A framework for formally verifying distributed systems implementations in Coq.
We recommend installing Verdi via OPAM, which will automatically build and install its dependencies:
opam repo add coq-extra-dev https://coq.inria.fr/opam/extra-dev opam install coq-verdi
To build Verdi manually, it is a good idea to first consult the
file for exact requirements.
./configure in the Verdi root directory. This will check
for the appropriate version of Coq and ensure all necessary
dependencies can be located. By default, the script assumes that
Cheerios are installed in Coq's
user-contrib directory, but this
can be overridden by setting the
make in the Verdi root directory. This will compile the
framework's core specifications and proofs, as well as some
simple example systems and their correctness proofs.
To run Verdi systems on real hardware, event handler code must be extracted to OCaml and linked with one of the shims in the Verdi runtime library that handles low-level network communication.
To install the runtime library via OPAM, use the following commands:
opam repo add distributedcomponents-dev http://opam-dev.distributedcomponents.net opam install verdi-runtime
To set up your own Verdi-based distributed systems verification project, we recommend basing it on Verdi LockServ.
Verdi LockServ contains a minimalistic implementation of a message-passing lock server and a proof that it maintains mutual exclusion between client nodes. At build time, extracted OCaml code is linked to a runtime library shim to produce an executable program that can be run in a cluster. There is also a simple script to interface with cluster nodes.
In addition to the example verified systems listed below, see the scientific papers and blog posts listed at the Verdi website.
- Core Verdi files:
Verdi.v: exporting of core Verdi theories, imported by systems
Net.v: core (unlabeled) network semantics
LabeledNet.v: labeled network semantics, for use in liveness reasoning
HandlerMonad.v: a monad for writing network/input handlers
StatePacketPacket.v: a technique for writing easily decomposable invariants
- Example systems:
Counter.v: counting server with backup
LockServ.v: lock server with proof of safety
LiveLockServ.v: lock server with proof of liveness
vard, a key-value store
- Verified system transformers:
SeqNumCorrect.v, a system transformer implementing sequence numbering
LockServSeqNum.v, the sequence numbering transformer applied to the lock server
PrimaryBackup.v, a system transformer implementing asynchronous primary-backup replication
VarDPrimaryBackup.v, the primary-backup transformer applied to the key-value store
Projects using Verdi
Verdi Raft: a verified implementation of the Raft distributed consensus protocol