidebound, but it could just be a case of overzealous/interfering/uncomprehending lawyers. A security policy that precluded cross-site service or data replication would likely be in contradiction with DR/BCP plans.

The classic irony for me was a service manager in just such an environment resisting a cloud move "because it's someone else's computer" - even though his (ancient) application was running on a rented partition of a remote, IBM owned & operated S/390...

No surprise therefore that the big clouds have country resources dedicated to moving the needle on cloud awareness in highly regulated environments.

(obdisclosure: I am former .au AWS manager)

[technion]

    lawyers

I've supported multiple legal firms who have assured me they cannot legally host their data in the cloud.

Noone ever seems to be able to refer to a specific law, but then, it's an IT person talking to lawyers, so there are some battles you just don't fight.

[Khaine]

I would not call APRA cloud friendly. Systems of record can not be in the cloud, and I don't know of any bank that is actually storing data in the cloud