From the security/production-readiness review (https://github.com/v-code01/ledge/blob/main/AUDIT.md §F-6, R-1).
Finding: the sharded-Raft layer has only been exercised on a single host with emulated WAN + clock skew (soak/wan-chaos.sh, 16/0); never on real separate machines.
Why it matters: a consensus system proven only on localhost is not proven for production (real clock skew, NIC faults, geo-latency, partition shapes the emulation can't reproduce). Biggest blocker to depending on a multi-node cluster.
Closes it: deploy 3+ nodes across real instances (multi-AZ/region), run the chaos/skew suite there, confirm leader-failover-no-data-loss + no split-brain + reconvergence. Severity: High.
From the security/production-readiness review (https://github.com/v-code01/ledge/blob/main/AUDIT.md §F-6, R-1).
Finding: the sharded-Raft layer has only been exercised on a single host with emulated WAN + clock skew (
soak/wan-chaos.sh, 16/0); never on real separate machines.Why it matters: a consensus system proven only on localhost is not proven for production (real clock skew, NIC faults, geo-latency, partition shapes the emulation can't reproduce). Biggest blocker to depending on a multi-node cluster.
Closes it: deploy 3+ nodes across real instances (multi-AZ/region), run the chaos/skew suite there, confirm leader-failover-no-data-loss + no split-brain + reconvergence. Severity: High.