Being only somewhat familiar, I wonder this: Why don't clients have a simple rule against accepting changes to history that are more than one or two blocks old?
Philosophically, what the distributed blockchain is, is an agreement for adjudicating whose record (eg list of transactions) should be agreed on as "the valid one". If you already have a different, reliable external way of deciding this question, that is (arguably) conceding that you didn't need a distributed blockchain in the first place.
EDIT: I've realised my comment could be misinterpreted as endorsing the blockchain 'movement'. I am among those who are quite skeptical of these many blockchain projects.
If anything, relying on an outside source is what we're doing now. Coinbase and others are telling us that they detected something and I haven't heard anybody question them. (And I'm assuming they're not lying). Though that's not to say the blockchain will be changed by this information.
You are correct of course. To me, problems like this recent attack highlight an inherent contradiction in the blockchain idea
- that the "51%+ of hashing power determines the truth" rule is really either lip-service; techno-marketing vanity that doesn't actually translate to practice (if corrective action ensues), and in which case the adjudicator is clearly not the blockchain tech we were told, but instead a nebulous collection of companies and stakeholders OR
- that "51% determines the truth" is indeed a strictly-held principle even though we have already demonstrated that there exist actors with enough resources and motivation to make attacks of this nature on some of the most prominent blockchain projects already - and we haven't even yet witnessed what a 51% attempt by an actor with the resources of a state would look like.
This is one of the reasons I have become skeptical of blockchain projects.
There's always dilution when an idea has been around long enough and I'm concerned about the copycats trying to use blockchain for other things.
If Raft is democracy, then blockchain is Might Makes Right. The guy with the biggest stick always wins. There is no actual vote. Everybody is an accomplice.
That doesn't sound like progress, that sounds like regression. A really big one.
Also, if you have clients follow the rule "never switch forks" to protect against 51% attacks, then any network partitions will cause a permanent fork.
I'd imagine that you would want to look for evidence of a double spend as that's the most likely goal of a 51% attack. But then the blockchain must have up-to-date knowledge of likely double spend targets (such as exchanges, OTC desks, etc), and be able to algorithmically and deterministically prove malicious intent with high certainty. Only then will you be able to maintain consensus and prevent unnecessary or accidental forks.
But since this is all open-source anyway, it would only be a matter of time before a slightly more sophisticated attacker read through the updated consensus algorithm and figured out how to game it. And so the cycle continues.
In truth, the only real strategy for mitigating attacks in PoW blockchains is hash power. It has proven to be very effective if you have enough of it (see BTC), and looking for other 51% resistance measures isn't really that productive unless you start from the ground up and rebuild the consensus mechanism on a different paradigm (e.g. PoS, which is still unproven afaik).
It is a pretty interesting solution, that seems to work out OK.
For example, most Bitcoin cash clients don't accept reorgs longer than 10 blocks, as such a reorg is 100% an attack. This protects the network against large reorgs.