Note: experienced it from the first row, cloud had an issue in a region with SSL...
Effectively unless they're down for more than a day, you're better off taking the hit and waiting for them to resolve everything.
lol. There's me worrying about a 2 minute downtime once every 10 years
I'm taking the piss too. Care to explain?
PS For a laugh (and I apologize for going dreadfully off topic), I asked ChatGPT a couple of questions regarding two mins and 10 years, just in case I'd missed a trick with your comment. I got two calculations within both answers that looked the same but have a factor of 10 difference in the result!
https://chat.openai.com/share/7aaa80e5-1c54-4ac2-9ddb-9e488d...
This is where it goes wrong:
"Total minutes in 10 years = 60 minutes/hour * 24 hours/day * 365 days/year * 10 years = 525,600 minutes"
"Total time in 10 years = 60 minutes/hour * 24 hours/day * 365 days/year * 10 years = 5,256,000 minutes"
LO Calc says that =6024365*10 = 5,256,000.
The worrying thing for me is that an awful lot of input training data might be badly wrong to cause this result or perhaps I've managed to excise a corner case in which case the training data is a bit too focussed in this particular regard. I suspect that arithmetic errors for these things will be awful because there are so many ways to screw up and the training data will have a lot of errors in it. Combine that with the number of subjects available and it will be a shit show.
As for GPT, it's not about the training data having errors in it - GPT doesn't parrot the training data exactly, there's randomness built into it (otherwise you'd always get the exact same output to the same input). It just generates a random plausible response, it doesn't actually know math.
To highlight this, I've just asked it the exact same thing you did, exact same words, and got an entirely different response (and this time it was correct): https://i.28hours.org/20230613-051307-0ae6.png
If DNS is up, you can change specific records to point to your origin or another provider instead of the proxy/CDN, provided that it can handle the load and doesn't need a setup similar to Cloudflare (i.e. repointing your nameserver).
?
But that doesn't mean Cloudflare is built in a centralised way (I assume it isn't), nor that there's something about the internet that is centralised around Cloudflare. Rather, people are choosing to include it as a dependency in their stack.
If all those websites are using Flask, that would be the equivalent of centralising on Flask. The opposite of that would be many Flask-compatible yet unrelated other frameworks being used in parallel. A bug in Flask would not affect those not using that very codebase.
The centralisation people are speaking of here is the amount of people all putting their eggs in Cloudflare's basket.
I agree on the equivalence between this and Cloudflare use, but not that this should be described as centralisation, which is a particularly potent word on the open web. Popularity isn't the same as centralisation. Each website could rewrite to remove Flask if they wanted.
Another example: if lots of people watch Squid Game, that doesn't indicate a centralisation of television programmes. No options are removed. People are just choosing to do a similar thing, but not in a way that centralises anything.
Here with Cloudflare, a single entity is responsible for the fix and will more or less fix the failure for every sites that uses it at the same time, by fixing it on their side. And website individually cannot do anything on their own.
So I would argue it makes sense to call that centralized, at least from a structural/operational perspective.
It is at the very least a form of contraction of the network.
> It is at the very least a form of contraction of the network.
I think it's that at most. No one has to use them, as they accelerate / enhance open protocols. That is the least lock-in one could hope for, so they don't contract anything in a negative way.
Contrast with, say, Etsy/Shopify, who actively try to replace the open space with closed ones.
Is a monoculture centralized? It doesn't have to be, and yet it can all fail due to a single fault.
If Cloudflare goes down, so does a significant portion of the web. Flask cannot have an outage like Cloudflare. All they can do is push a faulty update, but even then you can rollback or stay on the old version.
This is what was on the tip of my tongue. I'd argue you're missing the portion of control in this whole discussion, and how much process you can place in front of a component change.
If I have a flask dependency, I have a lot of control over this dependency. If flask screws up badly, I have many options: I can update. I can not update. I can downgrade. In fact, I could fork flask internally and fix it on my own and either be a good citizen and open up a PR, or I could be something else. I can test all of this in any number of environments before it hits a customer, and even more stages before it hits all customers.
With Cloudflare - or any number of Hosters as well, like AWS, Azure, Google Cloud, I have very little control. If I use Cloudflare as a CDN and Cloudflare goes down, I might not have the capacities at my upstream server to handle the load from all my customers, so I am down as long as Cloudflare is down. I wouldn't have the footprint available necessary to replace AWS in our own private DCs, even if we pooled all spare capacities - and then I'd still have to find a way to exfil our data from a downed AWS. (Which, yes, we have, but it'll take long hours)
And no matter how much I test, if my hoster fucks up, I'm immediately fucked as well, no matter what processes I might have. The only process around this would be provider independence, which is really expensive and a lot of effort even if you just have a luke-warm standby.
And you can do these things in parallel, unlike the Flask example, which you pretty much have to commit to using solely.
Remember when crypto promised a decentralized utopia of distributed systems that any interference wouldnt work, but slowly but surely it all congregated into a couple of oligarchical companies?
Surely by now the technology of the internet has matured enough that these conglomerates are going to do the same but with a bit less fraud involved.
BGP was standardized in 1989 and has been in use since 1994. The technology of the internet has matured in many ways, and remains almost identical in many others. Sometimes Microsoft is right, and backwards compatibility is the killer feature.
Stop putting every-bloody-thing behind Cloudflare, and the problem solves itself. I don't know whether to laugh or cry when I read of someone on HN seriously saying that they need a CDN for their personal website, or that they really need to use AWS or GCP for that matter. We tech workers have lost the plot, and it's our fault it's all in the hands of the few.
Cryptocurrencies promised many things but only wanted to replace oligarchical companies with oligarchical miners.