Musk said they fended off a DDOS.
He can safely get around being truthful because an organic user traffic outage would look exactly like a DDOS to an end user. He had the same outages when he brought Desantis on (last year? two years ago? I forget). It's not like he'll ever release a post-mortem, we'll never see if Libtard420 (/s) spun up a million boxes from a digitalocean cidr block and nuked twitters video endpoints vs. twitter just getting slammed globally once it started.
He fired all of the previous (great) engineers at twitter. I find it much more probable the systems just went down due to load.
This is my job for a big advertising company that does far more load than twitter ever did. We scale on load. We'd be a failure if we had to cap the audience that saw our advertisements. We do something like 20-40x (tens of billions req a minute) the amount of load on Black Friday/Christmas every year and don't even need to tweak anything.
IIRC Musk brought everything from the "cloud" (twitter was AWS iirc) to I guess some in-house datacenter where they couldn't quickly scale as they're dealing with physical hardware themselves. I'm not sure if they even have a CDN in front of them nowadays. I quit paying attention to their tech. IIRC their video service used to be provided by another vendor that Musk wound up getting in trouble with because he stopped paying them or something along those lines. I believe they stopped using that service?
edit: Oh, Periscope. https://www.businessinsider.com/elon-musk-twitter-building-l...
Musk claimed they tested with 8M, but the actual attendance was around 115k iirc. It makes me wonder if they tested with internal connections and not through the full production network path in a truly global distributed setup including network and packet failures.
The other thing to consider is the history of Spaces, notably the DeSantis incident which went poorly as well. It is notoriously a low quality audio experience.
I've worked for some large CDNs that have video services and I have no idea how I would test 8 million actual video streams at once. I mean, I have a general idea of how I could pull that off but I have absolutely never tested anything anywhere near to that degree and I don't think I could get every single cloud I have available to provision that many containers/serverless/lambdas to even test that many at once.
There was a point in my life where I had to deal with Lady Gaga (et al) releasing new music videos on Vevo.com which was probably the biggest load I've seen on any service and we absolutely did not stress test millions of streams prior to a release. I'm pretty sure we didn't even know they were releasing until I walked in that day.
AFAIK Twitter/Musk brought all of their systems in house so they'd be severely limited on hardware scaling because they lost aws/gcloud/etc scale speed.
It's absolutely possible to do this, I'm not saying it isn't, it just seems like a pain in the ass logistical nightmare so I doubt they did it. What would you do? 8 million HLS/RTC/TSP processes spread globally to not hit ulimits/socket limits/edge cases that are doing a single video stream GET and holding the connection open?
a. Run a number of Vegetas or similar
b. hire a bunch of yes men who will tell you tall tales