Yesterday we witnessed "Decentralization" fails
—————
Pyth Oracle was offline for 25 minutes. exposing a fundamental weaknesses in the current approach to oracle networks & decentralization.
The trigger? Wormhole - the messaging protocol Pyth relies on - went down completely.
—————
What we know:
--> Pyth Oracle offline for 25 minutes
--> Wormhole Guardian network failed
--> All 19 Guardians failed simultaneously
--> No post-mortem published yet
—————
While the crypto community speak about oracles provision of wrong price updates, we're missing an equally dangerous threat: liveness failures. When oracles stop updating, the entire DeFi stack built upon them becomes vulnerable.
The conventional thinking suggests that using external providers like Wormhole creates single points of failure.
But I believe this misses a deeper truth - even internal solutions can fail just as catastrophically when they share correlated failure modes.
—————
from Wormhole's docs:
—————
Well, why do "decentralized" networks often fail as one unit?
-> Shared Infrastructure: Many nodes using the same cloud providers or software
-> Cultural Homogeneity: Projects starting centralized create uniform operational practices (in oppose to utilising Ethereum’s diverse validator set through
@eigencloud)
-> Hidden Dependencies: Seemingly independent components sharing critical dependencies
-> Communication Patterns: Excessive coordination leading to synchronized behavior
—————
Two approaches emerge for creating truly uncorrelated operator networks for oracles:
1. Incentive design: such as Operator slashing
2. Ensuring Isolation of nodes: Using behaviourally and technologically diverse node operators (like
@eigencloud)
The second approach is hard because It requires:
> Complex coordination
> Higher operational overhead
> Dealing with diverse technical capabilities
>Managing multiple social relationships
But when correlation failures strike - and they will strike again - this difficulty pays for itself many times over.
—————
Pyth/Wormhole incident isn't just about one protocol's failure - it's a reminder for an industry which tends forget why decentralization matters.
We're building "decentralized" systems that are decentralized in name only, missing the core purpose: resilience through genuine independence.
The path forward isn't easy, but clearly necessary. We need to shift our focus from superficial decentralization metrics to deep structural independence.
This means EMBRACING the complexity of truly diverse systems, even when it seems inefficient, difficult or just not the most financially rewarding in the short-term.
Because in the end, a catastrophically cost of correlated failure is higher than the cost of preventing it as we’re still in day-1