Skip to main content
Version: v0.47

Node Client (Daemon)


The main endpoint of a Cosmos SDK application is the daemon client, otherwise known as the full-node client. The full-node runs the state-machine, starting from a genesis file. It connects to peers running the same client in order to receive and relay transactions, block proposals and signatures. The full-node is constituted of the application, defined with the Cosmos SDK, and of a consensus engine connected to the application via the ABCI.


Pre-requisite Readings

main function

The full-node client of any Cosmos SDK application is built by running a main function. The client is generally named by appending the -d suffix to the application name (e.g. appd for an application named app), and the main function is defined in a ./appd/cmd/main.go file. Running this function creates an executable appd that comes with a set of commands. For an app named app, the main command is appd start, which starts the full-node.

In general, developers will implement the main.go function with the following structure:

  • First, an encodingCodec is instantiated for the application.
  • Then, the config is retrieved and config parameters are set. This mainly involves setting the Bech32 prefixes for addresses.
  • Using cobra, the root command of the full-node client is created. After that, all the custom commands of the application are added using the AddCommand() method of rootCmd.
  • Add default server commands to rootCmd using the server.AddCommands() method. These commands are separated from the ones added above since they are standard and defined at Cosmos SDK level. They should be shared by all Cosmos SDK-based applications. They include the most important command: the start command.
  • Prepare and execute the executor.

See an example of main function from the simapp application, the Cosmos SDK's application for demo purposes:


start command

The start command is defined in the /server folder of the Cosmos SDK. It is added to the root command of the full-node client in the main function and called by the end-user to start their node:

# For an example app named "app", the following command starts the full-node.
appd start

# Using the Cosmos SDK's own simapp, the following commands start the simapp node.
simd start

As a reminder, the full-node is composed of three conceptual layers: the networking layer, the consensus layer and the application layer. The first two are generally bundled together in an entity called the consensus engine (CometBFT by default), while the third is the state-machine defined with the help of the Cosmos SDK. Currently, the Cosmos SDK uses CometBFT as the default consensus engine, meaning the start command is implemented to boot up a CometBFT node.

The flow of the start command is pretty straightforward. First, it retrieves the config from the context in order to open the db (a leveldb instance by default). This db contains the latest known state of the application (empty if the application is started from the first time.

With the db, the start command creates a new instance of the application using an appCreator function:


Note that an appCreator is a function that fulfills the AppCreator signature:


In practice, the constructor of the application is passed as the appCreator.


Then, the instance of app is used to instantiate a new CometBFT node:


The CometBFT node can be created with app because the latter satisfies the abci.Application interface (given that app extends baseapp). As part of the node.New method, CometBFT makes sure that the height of the application (i.e. number of blocks since genesis) is equal to the height of the CometBFT node. The difference between these two heights should always be negative or null. If it is strictly negative, node.New will replay blocks until the height of the application reaches the height of the CometBFT node. Finally, if the height of the application is 0, the CometBFT node will call InitChain on the application to initialize the state from the genesis file.

Once the CometBFT node is instantiated and in sync with the application, the node can be started:


Upon starting, the node will bootstrap its RPC and P2P server and start dialing peers. During handshake with its peers, if the node realizes they are ahead, it will query all the blocks sequentially in order to catch up. Then, it will wait for new block proposals and block signatures from validators in order to make progress.

Other commands

To discover how to concretely run a node and interact with it, please refer to our Running a Node, API and CLI guide.