New Jepsen analysis is out! We talk about etcd's key-value ops, transactions, watches, and locks. Things look pretty good: KV ops look strict serializable. Watches deliver all changes in order, though be careful about passing revision 0. Locks, however, are fundamentally unsafe, and failed to provide mutual exclusion immediately in our testing. No surprise there: distributed locks (in the sense of mutexing execution) are impossible in async networks.
My uncle (111 M) gifted me (33 M) a ring before leaving to go travelling. A close family friend (2019 M) told me to destroy the ring due to problematic associations with the jeweller who made it, but the ring is precious to me and I would feel guilty throwing it away. AITA?
I think in a couple decades we’ll look back and will think that IT/infosec professionals barely deserved the name, just like doctors before it became evidence based.
Instead of these gradients, I’d be surprised if there is more than 1-5% that we can call progress and the rest being utter quackery the current practitioners accept as normal.
We are in the era where we’re arguing that maybe doctors should wash hands (=software updates) and before randomized double-blind trials.
Infosec and IT in general is still in the era that would be the equivalent of pre germ theory in medicine.
It’s still a wild west when it comes to so many things, memory safety is not yet a thing, hardware microarchitectural bugs are just starting to get serious/recognized, we as a society have no idea what to do about data, machine learning and privacy.
People assume infosec is this gradient with some bad, some average and some good parts distributed in a fairly uniform way. 1/2