Raft orders.
MonsterFt uses standard, unmodified Raft to place every command in one sequence.
MISSION-CRITICAL SOFTWARE
CLOUD + EMBEDDED
A LOCK HOST PROJECT / MONSTERFT
IN ORBIT AND ON EARTH
Built on world-famous Raft and SQLite
READ THE THESIS ↓A transient hardware fault can alter application data without crashing the machine. Traditional replication will keep serving that result. MonsterFt adds a deliberate checkpoint above Raft: order the work, run it independently, compare the app change + response, and accept only when enough nodes agree.
DESIGNED FOR THE FAILURE
THAT LEAVES NO STACK TRACE.
Even formally verified software can fail due to hardware.1
per 100
server-years
100-server fleet = 1 error every year.
per 100 billion
server-years
100-server fleet = 1 error every billion years.
BIGGER NUMBERS LOOK UNBELIEVABLE (WE CAN GO HIGHER). MONSTERFT WILL BE ON TOP.
THE SYSTEM
02MonsterFt uses standard, unmodified Raft to place every command in one sequence.
Each available node runs your application state machine inside a SQLite transaction and captures the resulting row changes as a patchset.
The patchset, ordered return value, and any stable application error must match.
Nodes in the agreement quorum can restore a member that produced a different outcome.
THROUGHPUT PER CLUSTER2
COMMANDS / SECOND
MONSTERFT.APPEND
COMMANDS / SECOND
MONSTERFT.APPENDBATCH
Every command in appendBatch() remains individually atomic and isolated, just as with append(). If one command fails, it rolls back independently; successful commands still commit.
THE PRODUCT MODEL
0301 / AVAILABLE TODAY
The reference implementation.
02 / AVAILABLE TOMORROW
The high-performance implementation.
THE FOUNDATION
04Lock Host is joining the SQLite Consortium to support the database at the foundation of MonsterFt.
Lock Host’s AGPL-licensed VFS collection will be validated against SQLite’s private TH3 test suite.
EARLY PARTNERS / INVESTORS / ENGINEERS