Cold restart whole system after total outage
"What are folks’ views on systems so large where cold-starting the whole system is almost impossible?"... — M'colleague, Shivam, In A Slackroom Next Door.

… When all your systems go down because data center blew up or something else happened which caused all the fancy microservices, IAM, etc etc are dead.

… I got interested in this (sic) disaster recovery methods when Elon Musk started laying off the bird app’s employees and everyone was saying twitter will go down and it will be almost impossible to cold start it."

— M'colleague, Shivam Singhal , spake further


And so of course one ends up riffing off a blog-length wall of text. Which you can read, right here…

Well… a former boss told me this story about his time at a giant US telco, back in the mid/late 1990s.

People up-high initiated some major org restructure. In the course of housekeeping, he / his team found an absolutely giant shell script (10K LoC types, IIRC).

It was called DISASTER. Yes, in all caps.

There was no documentation, obviously. So they asked around and found nobody knew anything inside the org. And then some old timer said to contact "Bill", a retired old timer to ask if he knew anything. The moment "Bill" heard that someone was planning to trash the script he yelled "Don't do that!!!".

It turned out that the script was the only thing that knew how to bootstrap the Telco's entire system from scratch in the event of a total continent-wide disaster/outage. Right from finding switches and network equipment/hardware to recover/restart, in what sequence etc.

The script was authored sometime in the 1980s.

What are folks’ views on systems so large where cold-starting the whole system is almost impossible?

I'd be surprised if there existed a system where cold-starting it is impossible.

I'd agree it would be impossible to restore it to its former shape and state.

Evolution, for example, has shown itself to be surprisingly resilient, having come back from near-zero not just once, but several times.

It's a very interesting line of questioning.

Consider what would it take to reboot civilization as we know it? It is a system, make no mistake.

One insight that bowls me over is, we would need to recover the ability to make precision screws (as in precision threading).

So much of the modern world depends on our mastery over materials (to make a precision screw, you need a precision-machined harder material—diamond / titanium—to work on a softer material—steel), and our ability to turn rotary motion to linear motion (it's stupidly difficult to reliably precision-machine a harder material without even more precise linear + rotary motion—lathe/CNC machine). Hence, a bootstrap problem.

In fact, I'd recommend looking way outside software to see how people think about recovering from catastrophic losses / shutdowns. Electric grid systems, supply chains, political succession, military planning, medicine and so forth.

One of the common lessons is to compartmentalise, so that at least something can be restored. Another is to have some sort of self-diagnosis going on so that one can do something at least while the whole thing is going down. A third is to have a seed of a backup—memory and know-how—from which to restore the atomic parts of the thing.

None of this works if there is no disaster plan, btw. Even though nothing will go as planned, it's important to have the memory and expertise that did the planning, becuase that's what's going to be able to think through the as-yet- unknown-unknowns, when the inevitable FUBAR situation suddenly happens later.

Apropos long-range planet-scale disaster planning + execution, H.E.B.'s story is one of the more interesting stories I came across:

Inside the Story of How H-E-B Planned for the Pandemic

The grocer started communicating with its Chinese counterparts in January and was running tabletop simulations a few weeks later. (But nothing prepared it for the rush on toilet paper.)

— By Dan Solomon and Paula Forbes, March 26, 2020, Texas Monthly

Another colleague in the chat remarked up-thread (apropos cold reboot thinking):

I have seen this at <Indian eCommerce Giant> and at <a FAANG>. Most of it is related to cached data. Cold starts with empty caches causes too much load on databases. And then the failures cascade.

— Another M'colleague in the Slackroom.

Not at all ironically, I've seen this at home. Thanks to the Indian Navy background, we keep semi-annually-revised "death files" at home. A very practical, not at all morbid (despite the name), response to the uncertainty of the human organism's uptime.

On the occurrence of the inevitable, one can't restore the family system to it's former state, but one can restore a lot of mechanical functionality.

The same thinking applies — backups/nominations, succession planning and transfers of ownership and control, post-hoc admin action items etc.

What are folks’ views on systems so large where cold-starting the whole system is almost impossible?

Suffice to say, I think of it as systems so complex where cold-starting is highly improbable, and restoring to former state is impossible (because, entropy).

Ask a heart surgeon trying to save a person's life. They literally cold-reboot people!

I can be convinced to argue that—barring a few exceptions, like the Internet and the WWW—the software systems we create are tame complexity-wise, compared to organic / social systems.

In abstract terms, many Google-likes and Apple-likes and Amazon-likes are possible. i.e. if you hard-stopped one of those, odds are you can re-create things that behave like those. You are guaranteed to never get back exactly Google, or Apple, or Aamzon. But you can ground-up create verisimilar entities. They've happened, in fact (Yahoo, Nokia, Baidu, Yandex, Alibaba, Samsung, etc.).

The Internet/WWW is orders of magnitude harder to recreate. But maybe also orders of magnitude harder to prevent permanently…

To prevent the Internet from happening—the entire computer industry for that matter—we'd have to go back very far in history and punch people in the face just before they got excellent insights. Like electromagnetism, fourier transforms, quantum mechanics, and so forth—and that's just the electrical engineering parts.

And it still wouldn't guarantee the prevention of some kind of planet-scale Inter-networked communication system. (Due to the fallacy of the "Great man" hypothesis.)

Tying electrical + quantum mechanical engineering back up-thread to the point about precision screws, and rotory control, hard drives present themselves as an interesting case study…

Hard Disk Drives Have Made Precision Engineering Commonplace

Modern-day hard disk drives (HDDs) hold the interesting juxtaposition of being simultaneously the pinnacle of mass-produced, high-precision mechanical engineering, as well as the most scorned storage…

And at this point I realised…

Hey, I just wrote a blog post :p :)

and stopped the wall-of-texting. But you and I can keep riffing, if you like long-running email conversations…

See discussion at HN. Several cool anecdotes and some useful critiques.