Create a New Virtual Machine


The code below is slightly out of date. Some methods, interfaces and implementations are slightly different than in this tutorial. We’re going to leave this up because the current code is very similar and this tutorial is still useful in demonstrating how Avalanche’s VM model works.


One of the core features of the Avalanche network is the creation of new, custom blockchains, which are defined by Virtual Machines.

In this tutorial, we’ll create a very simple Virtual Machine. The blockchain defined by the Virtual Machine is a timestamp server. Each block in the blockchain contains the timestamp when it was created along with a 32 byte piece of data (payload.) Each block’s timestamp is after its parent’s timestamp.

Such a server is useful because it can be used to prove a piece of data existed at the time the block was created. For example, suppose that you have a book manuscript, and you want to be able to prove in the future that the manuscript exists today. You add a block to the blockchain where the block’s payload is a hash of your manuscript. In the future, you can prove that the manuscript existed today by showing that the block, which has a timestamp from today, has as its payload the hash of your manuscript. (This follows from the fact that finding the pre-image of a hash is impossible.)

Before we get to the implementation of the Virtual Machine, we’ll look at the interface that a Virtual Machine must implement to be compatible with the platform’s Avalanche consensus engine.

We’ll show and explain all the code that constitutes this Virtual Machine in snippets. In-line comments explain what’s going on in the code. At the bottom of some snippets, we expound further on some portions of the code. If you want to see the code in one place, rather than in snippets, you can see it in our Github repository.

The snowman.VM Interface

To reach consensus on linear blockchains (as opposed to DAG blockchains), Avalanche uses the Avalanche-powered Snowman consensus engine. The blockchain we’re defining is linear, so it will use Snowman. In order to be compatible with Snowman, the Virtual Machine that defines the blockchain must implement the snowman.VM interface, which we include below from its declaration in

The interface is big, but don’t worry. We’ll explain each method and see an implementation example, and it’s not necessary you understand every nuance.

// ChainVM defines the methods a Virtual Machine must implement to use the Snowman consensus engine.
// A Snowman VM defines the state contained in a linear blockchain,
// the state transition functions that modify the blockchain's state,
// the API exposed by the blockchain, as well as other aspects of the blockchain.
type ChainVM interface {
    // Initialize an instance of the blockchain defined by this VM.
    // [ctx]: Run-time context and metadata about the blockchain.
    //     [ctx.networkID]: The ID of the network this blockchain exists on.
    //     [ctx.chainID]: The unique ID of this blockchain.
    //     [ctx.Log]: Used to log messages
    //     [ctx.NodeID]: The ID of this node.
    // [db]: The database the blockchain persists data to.
    // [genesisBytes]: The byte representation of the genesis state of this blockchain.
    //                 If this VM were an account-based payments system, for example
    //                 `genesisBytes` would probably be a genesis
    //                 transaction that gives coins to some accounts, and this
    //                 transaction would be in the genesis block.
    // [toEngine]: The channel used to send messages to the consensus engine.
    // [fxs]: Feature extensions that attach to this VM.
    // In this release, we do not document feature extensions. You can ignore them.
        ctx *snow.Context,
        db database.Database,
        genesisBytes []byte,
        toEngine chan<- Message,
        fxs []*Fx,
    ) error

    // Shutdown this blockchain.

    // Creates the HTTP handlers for this blockchain's API
    // and specifies the endpoint where they handle traffic.
    // Each handler handles traffic to a specific endpoint.
    // Each endpoint begins with:
    // [Node's address]:[Node's HTTP port]/ext/bc/[blockchain ID]
    // A handler may handle traffic at an *extension* of the above endpoint.
    // The method returns a mapping from an extension to the HTTP handler at that extension.
    // For example, if this VM implements an account-based payments system,
    // CreateHandlers might return this map:
    // "accounts" --> [handler for API calls that pertain to accounts]
    // "transactions" --> [handler for API calls that pertain to transactions]
    // The accounts handler would have endpoint [Node's address]:[Node's HTTP port]/ext/bc/[blockchain ID]/accounts
    // The trasnsactions handler would have endpoint [Node's address]:[Node's HTTP port]/ext/bc/[blockchain ID]/trasnsactions
    // If a handler is mapped to by the empty string, it has no extension.
    // It handles traffic at [Node's address]:[Node's HTTP port]/ext/bc/[blockchain ID]
    CreateHandlers() map[string]*HTTPHandler

    // Attempt to create a new block from pending data in the blockchain's mempool.
    // If there is no new block to be created, returns an error.
    BuildBlock() (snowman.Block, error)

    // Attempt to create a block from its byte representation.
    ParseBlock([]byte) (snowman.Block, error)

    // Attempt to fetch a block by its ID.
    // If the block does not exist, returns an error.
    GetBlock(ids.ID) (snowman.Block, error)

    // Set the preferred block to the one with the specified ID.
    // New blocks will be built atop the preferred block.
    // This should always be a block that has no children known to consensus.

    // LastAccepted returns the ID of the last accepted block.
    // If no blocks have been accepted yet, should return the genesis block's ID.
    LastAccepted() ids.ID

The snowman.Block Interface

You may have noticed the snowman.Block type referenced in the snowman.VM interface. It describes the methods that a block must implement to be a block in a linear (Snowman) chain.

Let’s look at this interface and its methods, which we copy from Again, it’s OK if you don’t understand every detail. We’ll see an example soon.

// Block is a block in a blockchain.
// Blocks are guaranteed to be Verified, Accepted, and Rejected in topological
// order. Specifically, if Verify is called, then the parent has already been
// verified. If Accept is called, then the parent has already been accepted. If
// Reject is called, the parent has already been accepted or rejected.
// If the status of the block is Unknown, ID is assumed to be able to be called.
// If the status of the block is Accepted or Rejected; Parent, Verify, Accept,
// and Reject will never be called.
type Block interface {
    // ID returns this block's unique ID.
    // Typically, a block's ID is a hash of its byte representation.
    // A block should return the same ID upon repeated calls.
    ID() ids.ID

    // Accept this block.
    // This block will be accepted by every correct node in the network.

    // Reject this block.
    // This block will not be accepted by any correct node in the network.

    // Status returns this block's current status.
    // If Accept has been called on n block with this ID, Accepted should be
    // returned. Similarly, if Reject has been called on a block with this
    // ID, Rejected should be returned. If the contents of this block are
    // unknown, then Unknown should be returned. Otherwise, Processing should be
    // returned.
    Status() Status

    // Parent returns this block's parent.
    // If the parent block is not known, a Block should be returned with the
    // status Unknown.
    Parent() Block

    // Verify that the state transition this block would make if accepted is
    // valid. If the state transition is invalid, a non-nil error should be
    // returned.
    // It is guaranteed that the Parent has been successfully verified.
    Verify() error

    // Bytes returns the binary representation of this block.
    // This is used for sending blocks to peers. The bytes should be able to be
    // parsed into the same block on another node.
    Bytes() []byte


We’ve created some types that your Virtual Machine implementation can embed (embedding is like Go’s version of inheritance) in order to handle boilerplate code.

In our example we use both of the below library types, and we encourage you to use them too.


This type, a struct, contains methods and fields common to all implementations of the snowman.ChainVM interface.


This type implements the following methods, which are part of the snowman.ChainVM interface:

  • SetPreference
  • Shutdown
  • LastAccepted

If your Virtual Machine implementation embeds a core.SnowmanVM, you need not implement any of these methods because they are already implemented by core.SnowmanVM. You may, if you want, override these inherited methods.


This type contains several fields that you’ll want to include in your Virtual Machine implementation. Among them:

  • DB: the blockchain’s database
  • Ctx: the blockchain’s runtime context
  • preferred: ID of the preferred block, which new blocks will be built on
  • lastAccepted: ID of the most recently accepted block
  • toEngine: the channel through which messages are sent to the consensus engine powering the blockchain
  • State: used to persist data such as blocks. Can be used to put/get any bytes.


This type, a struct, contains methods and fields common to all implementations of the snowman.Block interface.


This type implements the following methods, which are part of the snowman.Block interface:

  • ID
  • Parent
  • Accept
  • Reject
  • Status

Your Virtual Machine implementation will probably override Accept and Reject so that these methods cause application-specific state changes.


core.Block has a field VM, which is a reference to a core.SnowmanVM. This means that a core.Block has access to all of the fields and methods of that type.

Timestamp Server Implementation

Now we know the interface our Virtual Machine must implement and the libraries we can use to build a Virtual Machine.

Let’s write our Virtual Machine, which implements snowman.VM, and whose blocks implement snowman.Block.


First, let’s look at our block implementation.

The type declaration is:

// Block is a block on the chain.
// Each block contains:
// 1) A piece of data (the block's payload)
// 2) The (unix) timestamp when the block was created
type Block struct {
    *core.Block           `serialize:"true"`
    Data        [32]byte  `serialize:"true"`
    Timestamp   int64     `serialize:"true"`

The serialize:"true" tag indicates that when a block is serialized (when it’s persisted in the database or sent to other nodes, for example), the field with the tag is included in the serialized representation.


// Verify returns nil iff this block is valid.
// To be valid, it must be that:
// b.parent.Timestamp < b.Timestamp <= [local time] + 1 hour
func (b *Block) Verify() error {
    // Check to see if this block has already been verified by calling Verify on the
    // embedded *core.Block.
    // If there is an error while checking, return an error.
    // If the core.Block says the block is accepted, return accepted.
    if accepted, err := b.Block.Verify(); err != nil || accepted {
        return err

    // Get [b]'s parent
    parent, ok := b.Parent().(*Block)
    if !ok {
        return errors.New("error while retrieving block from database")

    // Ensure [b]'s timestamp is after its parent's timestamp.
    if b.Timestamp < time.Unix(parent.Timestamp, 0).Unix() {
        return errors.New("block's timestamp is more than 1 hour ahead of local time")

    // Ensure [b]'s timestamp is not more than an hour 
    // ahead of this node's time
    if b.Timestamp >= time.Now().Add(time.Hour).Unix() {
        return errors.New("block's timestamp is more than 1 hour ahead of local time")

    // Our block inherits VM from *core.Block.
    // It holds the database we read/write, b.VM.DB
    // We persist this block to that database using VM's SaveBlock method.
    b.VM.SaveBlock(b.VM.DB, b)

    // Then we flush the database's contents
    return b.VM.DB.Commit()

That’s all the code for our block implementation! All of the other methods of snowman.Block, which our Block must implement, are inherited from *core.Block.

Virtual Machine

Now let’s look at the implementation of VM, which implements the snowman.VM interface.

The declaration is:

// This Virtual Machine defines a blockchain that acts as a timestamp server
// Each block contains a piece of data (payload) and the timestamp when it was created
type VM struct {

    // codec serializes and de-serializes structs to/from bytes
    codec codec.Codec

    // Proposed pieces of data that haven't been put into a block and proposed yet
    mempool [][32]byte


// Initialize this vm
// [ctx] is the execution context
// [db] is this database we read/write
// [toEngine] is used to notify the consensus engine that new blocks are
//   ready to be added to consensus
// The data in the genesis block is [genesisData]
func (vm *VM) Initialize(
    ctx *snow.Context,
    db database.Database,
    genesisData []byte,
    toEngine chan<- common.Message,
    _ []*common.Fx,
) error {
    // First, we initialize the core.SnowmanVM.
    // vm.ParseBlock, which we'll see further on, tells the core.SnowmanVM how to deserialize
    // a block from bytes
    if err := vm.SnowmanVM.Initialize(ctx, db, vm.ParseBlock, toEngine); err != nil {
        ctx.Log.Error("error initializing SnowmanVM: %v", err)
        return err
    // Set vm's codec to a new codec, which we can use to 
    // serialize and deserialize blocks
    vm.codec = codec.NewDefault()

    // If the database is empty, initialize the state of this blockchain
    // using the genesis data
    if !vm.DBInitialized() {
        // Ensure that the genesis bytes are no longer than 32 bytes
        // (the genesis block, like all blocks, holds 32 bytes of data)
        if len(genesisData) > 32 {
            return errors.New("genesis data should be bytes (max length 32)")

        // genesisData is a byte slice (because that's what the snowman.VM interface says)
        // but each block contains an byte array.
        // To make the types match, take the first [dataLen] bytes from genesisData
        // and put them in an array
        var genesisDataArr [dataLen]byte
        copy(genesisDataArr[:], genesisData)

        // Create the genesis block
        // Timestamp of genesis block is 0. It has no parent, so we say the parent's ID is empty.
        // We'll come to the definition of NewBlock later.
        genesisBlock, err := vm.NewBlock(ids.Empty, genesisDataArr, time.Unix(0, 0))
        if err != nil {
            vm.Ctx.Log.Error("error while creating genesis block: %v", err)
            return err

        // Persist the genesis block to the database.
        // Normally, a block is saved to the database when Verify() is called on the block.
        // We don't call Verify on the genesis block, though. (It has no parent so
        // it wouldn't pass verification.)
        // vm.DB is the database, and was set when we initialized the embedded SnowmanVM.
        if err := vm.SaveBlock(vm.DB, genesisBlock); err != nil {
            vm.Ctx.Log.Error("error while saving genesis block: %v", err)
            return err

        // Accept the genesis block.
        // Sets [vm.lastAccepted] and [vm.preferred] to the genesisBlock.

        // Mark the database as initialized so that in the future when this chain starts
        // it pulls state from the database rather than starting over from genesis

        // Flush the database
        if err := vm.DB.Commit(); err != nil {
            vm.Ctx.Log.Error("error while commiting db: %v", err)
            return err
    return nil


This method adds a piece of data to the mempool and notifies the consensus layer of the blockchain that a new block is ready to be built and voted on. We’ll see where this is called later.

// proposeBlock appends [data] to [p.mempool].
// Then it notifies the consensus engine
// that a new block is ready to be added to consensus
// (namely, a block with data [data])
func (vm *VM) proposeBlock(data [dataLen]byte) {
    vm.mempool = append(vm.mempool, data)


// ParseBlock parses [bytes] to a snowman.Block
// This function is used by the vm's state to unmarshal blocks saved in state
// and by the consensus layer when it receives the byte representation of a block
// from another node
func (vm *VM) ParseBlock(bytes []byte) (snowman.Block, error) {
    // A new empty block
    block := &Block{}

    // Unmarshal the byte repr. of the block into our empty block
    err := vm.codec.Unmarshal(bytes, block)

    // Initialize the block
    // (Block inherits Initialize from its embedded *core.Block)
    block.Initialize(bytes, &vm.SnowmanVM)
    return block, err


// NewBlock returns a new Block where:
// - the block's parent has ID [parentID]
// - the block's data is [data]
// - the block's timestamp is [timestamp]
func (vm *VM) NewBlock(parentID ids.ID, data [dataLen]byte, timestamp time.Time) (*Block, error) {
    // Create our new block
    block := &Block{
        Block:     core.NewBlock(parentID),
        Data:      data,
        Timestamp: timestamp.Unix(),

    // Get the byte representation of the block
    blockBytes, err := vm.codec.Marshal(block)
    if err != nil {
        return nil, err

    // Initialize the block by providing it with its byte representation
    // and a reference to SnowmanVM
    block.Initialize(blockBytes, &vm.SnowmanVM)

    return block, nil


This method is called by the consensus layer after the application layer tells it that a new block is ready to be built (ie when vm.NotifyConsensus() is called).

// BuildBlock returns a block that this VM wants to add to consensus
func (vm *VM) BuildBlock() (snowman.Block, error) {
    // There is no data to put in a new block
    if len(vm.mempool) == 0 { 
        return nil, errors.New("there is no block to propose")

    // Get the value to put in the new block
    value := vm.mempool[0]
    vm.mempool = vm.mempool[1:]

    // Notify consensus engine that there are more pending data for blocks
    // (if that is the case) when done building this block
    if len(vm.mempool) > 0 {
        defer vm.NotifyBlockReady()

    // Build the block
    block, err := vm.NewBlock(vm.Preferred(), value, time.Now())
    if err != nil {
        return nil, err
    return block, nil


// CreateHandlers returns a map where:
// Keys: The path extension for this blockchain's API (empty in this case)
// Values: The handler for the API
// In this case, our blockchain has only one API, which we name timestamp,
// and it has no path extension, so the API endpoint:
// [Node IP]/ext/bc/[this blockchain's ID]
// See API section in documentation for more information
func (vm *VM) CreateHandlers() map[string]*common.HTTPHandler {
    // Create the API handler (we'll see the declaration of Service further on)
    handler := vm.NewHandler("timestamp", &Service{vm})
    return map[string]*common.HTTPHandler{
        "": handler,


AvalancheGo uses Gorilla’s RPC library. to implement APIs.

Using Gorilla, there is a struct for each API service. In the case of this blockchain, there’s only one API service.

The service struct’s declaration is:

// Service is the API service for this VM
type Service struct{ vm *VM }

For each API method there is: * A struct that defines the method’s arguments * A struct that defines the method’s return values * A method that implements the API method, and is parameterized on the above 2 structs


This API method allows clients to add a block to the blockchain.

// ProposeBlockArgs are the arguments to ProposeValue
type ProposeBlockArgs struct {
    // Data for the new block. Must be base 58 encoding (with checksum) of 32 bytes.
    Data string

// ProposeBlockReply is the reply from function ProposeBlock
type ProposeBlockReply struct{ 
    // True if the operation was successful
    Success bool

// ProposeBlock is an API method to propose a new block whose data is [args].Data.
func (s *Service) ProposeBlock(_ *http.Request, args *ProposeBlockArgs, reply *ProposeBlockReply) error {
    // Parse the data given as argument to bytes
    byteFormatter := formatting.CB58{}
    if err := byteFormatter.FromString(args.Data); err != nil {
        return errBadData
    // Ensure the data is 32 bytes
    dataSlice := byteFormatter.Bytes
    if len(dataSlice) != 32 {
        return errBadData
    // Convert the data from a byte slice to byte array
    var data [dataLen]byte             
    copy(data[:], dataSlice[:dataLen])
    // Invoke proposeBlock to trigger creation of block with this data
    reply.Success = true
    return nil


This API method allows clients to get a block by its ID.

// APIBlock is the API representation of a block
type APIBlock struct {
    Timestamp int64  `json:"timestamp"` // Timestamp of most recent block
    Data      string `json:"data"`      // Data in the most recent block. Base 58 repr. of 5 bytes.
    ID        string `json:"id"`        // String repr. of ID of the most recent block
    ParentID  string `json:"parentID"`  // String repr. of ID of the most recent block's parent

// GetBlockArgs are the arguments to GetBlock
type GetBlockArgs struct {
    // ID of the block we're getting.
    // If left blank, gets the latest block
    ID string

// GetBlockReply is the reply from GetBlock
type GetBlockReply struct {

// GetBlock gets the block whose ID is [args.ID]
// If [args.ID] is empty, get the latest block
func (s *Service) GetBlock(_ *http.Request, args *GetBlockArgs, reply *GetBlockReply) error {
    // If an ID is given, parse its string representation to an ids.ID
    // If no ID is given, ID becomes the ID of last accepted block
    var ID ids.ID
    var err error
    if args.ID == "" {
        ID = s.vm.LastAccepted()
    } else {
        ID, err = ids.FromString(args.ID)
        if err != nil {
            return errors.New("problem parsing ID")

    // Get the block from the database
    blockInterface, err := s.vm.GetBlock(ID)
    if err != nil {
        return errors.New("error getting data from database")

    block, ok := blockInterface.(*Block)
    if !ok { // Should never happen but better to check than to panic
        return errors.New("error getting data from database")

    // Fill out the response with the block's data
    reply.APIBlock.ID = block.ID().String()
    reply.APIBlock.Timestamp = block.Timestamp
    reply.APIBlock.ParentID = block.ParentID().String()
    byteFormatter := formatting.CB58{Bytes: block.Data[:]}
    reply.Data = byteFormatter.String()

    return nil

Wrapping Up

That’s it! That’s the entire implementation of a Virtual Machine which defines a blockchain-based timestamp server.

In this tutorial we learned:

  • The snowman.ChainVM interface, which all Virtual Machines that define a linear chain must implement
  • The snowman.Block interface, which all blocks that are part of a linear chain must implement
  • The core.SnowmanVM and core.Block library types, which make defining Virtual Machines faster