50 comments

[ 5.8 ms ] story [ 98.0 ms ] thread
Brutal. We are seeing requests that originate from heroku time out because response times are well over 30 seconds.
Same. It took over an hour for this to show up as an incident somewhere. I was going crazy.
Everything Google-related is down for me, so I can't open that page.
The issue started occurring intermittently at 08:26 US/Pacific. The issue is impacting Google’s Backbone network and may impact various services when accessing them from a different region or from the internet. Impacted services include Cloud Services (Workspace, Firebase, GCP) as well as other Google properties.

Connectivity within a zone should not be impacted.

Our engineering team has implemented a mitigation and is now monitoring the effectiveness of the change.

Hosting your status page on the same infrastructure it is reporting on is the most idiotic thing a service can do.
It's tough, though. The PR embarrassment of hosting your Google Cloud status page on AWS, say, would be substantial.
I’m sure google can afford some colo space (probably would be cheaper too)
Sure, but what’s more likely to be reliable: GCP/AWS or a colo?

It’s also embarrassing (and can cause stress for your customers) when your status page is down.

A separate failure domain / uncorrelated failure can be more important than the absolute rate of failure.
Imo for simple deployment colo with a major provider (equinix, coresite, etc) and redundant transit beats any cloud on reliability hands down
How do you know that there isn't a disaster plan under which Google routes requests for status.cloud.google.com to some other, non-Google host?
I honestly have no idea, hope they do. My point is aws is not the only alternative (and not even the best one) for setting this up.
Makes me think of a calculus word problem: As the things Google cannot afford approaches 0...
(comment deleted)
Seems like the status page was on separate infra (I could access it while the GCP services were down), but Google Public DNS (8.8.8.8) was also down.

Perhaps alexdumitru was using 8.8.8.8?

(comment deleted)
All Google services have been down for me for the past 30 minutes, at least.
Miami: Google services (Mail, Drive, Calendar, Maps, DNS) available to me.
Sweden: Google's public web-based services all seem to be available, GCP services are not. Except the console.

Edit: things started working now.

Try not using 8.8.8.8, it's down from some places.
My home internet connection is down, maybe some larger scale backbone problem?
OVH last Wednesday, Oracle the very next day, Azure yesterday, GCP today... will a crime syndicate employing arsonists as well as hackers claim all these in a month as a demonstration of their capabilities?
These comments crop up regularly on outage posts. There's probably some psychological bias that explains it but I can't remember... I think the last time Google went down there were comments blaming it on Summer Interns :)
No, it was SolarWinds that blamed the intern for their security issues.

In all seriousness... if a single employee can accidentally single-handedly screw things up you've got a process problem. There should be checks and balances to make sure that what you're trying to do make sense. These might be in the form of a code-review, automated validation, a staging environments, etc. Yes, these might be things that don't exist at a scrappy startup and that is fine as they'll get there eventually but for any large company not having good processes is reckless.

These issues happen despite all those things. You can bet that Google has that and more in place. Just like in aerospace engineering, things go wrong when dozens of failures happen across the critical path, not any individual piece.
Yes, and issues will always occur. What I was trying to point out that it's never an individual's fault alone. Sure someone might have tripped the final domino but they might not have had anything to do with everything else that happened. That's why blameless postmortems are a thing. Blameless (to me) doesn't mean that you don't name any person but rather that there's no blame assigned to anyone. I want to know what Person A did to cause the incident but there shouldn't be any retribution against Person A.
Anybody know the current where abouts of Mr. Robot?
(comment deleted)
Does Cloudflare use Google Cloud? I know they have a ton of their own hardware, but maybe they outsource some things?

I had my first ever Cloudflare outage earlier today. It's just a free site that has a single page rule that redirects to a different domain, but it started refusing connections for a little while today.

Why the most GCP outages are global? Even though it says issue started in US/Pacific, I'm not sure if it affects other regions and the status page of them does not make it clear.
they aren't, everything from germany works.
Because Google engineers think that they are the smartest people in the world which means they don't actually engineer for minimization of a blast radius.

You should just see how Google builds the prefix lists that it allows customers to advertise to Google via PNIs. You would think they are building it off some registry lists because why else would Google insists that the routes are registered? Oh no, that's for small people. At Google they just bring up a session to you, let you advertise and take whatever you advertised to them during the setup as the allow list because everything is done via custom automated software!

"US/Pacific" is detailing the timezone they are using for start/end times. It's not saying the issue is only in that location.
Seems to be resolved according to the same source.
> Workaround: None at this time.

I wish I could say that to my customers...