mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
663cc83ca3
Companion LaTeX papers to luxfi/LPs: - lp-200-zap-stack: full ZAP-as-substrate paper (10-layer model, deco- mplected concerns, performance). - lp-200-lux-vs-avalanche-diff: code-translation appendix. - lp-203-gpu-native-verify, lp-205-pipelined-quasar, lp-208-dag-mempool, lp-209-mysticeti-ordering, lp-210-block-stm, lp-211-cross-shard-atomic, lp-217-cert-profile-modes, lp-218-zap-p3q-pq-rollup, lp-220-p3q-corona-magnetar, lp-221-stark-fri-verifier, lp-300-schema-registry. Pitch / outreach decks (luxcover.sty templated): - blackrock-nahmii-pitch, corporate-dao-stack-analysis, nahmii-patent-acquisition, pq-patent-family-future-infringers, lux-substrate-platform-pitch, w3a-foundation-membership, multi-validator-scale, vfs-archive-cost-study. Edits aligning to LP-217 / LP-200: - lp-020-quasar-consensus 02-system-model + 08-performance. - lp-073-pulsar 11-quasar. - lp-103-omnichain-explorer 03-chain-support. - lp-104-graph-engine. - lp-201-zap-universal-wire/wire-inventory.csv. - lp-306-lx-social-dex sections/02. - lux-adoption-roadmap, lux-chain-architecture, lux-consensus-engines/08-composition, lux-crypto-agility, lux-evm-precompile-engineering/02, lux-fhe-smart-contracts/05, lux-quasar-composition/04+06+10, lux-reproducible-builds. - INDEX.md cross-links the new papers.
42 lines
2.5 KiB
TeX
42 lines
2.5 KiB
TeX
\section{Failure Modes}
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\small
|
|
\begin{tabular}{@{}p{3.5cm}p{3cm}p{4cm}p{3.5cm}@{}}
|
|
\toprule
|
|
\textbf{Failure} & \textbf{Trigger} & \textbf{Effect} & \textbf{Recovery} \\
|
|
\midrule
|
|
Leader missed (no header at round $2k$) & leader offline, partition, equivocation suppressed by reputation & wave skips; round $2k{+}1$ votes do not anchor & next leader at $2k{+}2$ anchors a 4-round wave covering $2k..2k{+}2$ \\
|
|
Leader equivocates (two distinct headers at $2k$) & byzantine & DAG retains both; neither receives $\geq 2f{+}1$ references (Mysticeti slashing) & wave skips; next $2k{+}2$ leader anchors \\
|
|
$<2f{+}1$ references on leader vertex & partition, slow validators & fast-path quorum not met & 3-RTT recovery: round $2k{+}2$ votes for round $2k{+}1$ votes for round $2k$ \\
|
|
Cert leg fails to aggregate (LP-217) & insufficient threshold sigs & cert tier downgrade per LP-202 & relying parties at the mode wait or fall back; consensus liveness preserved \\
|
|
LP-210 execution fails on a wave tx & sig invalid, balance underflow & per-tx MVCC rollback; other txs commit & invalid tx dropped; commit proceeds \\
|
|
Partition splits validators & dual halves cannot reach $\geq 2f{+}1$ & both halves stall at anchor decision; no double-commit (Mysticeti safety) & partition heals; quorum-side advances \\
|
|
Reorg at committed wave & impossible & n/a & Q-Chain finality (LP-079) + LP-217 cert PQ-strict$+$ is content-addressed; no alternative ordering produces same cert \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Failure modes and recovery.}
|
|
\label{tab:failures}
|
|
\end{table}
|
|
|
|
\subsection{Safety property}
|
|
|
|
The Mysticeti safety property holds: at most one wave anchor is
|
|
committed per even round across the entire honest quorum. Mysticeti
|
|
\cite{mysticeti2024} proves the safety theorem in the partial-synchrony
|
|
model; LP-209 inherits the proof, with the only adaptation being that
|
|
the cert ceremony following the ordering is the LP-182 QuasarCert
|
|
rather than a single-leg BFT cert.
|
|
|
|
\subsection{Liveness under partition}
|
|
|
|
Under network partition $> 1/3$ stake, the wave anchoring halts:
|
|
neither half can produce $\geq 2f{+}1$ references on the leader
|
|
vertex. When the partition heals, the half that resumes with
|
|
$\geq 2f{+}1$ advances; the other half catches up via the LP-202
|
|
bootstrap pipeline. No double-commit is possible because each
|
|
honest validator can be on only one side of the partition, and the
|
|
$2f{+}1$ quorum cannot have members from both sides simultaneously
|
|
in a valid validator-set partitioning.
|