Wisqnet

Cripes, people. WHO CARES?

So I SSHed in. I got the message. Then I saw all the people freaking out here and couldn't believe my eyes.

Ask yourself this:

How often do you SSH in to arbitrary hosts? Ones that you don't control, or work for, or trust with your source code?

Did you really expect that you could give the same long unique base64 string to a bunch of different hosts and NOT have them connect your identity between them?

Do you not understand that the whole point of public keys is to uniquely and reliably identify yourself to an arbitrarily large number of parties?

Honestly, the only real eye-opener here is that Github gives out your public keys and can be scraped to collect all of them. But some of us have been using that functionality to share / snag each other's public keys for a long time now anyway.

Shopify itself is written in Ruby using the Rails framework. I believe we're one of the oldest and largest Rails projects out there.

Some parts of our infrastructure are written in other frameworks (e.g. Thin) and languages (e.g. Go).

I'm honestly not really seeing the fuss here. Doesn't every company want "really really really good programmers"? Don't most companies offer stock options? Am I really supposed to see a lack of info about how hard they work and assume the worst?

If there were some inside knowledge from people who have taken these offers and lived to regret doing so, then maybe this would have some substance. To me, without more info and being willing to give people the benefit of the doubt, it just seems like speculation and pessisism.

The problem is that DSL around here is PPP over Ethernet, rather than just plain Ethernet. So there's no MII, no ARP, no regular ethernet link bonding. It's hard to escape PPPoE for DSL around here due to phone company politics.

The type of bonding I talked about in the article was all PPP link bonding, which apparently still has a lot of issues to work out in Linux. Our two-line bond was pretty unreliable, and our attempts at a three-line bond ended up with a lot of return packets going nowhere. Didn't matter which lines we picked, either.

The problem was that all cabling had to be done underground at our place, and phone lines were the only thing we had in abundance. Coax, fibre, anything else basically meant busting up a path to the road in a major urban area. Plus it was the middle of winter which makes road work even harder.

Not every internet provider realised this immediately. We started down the path to better internet a lot of times with companies that said "no problem, we can get a wire to your building just fine" -- and then backed down weeks later when they saw our situation.

After one or two of these delays, we applied the shotgun approach, asking every single internet provider to give us an estimate all at once, to make the delays concurrent rather than successive. Every last one came back negative, or just stopped returning our calls(!). So Potato was my response to our complete lack of options.

The multiple IPs thing was an issue for some sites, but it turned out to work fine in general. If there was a site that really had a problem with it, it was also typically a low volume site, and didn't need heavy load balancing. So we had one of our proxies set up to not be balanced and just pick a single working link, and used that for those sites.