Babble is a distributed consensus engine designed to easily plug into any application. It uses peer-to-peer networking and a consensus algorithm to guarantee that a group of connected computers process the same commands in the same order.
-
Asynchronous: Participants have the freedom to process commands at different times.
-
Leaderless: No participant plays a special role.
-
Byzantine Fault-Tolerant: Supports one third of faulty nodes, including malicious behavior.
-
Finality: Babble’s output can be used immediately, no need for block confirmations.
-
Dynamic Membership: Members can join or leave a Babble network without undermining security.
-
Fast Sync: Joining nodes can sync directly to the current state of a network.
-
Accountability: Auditable history of the consensus algorithm’s output.
-
Language Agnostic: Integrate with applications written in any programming language.
-
Mobile: Bindings for Android and iOS.
-
WebRTC: Supports WebRTC connections for practical p2p connections.
We use an adaptation of the Hashgraph consensus algorithm, invented by Leemon Baird, to which we added important features. Hashgraph is best described in the white-paper and its accompanying document. The original Hashgraph algorithm is protected by patents in the USA, so anyone intending to use this software in the USA should take this into consideration. For a high level overview of the concepts behind Babble, please refer to this document.
Babble's major departure from the original Hashgraph algorithm is the introduction of blocks, which represent self-contained sections of the Hashgraph, and which are instrumental in the implementation of two important new features that were alluded to in Baird's paper, but not specified:
-
A dynamic membership protocol, which enables peers to join or leave a group on demand.
-
A fast-sync protocol which enables joining nodes to fast-forward directly to a point in the hashgraph without downloading the entire history.
Babble communicates with the App through an AppProxy interface, which has two
implementations:
-
InmemProxy: An InmemProxy uses native callback handlers to integrate Babble as a regular Go dependency. -
SocketProxy: A SocketProxy connects to an App via TCP sockets. It enables the application to run in a separate process or machine, and to be written in any programming language.
Refer to the dummy package for an example that implements both proxies.
// Start from default Babble configuration.
babbleConfig := config.NewDefaultConfig()
// Create dummy InmemProxy
dummy := NewInmemDummyClient(babbleConfig.Logger())
// Set the proxy in the Babble configuration.
babbleConfig.Proxy = dummy
// Instantiate Babble.
babble := babble.NewBabble(babbleConfig)
// Read in the configuration and initialise the node accordingly.
if err := babble.Init(); err != nil {
babbleConfig.Logger().Error("Cannot initialize babble:", err)
os.Exit(1)
}
// The application can submit transactions to Babble using the proxy's
// SubmitTx. Babble will broadcast the transactions to other nodes, run them
// through the consensus algorithm, and eventually call the callback methods
// implemented in the handler.
go func() {
dummy.SubmitTx([]byte("the test transaction"))
}()
// Run the node aynchronously.
babble.Run()
// Babble reacts to SIGINT (Ctrl + c) and SIGTERM by calling the leave
// method to politely leave a Babble network, but it can also be called
// manually.
defer babble.Node.Leave()Babble configuration is defined in the config package.
Babble reads configuration files from its data directory which defaults to
~/.babble on Linux. It can be overwritten with DataDir in the Config object
or --datadir from the CLI.
Every Babble validator requires a cryptographic key-pair to encrypt, sign and verify messages. The private key is secret but the public key is used by other nodes to verify messages signed with the private key. The encryption scheme used by Babble is ECDSA with the secp256k1 curve (like Bitcoin and Ethereum).
To pass a private key to Babble, either set it directly in the Config object,
or dump it to a priv_key file in the data directory. Babble's keygen command
may be used to generate key-pairs in the appropriate format.
Babble needs to know the other peers in the network. This is specified by adding two JSON files in the data directory.
-
genesis.peers.jsoncorresponds to the initial validator-set; the one that the hashgraph was started with. Ifgenesis.peers.jsonis not provided, Babble will usepeers.jsonas the genesis validator-set. -
peers.jsoncorresponds to the set of peers that the node should attempt to connect to upon starting.
peers.json and gensesis.peers.json are not necessarily equal because
the dynamic membership
protocol enables new nodes to join or leave a live Babble network dynamically.
It is important for a joining node to know the initial validator-set in order to
replay and verify the hashgraph up to the point where it joins.
It is possible to start a Babble network with just a single node, or with a
predefined validator-set composed of multiple nodes. In the latter case,
someone, or some process, needs to aggregate the public keys and network
addresses of all participants into a single file (peers.genesis.json), and
ensure that everyone has a copy of this file. It is left to the user to derive a
scheme to produce the configuration files but the docker demo
scripts are a good place to start.
To join an existing network, a peer must first obtain the JSON peers files from
an existing node and place them in the data directory. One way to obtain the
peers files is to query the /peers and /genesispeers functions exposed by
the HTTP API service. Please refer to the
join script in the demo for an example.
for an example.
Implementations of the Transport interface determine how nodes communicate with one-another.
The TCP transport is suitable when nodes are in the same local network, or when users are able to configure their connections appropriately to avoid NAT issues.
To use a TCP transport, set the following configuration properties:
-
BinAdddror--listen: the IP:PORT of the TCP socket that Babble binds to. By defaultBindAddris127.0.0.1:1337, meaning that Babble will bind to the loopback address on the local machine. -
AdvertiseAddror--advertise: (optional) The address that is advertised to other nodes. IfBindAddris a local address not reachable by other peers, it is necessary to setAdvertiseAddrto something else. IfAdvertiseAddris not set, it defaults to theBindAddr.
For example, when running a node from a local network behind a NAT, the
BindAddr might be 192.168.1.10 which is not reachable from outside the local
network. So it is necessary to set AdvertiseAddr to the public IP of the
router, and to setup port-forwarding in the NAT.
Note that the advertise address (which defaults to bind address if not set) must
match the address of the peer in the peers.genesis.json or peers.json files.
Because Babble is a peer-to-peer application, it can run into issues with NATs and firewalls. The WebRTC transport addresses the NAT traversal issue, but it requires centralised servers for peers to exchange connection information and to provide STUN/TURN services.
To use a WebRTC transport, use the following configuration properties:
-
WebRTCor--webrtc: tells Babble to use a WebRTC transport. -
SignalAddror--signal-addr: address of the WebRTC signaling server. -
SignalRealmor--signal-realm: routing domain within the signaling server. -
ICEAddressor--ice-addr: URL of a server providing ICE services such as STUN and TURN. -
ICEUsernameor--ice-username: Username to authenticate to the ICE server. -
ICEPasswordor--ice-password: Password to authenticate to the ICE server.
Users of the library can also manipulate the ICE servers configuration directly
by manually modifying the list returned by Config.ICEServers().
WebRTC requires a signaling mechanism for peers to exchange connection information. This requires a central server, so when the WebRTC transport is used, Babble is not fully p2p anymore. That being said, all the computation and data at the application layer remains p2p; the signaling server is only used as a sort of peer-discovery mechanism. We povide the code for a signaling server here. The demo has a WebRTC option that illustrates the usage of WebRTC.
It is not necessary to specify network addresses in the JSON peer files when
WebRTC is enabled because this information will be exchanged over the signaling
server. Likewise, the BindAddr and AdvertiseAddr options will be ignored.
The default ICEServers points to a public STUN server hosted by Google
(stun:stun.l.google.com:19302). It does not include a TURN server, so not all
p2p connections will be possible. For a full ICE server, have a look at our
Disco server.
We can choose to run Babble with a database backend or only with an in-memory
cache. With the Store (--store) option, Babble will look for a database in
datadir/babdger_db or in the path specified by DatabaseDir (--db). If the
database already exists, and the Bootstrap (--boostrap) option is set, the
node will load the database and bootstrap itself to a state consistent with the
database and it will be able to proceed with the consensus algorithm from there.
If the database does not exist yet, or the Bootstrap option is not set, a new
one will be created and the node will start from a clean state.
The node can also be started in maintenance-mode with the homonymous flag. The
node is started normally but goes directly into the Suspended state, where it
still responds to sync-requests, and service API requests, but does not produce
or insert new Events in the underlying hashgraph. The Suspended state is also
triggered automatically when more than suspend-limit, multiplied by the number
of validators, undetermined-events were created since last starting the node.
This is a safeguard against runaway conditions when a network does not have a
strong majority and produces undetermined-events ad infinitum.
We can also specify where Babble exposes its HTTP API which provides information
about the internals of the hashgraph and data store. This is controlled by the
ServiceAddr (--service-listen) option. It can also be disabled altogether
with the NoService (--no-service) option.
When we use Babble as a native Go library, we set the InmemProxy directly in the
Config object's Proxy field.
Instead, when Babble and the application are connected by a TCP interface, we start Babble as a standalone executable and we specify the endpoints of the connection:
--proxy-listen: where Babble listens for transactions from the App.--client-connect: where the App listens for blocks from Babble
EnableFastSync (--fast-sync) tells Babble to attempt to fast-forward to the
tip of the hashgraph when joining, instead of downloading and replaying the
entire hashgraph from start. More on this in
fast-sync. This options
requires the Snapshot and Restore handlers to be carefully implemented in the
AppProxy.
-
LogLevel(--log): Determines the chattiness of the log output. -
HeartbeatTimeout(--heartbeat): Frequency of the gossip timer when there is something to gossip about. -
SlowHeartbeatTimeout(--slow-heartbeat): Frequency of the gossip timer when there is nothing to gossip about. -
MaxPool(--max-pool): Controls how many connections are pooled per target in the gossip routines. -
TCPTimeout(--timeout): Timeout of gossip RPC connections (also applies for WebRTC connections). -
JoinTimeout(--join_timeout): Timeout of join requests. -
SyncLimit(--sync-limit): Max number of hashgraph events to include in a SyncResponse or EagerSyncRequest. -
CacheSize(--cache-size): Max number of items in the in-memory caches. -
SuspendLimit(--suspend-limit): Multiplier applied to the number of validators to determine the limit of undetermined events that will cause a node to become suspended. -
Moniker(--moniker): Friendly name for this node. It takes precedence over the moniker defined in JSON peers files. -
SignalSkipVerify(--signal-skip-verify): (insecure) Controls whether the signal client verifies the server's certificate chain and hostname when WebRTC is activated.
Babble is written in Golang 1.14. Hence, the first step is to install Go version 1.14 or above which is both the programming language and a CLI tool for managing Go code. Go is very opinionated and will require you to define a workspace where all your go code will reside.
Fetch Babble from github:
$ go get github.com/mosaicnetworks/babbleDownload all dependencies and put them in the vendor folder.
[...]/babble$ make vendorBabble uses go mod to manage dependencies.
Bash scripts used in this project assume the use of GNU versions of coreutils. Please ensure you have GNU versions of these programs installed:-
example for macOS:
# --with-default-names makes the `sed` and `awk` commands default to gnu sed and gnu awk respectively.
brew install gnu-sed gawk --with-default-namesBabble has extensive unit-testing.
[...]/babble$ make testIf everything goes well, it should output something along these lines:
? github.com/mosaicnetworks/babble/src/babble [no test files]
ok github.com/mosaicnetworks/babble/src/common 0.015s
ok github.com/mosaicnetworks/babble/src/crypto 0.122s
ok github.com/mosaicnetworks/babble/src/hashgraph 10.270s
? github.com/mosaicnetworks/babble/src/mobile [no test files]
ok github.com/mosaicnetworks/babble/src/net 0.012s
ok github.com/mosaicnetworks/babble/src/node 19.171s
ok github.com/mosaicnetworks/babble/src/peers 0.038s
? github.com/mosaicnetworks/babble/src/proxy [no test files]
ok github.com/mosaicnetworks/babble/src/dummy 0.013s
ok github.com/mosaicnetworks/babble/src/proxy/inmem 0.037s
ok github.com/mosaicnetworks/babble/src/proxy/socket 0.009s
? github.com/mosaicnetworks/babble/src/proxy/socket/app [no test files]
? github.com/mosaicnetworks/babble/src/proxy/socket/babble [no test files]
? github.com/mosaicnetworks/babble/src/service [no test files]
? github.com/mosaicnetworks/babble/src/version [no test files]
? github.com/mosaicnetworks/babble/cmd/babble [no test files]
? github.com/mosaicnetworks/babble/cmd/babble/commands [no test files]
? github.com/mosaicnetworks/babble/cmd/dummy [no test files]
? github.com/mosaicnetworks/babble/cmd/dummy/commands [no test files]
The easiest way to build binaries is to do so in a hermetic Docker container. Use this simple command:
[...]/babble$ make distThis will launch the build in a Docker container and write all the artifacts in the build/ folder.
[...]/babble$ tree build
build/
├── dist
│ ├── babble_0.1.0_darwin_386.zip
│ ├── babble_0.1.0_darwin_amd64.zip
│ ├── babble_0.1.0_freebsd_386.zip
│ ├── babble_0.1.0_freebsd_arm.zip
│ ├── babble_0.1.0_linux_386.zip
│ ├── babble_0.1.0_linux_amd64.zip
│ ├── babble_0.1.0_linux_arm.zip
│ ├── babble_0.1.0_SHA256SUMS
│ ├── babble_0.1.0_windows_386.zip
│ └── babble_0.1.0_windows_amd64.zip
└── pkg
├── darwin_386
│ └── babble
├── darwin_amd64
│ └── babble
├── freebsd_386
│ └── babble
├── freebsd_arm
│ └── babble
├── linux_386
│ └── babble
├── linux_amd64
│ └── babble
├── linux_arm
│ └── babble
├── windows_386
│ └── babble.exe
└── windows_amd64
└── babble.exeTo see Babble in action, we have provided a series of scripts to bootstrap a local test network with the dummy application and the SocketProxy.
NOTE: This has been tested on Ubuntu 18.04 and macOS.
Make sure you have Docker installed.
Then, run the testnet:
[...]/babble$ cd demo
[...]/babble/demo$ make
# or using webrtc
[...]/babble/demo$ make webrtc=trueOnce the testnet is started, a script is automatically launched to monitor each Babble node:
consensus_events:180 consensus_transactions:40 events_per_second:0.00 id:1 last_block_index:3 last_consensus_round:17 num_peers:3 round_events:7 rounds_per_second:0.00 state:Babbling sync_rate:1.00 transaction_pool:0 undetermined_events:18
consensus_events:180 consensus_transactions:40 events_per_second:0.00 id:3 last_block_index:3 last_consensus_round:17 num_peers:3 round_events:7 rounds_per_second:0.00 state:Babbling sync_rate:1.00 transaction_pool:0 undetermined_events:20
consensus_events:180 consensus_transactions:40 events_per_second:0.00 id:2 last_block_index:3 last_consensus_round:17 num_peers:3 round_events:7 rounds_per_second:0.00 state:Babbling sync_rate:1.00 transaction_pool:0 undetermined_events:21
consensus_events:180 consensus_transactions:40 events_per_second:0.00 id:0 last_block_index:3 last_consensus_round:17 num_peers:3 round_events:7 rounds_per_second:0.00 state:Babbling sync_rate:1.00 transaction_pool:0 undetermined_events:20
Running docker ps -a will show you that 9 docker containers have been launched:
[...]/babble/demo$ docker ps -a
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
ba80ef275f22 mosaicnetworks/watcher "/watch.sh" 48 seconds ago Up 7 seconds watcher
4620ed62a67d mosaicnetworks/dummy "dummy '--name=client" 49 seconds ago Up 48 seconds 1339/tcp client4
847ea77bd7fc mosaicnetworks/babble "babble run --cache_s" 50 seconds ago Up 49 seconds 80/tcp, 1337-1338/tcp node4
11df03bf9690 mosaicnetworks/dummy "dummy '--name=client" 51 seconds ago Up 50 seconds 1339/tcp client3
00af002747ca mosaicnetworks/babble "babble run --cache_s" 52 seconds ago Up 50 seconds 80/tcp, 1337-1338/tcp node3
b2011d3d65bb mosaicnetworks/dummy "dummy '--name=client" 53 seconds ago Up 51 seconds 1339/tcp client2
e953b50bc1db mosaicnetworks/babble "babble run --cache_s" 53 seconds ago Up 52 seconds 80/tcp, 1337-1338/tcp node2
0c9dd65de193 mosaicnetworks/dummy "dummy '--name=client" 54 seconds ago Up 53 seconds 1339/tcp client1
d1f4e5008d4d mosaicnetworks/babble "babble run --cache_s" 55 seconds ago Up 54 seconds 80/tcp, 1337-1338/tcp node1Indeed, each replica is composed of a dummy application coupled to a Babble node running in a different container.
Run the demo script to play with the Dummy App which is a simple chat application powered by the Babble consensus platform:
[...]/babble/demo$ make demo
Finally, stop the testnet:
[...]/babble/demo$ make stop