I always got a chuckle out of the joke that StackOverflow devs must be the best devs because when StackOverflow goes down, they don't have StackOverflow to look at for answers, haha.
"Those who most need coffee are the least qualified to brew it." -- me, after forgetting to put the pot under the filter when attempting to make coffee in the office.
The joke is not that Google developers are so good. Rather, the code base and infrastructure at Google are not the industry standard, so googling to solve your issues wouldn't help you very much. However, Google does have its own internal search engine, Moma, that fulfills a similar role, and code search is also heavily relied upon as well. Google developers fear the day when both of these could go down.
DNS should have a mechanism for designating failover IPs, so you could e.g. host a static HTML saying "it's down" on some CDN as far as possible topologically from the rest of your network, so it displays when your status page is down too.
However, status pages are first and foremost marketing tools, so I wouldn't expect anyone setting such mechanism up.
You know, there's a proposal of treating HTTP names the same way SMTP works. That proposal includes adding low-priority servers that should be used only when the high priority fails.
All the mainstream browsers have already said they won't ever implement it.
Oh, neat! Thanks for pointing that out; I somehow got this far in life without realizing that SRV records have priority and weight fields (https://en.wikipedia.org/wiki/SRV_record has a decent summary) - I only knew that they let you specify a port, which would already be a huge advantage. Darn, now I'm even more annoyed that browsers don't handle them:(
The one rationale I know is that there is a competitive issue, where the first browser to adopt it will have slightly slower page loads, but won't gain anything yet because no site uses it.
The slightly slower page loads is one extra round trip to the DNS server (that is supposed to be close to you). It can be solved at the DNS side, and last time I looked, about a decade ago, was indeed solved at the default configuration of most open-source servers. But the large providers don't use the default configuration of their servers.
Yeah. There was someone a few years ago who said something like, "If your website doesn't have a status page, it does have one on Twitter". Wish I could find the quote source.
Likely it's hosted on the same infrastructure. The general recommendation is to host the status page on different provider so it's more likely to stay up.
Start time: June 15, 2023 | 9:00pm
End time: June 15, 2023 | 10:00pm
We are performing a software upgrade
on network equipment that will cause
some traffic to be dropped for about
1 minute, during the device restart.
I wonder which time zone. No idea where they are in the USA.
This is one of those UI conventions that would be good, but is actually unusable and ruined because there's too much other bad UIs that don't even think about time zones.
They did not discountinue it. They are uploading June data dump and it's expected that the automatic data dumps are re-enabled by end of day Friday: https://meta.stackexchange.com/a/390200
I'm probably doing my whole profession wrong. I should just shit out cookie cutter web apps and try to sucker VCs into funding them, just to make off with the dough. I could do it all from my couch now. ChatGPT webapp, ChatGPT some slides and stuff and just hire an actor to go do the pitch meeting in California for me.
One unexpected benefit of LLM (to me) is that it memorized large parts of the internet, so it has some backup like properties. SO is down, but ChatGPT memorized it already.
memorized SO, also known as "stole". when i (and everyone else) write for SO, i give them permission to use what i write, but never did I sign up for the GPT theft of that data.
You gave them permission to use it, and they used it in such a way that GPT benefits from it as training data. I'm sorry you misplaced your trust in Stack Overflow for not protecting your public writings. Knowledge is awesome, though, in that it's like a candle and by lighting a new flame, your initial candle is not diminished in any way. It's not even your sole candle but merely one in a roaring bonfire.
that's a great way to put it, I'm very disappointed they aren't the ones to innovate in this new direction, even though people gave them all the power (and data) to do it. Which they chose to give away to "Open" AI aka Microsoft and now spend their time bickering over GPT detectors and writing nonsensical CEO "vision" posts.
Basically, they really are now in a position to fizzle away at this rate, which is a shame, because I believe a community of acknowledged authors is a much more appealing way to advance knowledge than Microsoft anonymously milking the world for their sole profit and credit.
Yeah definitely. My default thought process was always, ok I'll just get a cached version, but now it's like - lemme see what GPT-4 thinks about this problem first.
One day ALL websites will go down worldwide and stay down. Could be today, could be 50 years in the future. Make sure you have a LOT of books you look forward to reading. "The Road" is not just a parable.
Interestingly, I’ve noticed a significant drop in stackoverflow results when I Google in the recent months. Can’t really attribute it to anything, but an honest word sometimes I don’t use SO for weeks.
Sometimes your separated infrastructure is not as separate as you think. True story. Northwest Airlines (prior to merger with Delta) was grounded for a couple of days in the 90's due to a backhoe cutting through their data and communication lines about 300 yards from their control center. Both vendors were using the same utility right of way. So separate systems, separate vendors, separate cables buried in the ground next to each other. One backhoe operator took them both out.
Looks like they use FireHydrant (https://firehydrant.com/) for their status page. But if FireHydrant is having issues, they'd need Stack Overflow. It's a terrible coincidence.
When Twitter was new, it was down every few weeks. It wasn't a big deal though. Turns out you can be down for 87 hours per year and still have >99% uptime.
With all that's going on around LLMs and the fact that SO has been a major source of data, when I see SO down now, I wonder if they haven't just pulled the plug. I know it's a dumb thought, I'm curious if other people feel the same way.
There's limited value if the site isn't continuing to amass new human-generated content. An "AI" that can only answer questions about technologies used until June 2023 will probably not change the world in the long term.
GPT probably generates a lot of its programming related answers based on StackOverflow. I wonder what the quality of GPT would be for future developers if SO did retire!
It's kind of surprising how relevant so many of my queries are to SO answers these days. This was kind of eye-opening. I wonder if I should start trying Bard or ChatGPT first?
I was surprised to see how many of my searches were landing on SO this morning. It stood out because the site was not responding. Apparently, it was a stupid thing to say...
I agree. ChatGPT is a really good first place to start with a technical question, so long as the tech is mature enough. I’ve even given the LLM updated docs for a library that had changed since it had run its learning, and it was able to update the example to work with the new api based on the docs.