Skip to content

Latest commit

 

History

History
826 lines (642 loc) · 21.2 KB

morpheusvm.md

File metadata and controls

826 lines (642 loc) · 21.2 KB

Introduction

In this tutorial, we will go through how to build MorpheusVM, a HyperVM which supports token transfers. Although an implementation of MorpheusVM already exists, we will set out to build our own version from scratch. The outline of this tutorial is as follows:

  • Prerequisites
  • Initializing MorpheusVM
  • Creating Transfer, Pt. 1
  • Implementing Storage
    • Read/write functions
    • StateManager
  • Creating Transfer, Pt. 2
  • Bringing Everything Together
  • Options
  • Workload Tests

This introduction will go over everything up to and including the section "Bringing Everything Together".

Prerequisites

  • You are using Go 1.22.8

To confirm, you can run:

go version

And you should see the ouptut (slightly different depending on your machine):

go version go1.22.8 darwin/arm64

Cloning HyperSDK

To get started, clone the HyperSDK:

git clone git@github.com:ava-labs/hypersdk.git
cd hypersdk

Initializing MorpheusVM

To start, we'll create a new directory in examples and initialize the go mod:

mkdir examples/tutorial
cd examples/tutorial
go mod init github.com/ava-labs/hypersdk/examples/tutorial

We now define our VM constants. We'll create a new package to store these values:

mkdir consts

Inside consts/, we'll add a consts.go file to declare the name and the initial version of our VM:

package consts

import (
	"github.com/ava-labs/avalanchego/ids"
	"github.com/ava-labs/avalanchego/version"
)

const (
	Name = "tutorialvm"
)

var ID ids.ID

func init() {
	b := make([]byte, ids.IDLen)
	copy(b, []byte(Name))
	vmID, err := ids.ToID(b)
	if err != nil {
		panic(err)
	}
	ID = vmID
}

var Version = &version.Semantic{
	Major: 0,
	Minor: 0,
	Patch: 1,
}

This will create a warning because we have not imported AvalancheGo into this workspace yet. To fix that, let's add AvalancheGo as a dependency:

go get github.com/ava-labs/avalanchego && go mod tidy

Creating Transfer Pt. 1

We'll implement a single Transfer action to handle payments in a native token. Actions are wrapped into the HyperSDK chain.Transaction type and executed on-chain.

A HyperSDK transaction includes a slice of actions. When the transaction is executed, it verifies the transaction validity and then executes each action included in the transaction sequentially. If any action execution fails, then all of the state changes from the transaction will be reverted.

To find more details on how actions are wrapped into HyperSDK transactions, see here.

You can think of each field in the Transfer type as if it's a parameter to the function you want to execute onchain.

To start, make sure you're in tutorial/ and execute the following:

mkdir actions

In actions/, create transfer.go. This file will declare the struct and add a type assertion that it implements the chain.Action interface:

package actions

import (
	"github.com/ava-labs/hypersdk/chain"
	"github.com/ava-labs/hypersdk/codec"
)

var _ chain.Action = (*Transfer)(nil)

type Transfer struct {
	// To is the recipient of the [Value].
	To codec.Address `serialize:"true" json:"to"`

	// Amount are transferred to [To].
	Value uint64 `serialize:"true" json:"value"`

	// Optional message to accompany transaction.
	Memo []byte `serialize:"true" json:"memo"`
}

This imports the HyperSDK, so we'll import it and redirect it to our local version of the HyperSDK:

go get github.com/ava-labs/hypersdk
go mod edit -replace github.com/ava-labs/hypersdk=../../
go mod tidy

Note: we include JSON and serialize tags in the struct to provide instructions on how the Transfer action should be marshalled to JSON or a binary representation. The JSON tags provide the key in the resulting JSON and serialize:"true" indicates that the codec should include each field when it marshals the struct to its binary representation.

Right now, the chain.Action type assertion should fail because we haven't implemented the interface yet. To fix that for now, we'll include stubs for each of the required functions that call panic("unimplemented"), so we can come back to fill these in later in the tutorial.

You can copy-paste the code below:

func (t *Transfer) GetTypeID() uint8 {
	panic("unimplemented")
}

func (t *Transfer) StateKeys(actor codec.Address, actionID ids.ID) state.Keys {
	panic("unimplemented")
}

func (t *Transfer) Execute(ctx context.Context, r chain.Rules, mu state.Mutable, timestamp int64, actor codec.Address, actionID ids.ID) (output codec.Typed, err error) {
	panic("unimplemented")
}

func (t *Transfer) ComputeUnits(chain.Rules) uint64 {
	panic("unimplemented")
}

func (t *Transfer) ValidRange(chain.Rules) (start int64, end int64) {
	panic("unimplemented")
}

Now that we've defined our action, we'll move on to implementing the state layout for our VM, so that we can use state helpers to update the state instead of handling constructing and writing key-value pairs directly from our action.

Implementing Storage

When building a VM with the HyperSDK, the VM developer defines their own state storage layout. The HyperSDK also stores metadata in the current state, and defers to the VM where to store that data. Note: this will be changed in the future, so that developers get a default out of the box and can choose to override it if needed instead of needing to provide it.

With that in mind, we'll break the storage down into two components:

  • Read/Write Helper Functions for VM specific state
  • StateManager interface to tell the HyperSDK where to store its metadata

Before that, in tutorial/, create a new folder called storage.

Separating the State into Partitions

To start off, we'll add single byte prefixes to separate out the state into metadata required by the HyperSDK and an address -> balance mapping.

Let's create the file storage.go in storage/ with prefixes for balance, height, timestamp, and fees:

package storage

const (
	// Active state
	balancePrefix   = 0x0
	heightPrefix    = 0x1
	timestampPrefix = 0x2
	feePrefix       = 0x3
)

var (
	heightKey    = []byte{heightPrefix}
	timestampKey = []byte{timestampPrefix}
	feeKey       = []byte{feePrefix}
)

Implementing HyperSDK Metadata Handlers

Now, we'll implement functions that return the key where the HyperSDK should store the chain height, timestamp, and fee dimensions. This is currently defined by the VM, so that it has full control of its own state layout, but could be abstracted away and moved into the HyperSDK in the future.

Let's add the following functions to storage.go:

func HeightKey() (k []byte) {
	return heightKey
}

func TimestampKey() (k []byte) {
	return timestampKey
}

func FeeKey() (k []byte) {
	return feeKey
}

Next, we'll define the BalanceKey function. BalanceKey will return the state key that stores the provided address' balance.

The HyperSDK requires using size-encoded storage keys, which include a "chunk suffix." The "chunk suffix" is encoded as a uint16. The length encoded in the "chunk suffix" sets the maximum length of any value that can ever be placed in the corresponding key-value pair. This also means trying to change the suffix will change the key itself. In other words, if you write to the same key with a different suffix, it will write to a different location instead of overwriting. This means that the suffix must provide an upper bound for the value size forever or the VM would need to explicitly handle migrating from the key from size A to size B.

Chunks are denominated in 64 bytes, so a chunk size of 1 means the value at that key will never exceed 64 bytes (2 -> max of 128 bytes, 3 -> max of 192, etc).

With that in mind, we'll add BalanceChunks as a constant to the top of the file:

const BalanceChunks uint16 = 1

This means we're committing that to always store a value (the balance) in a byte array <= 64 bytes.

Now, we can implement BalanceKey using the constant:

// [balancePrefix] + [address] + [BalanceChunks suffix]
func BalanceKey(addr codec.Address) (k []byte) {
	k = make([]byte, 1+codec.AddressLen+consts.Uint16Len)
	k[0] = balancePrefix
	copy(k[1:], addr[:])
	binary.BigEndian.PutUint16(k[1+codec.AddressLen:], BalanceChunks)
	return
}

Balance Helper Functions

Next, we'll implement balance getter and setter functions.

Let's start by adding setBalance to set the balance for a pre-computed key:

func setBalance(
	ctx context.Context,
	mu state.Mutable,
	key []byte,
	balance uint64,
) error {
	return mu.Insert(ctx, key, binary.BigEndian.AppendUint64(nil, balance))
}

Note: we use key instead of a codec.Address type, so that the higher level helpers only need to compute balance keys once if they call setBalance more than once.

Now, let's add getBalance. To re-use later, we'll start by implementing an innerGetBalance function that will interpret a raw value and error:

func innerGetBalance(
	v []byte,
	err error,
) (uint64, bool, error) {
	if errors.Is(err, database.ErrNotFound) {
		return 0, false, nil
	}
	if err != nil {
		return 0, false, err
	}
	val, err := database.ParseUInt64(v)
	if err != nil {
		return 0, false, err
	}
	return val, true, nil
}

Having defined the function to interpret the balance value, we'll implement an exported GetBalance and a private version that returns additional details:

func getBalance(
	ctx context.Context,
	im state.Immutable,
	addr codec.Address,
) ([]byte, uint64, bool, error) {
	k := BalanceKey(addr)
	bal, exists, err := innerGetBalance(im.GetValue(ctx, k))
	return k, bal, exists, err
}

func GetBalance(
	ctx context.Context,
	im state.Immutable,
	addr codec.Address,
) (uint64, error) {
	_, bal, _, err := getBalance(ctx, im, addr)
	return bal, err
}

Higher Level Balance Functions

Now that we've implemented the base level storage helpers for balances, we'll implement AddBalance and SubBalance utilizing the functions we just wrote:

func AddBalance(
	ctx context.Context,
	mu state.Mutable,
	addr codec.Address,
	amount uint64,
	create bool,
) (uint64, error) {
	key, bal, exists, err := getBalance(ctx, mu, addr)
	if err != nil {
		return 0, err
	}
	// Don't add balance if account doesn't exist. This
	// can be useful when processing fee refunds.
	if !exists && !create {
		return 0, nil
	}
	nbal, err := smath.Add(bal, amount)
	if err != nil {
		return 0, fmt.Errorf(
			"%w: could not add balance (bal=%d, addr=%v, amount=%d)",
			ErrInvalidBalance,
			bal,
			addr,
			amount,
		)
	}
	return nbal, setBalance(ctx, mu, key, nbal)
}

func SubBalance(
	ctx context.Context,
	mu state.Mutable,
	addr codec.Address,
	amount uint64,
) (uint64, error) {
	key, bal, ok, err := getBalance(ctx, mu, addr)
	if !ok {
		return 0, ErrInvalidAddress
	}
	if err != nil {
		return 0, err
	}
	nbal, err := smath.Sub(bal, amount)
	if err != nil {
		return 0, fmt.Errorf(
			"%w: could not subtract balance (bal=%d, addr=%v, amount=%d)",
			ErrInvalidBalance,
			bal,
			addr,
			amount,
		)
	}
	if nbal == 0 {
		// If there is no balance left, we should delete the record instead of
		// setting it to 0.
		return 0, mu.Remove(ctx, key)
	}
	return nbal, setBalance(ctx, mu, key, nbal)
}

This will give us a few warnings for smath, mconsts, and two errors that we have not defined yet. Let's go ahead and import the AvalancheGo safe math package and define those two errors in another file.

First, let's add the math package from AvalancheGo to our imports:

	smath "github.com/ava-labs/avalanchego/utils/math"

Next, we'll add the consts package we defined earlier:

	mconsts "github.com/ava-labs/hypersdk/examples/tutorial/consts"

We now define the two errors from earlier. In storage/, create a file called errors.go and paste the following:

package storage

import "errors"

var (
	ErrInvalidAddress = errors.New("invalid address")
	ErrInvalidBalance = errors.New("invalid balance")
)

Going back to storage.go, you should see that the errors from behind are now gone.

State Manager

Now we'll implement the chain.StateManager interface to tell the HyperSDK how to modify our VM's state when it needs to store metadata or charge fees.

Let's start off by creating a new state_manager.go file in storage/ and adding a new StateManager type with function stubs for each of the required functions:

package storage

import (
	"context"

	"github.com/ava-labs/hypersdk/chain"
	"github.com/ava-labs/hypersdk/codec"
	"github.com/ava-labs/hypersdk/state"
)

var _ chain.StateManager = (*StateManager)(nil)

type StateManager struct{}

func (s *StateManager) FeeKey() []byte {
	panic("unimplemented")
}

func (s *StateManager) HeightKey() []byte {
	panic("unimplemented")
}

func (s *StateManager) TimestampKey() []byte {
	panic("unimplemented")
}

func (s *StateManager) AddBalance(ctx context.Context, addr codec.Address, mu state.Mutable, amount uint64, createAccount bool) error {
	panic("unimplemented")
}

func (s *StateManager) CanDeduct(ctx context.Context, addr codec.Address, im state.Immutable, amount uint64) error {
	panic("unimplemented")
}

func (s *StateManager) Deduct(ctx context.Context, addr codec.Address, mu state.Mutable, amount uint64) error {
	panic("unimplemented")
}

func (s *StateManager) SponsorStateKeys(addr codec.Address) state.Keys {
	panic("unimplemented")
}

The type assertion should pass, so now we can go through and implement each function correctly.

For each of the metadata functions, we'll simply return the state keys we already defined when partitioning our state:

func (*StateManager) HeightKey() []byte {
	return HeightKey()
}

func (*StateManager) TimestampKey() []byte {
	return TimestampKey()
}

func (*StateManager) FeeKey() []byte {
	return FeeKey()
}

Now, we'll implement the balance handler functions re-using the helpers we've already implemented:

func (*StateManager) CanDeduct(
	ctx context.Context,
	addr codec.Address,
	im state.Immutable,
	amount uint64,
) error {
	bal, err := GetBalance(ctx, im, addr)
	if err != nil {
		return err
	}
	if bal < amount {
		return ErrInvalidBalance
	}
	return nil
}

func (*StateManager) Deduct(
	ctx context.Context,
	addr codec.Address,
	mu state.Mutable,
	amount uint64,
) error {
	return SubBalance(ctx, mu, addr, amount)
}

func (*StateManager) AddBalance(
	ctx context.Context,
	addr codec.Address,
	mu state.Mutable,
	amount uint64,
	createAccount bool,
) error {
	_, err := AddBalance(ctx, mu, addr, amount, createAccount)
	return err
}

Finally, we need to implement SponsorStateKeys.

The HyperSDK uses state.Keys to determine what state keys may be touched during the execution of a transaction or block. state.Keys includes explicit Read/Write/Allocate permissions, so that it can determine how to safely execute transactions in parallel.

This enables parallel transaction execution both for prefetching state and executing transactions.

SponsorStateKeys provides the state keys required by CanDeduct and Deduct when the HyperSDK charges fees to the transaction sponsor (see account abstraction for more details on the Auth module).

We'll re-use our BalanceKey function and specify that both read and write permissions, since the HyperSDK may need to both read/write this balance when it handles fees:

func (*StateManager) SponsorStateKeys(addr codec.Address) state.Keys {
	return state.Keys{
		string(BalanceKey(addr)): state.Read | state.Write,
	}
}

Creating Transfer, Pt. 2

Now it's time to put them into action. Going back to transfer.go we'll update the stubbed out functions implementing the chain.Action interface with a real implementation.

GetTypeID

We need to include a typeID for our codec, so we'll implement GetTypeID to return a constant typeID.

In consts/, create a new file named types.go and paste the following:

package consts

const (
	// Action TypeIDs
	TransferID uint8 = 0
)

Next, go to actions/transfer.go where we'll update our GetTypeID function stub to return the constant:

func (*Transfer) GetTypeID() uint8 {
	return mconsts.TransferID
}

We import the following into transfer.go:

    mconsts "github.com/ava-labs/hypersdk/examples/tutorial/consts"

StateKeys()

For Transfer, we need read/write access to both the actor address and the to address (sender and receiver). Additionally, if we're transferring to a new address, we may allocate a new account, which means we need to include permission to allocate a new state. So we'll use state.Read | state.Write for the actor and state.All for the to address:

Beforehand, we need to import the storage package from earlier:

	"github.com/ava-labs/hypersdk/examples/tutorial/storage"
func (t *Transfer) StateKeys(actor codec.Address, _ ids.ID) state.Keys {
	return state.Keys{
		string(storage.BalanceKey(actor)): state.Read | state.Write,
		string(storage.BalanceKey(t.To)):  state.All,
	}
}

Execute()

We start off by first defining the values necessary for this function at the top of the file:

const MaxMemoSize = 256

var (
	ErrOutputValueZero    = errors.New("value is zero")
	ErrOutputMemoTooLarge = errors.New("memo is too large")
)

We can now define Execute(). Recall that we should first check any invariants and then execute the necessary state changes. Therefore, we have the following:

func (t *Transfer) Execute(
	ctx context.Context,
	_ chain.Rules,
	mu state.Mutable,
	_ int64,
	actor codec.Address,
	_ ids.ID,
) (codec.Typed, error) {
	if t.Value == 0 {
		return nil, ErrOutputValueZero
	}
	if len(t.Memo) > MaxMemoSize {
		return nil, ErrOutputMemoTooLarge
	}
	senderBalance, err := storage.SubBalance(ctx, mu, actor, t.Value)
	if err != nil {
		return nil, err
	}
	receiverBalance, err := storage.AddBalance(ctx, mu, t.To, t.Value, true)
	if err != nil {
		return nil, err
	}

	return &TransferResult{
		SenderBalance:   senderBalance,
		ReceiverBalance: receiverBalance,
	}, nil
}

ComputeUnits

The ComputeUnits function specifies how many units of computation are consumed by this action. The HyperSDK uses dynamic, multi-dimensional fees and specifies its own resource limits ie. units of compute, storage, and bandwidth. This means it's more important that ComputeUnits of different actions is proportional to the maximum, target, and the cost of other actions than it is to be exact. So, we'll simply return 1 for Transfer.

Let's define a constant, TransferComputeUnits, at the top of the file.

const TransferComputeUnits = 1

Now we can implement our function:

func (*Transfer) ComputeUnits(chain.Rules) uint64 {
	return TransferComputeUnits
}

ValidRange

Finally, we need to specify the range of timestamps where this action will be considered valid. We'll use -1 to denote that there's no bound ie. -1, -1 indicates that the action is valid forever.

Now, we'll update our function stub:

func (*Transfer) ValidRange(chain.Rules) (int64, int64) {
	return -1, -1
}

Bringing Everything Together

So far, we’ve implemented the following:

  • Actions
  • Storage

We can now finish this MorpheusVM tutorial by implementing the vm interface. In tutorial/, create a subdirectory called vm. In that folder, let's create the file vm.go. Here, we first define the “registry” of MorpheusVM:

package vm

import (
	"github.com/ava-labs/avalanchego/utils/wrappers"

	"github.com/ava-labs/hypersdk/auth"
	"github.com/ava-labs/hypersdk/chain"
	"github.com/ava-labs/hypersdk/codec"
	"github.com/ava-labs/hypersdk/examples/tutorial/actions"
)

var (
	ActionParser *codec.TypeParser[chain.Action]
	AuthParser   *codec.TypeParser[chain.Auth]
	OutputParser *codec.TypeParser[codec.Typed]
)

// Setup types
func init() {
	ActionParser = codec.NewTypeParser[chain.Action]()
	AuthParser = codec.NewTypeParser[chain.Auth]()
	OutputParser = codec.NewTypeParser[codec.Typed]()

	errs := &wrappers.Errs{}
	errs.Add(
		ActionParser.Register(&actions.Transfer{}, actions.UnmarshalTransfer),

		AuthParser.Register(&auth.ED25519{}, auth.UnmarshalED25519),
		AuthParser.Register(&auth.SECP256R1{}, auth.UnmarshalSECP256R1),
		AuthParser.Register(&auth.BLS{}, auth.UnmarshalBLS),

		OutputParser.Register(&actions.TransferResult{}, nil),
	)
	if errs.Errored() {
		panic(errs.Err)
	}
}

By “registry”, we mean the ActionParser and AuthParser which tell our VM which actions and cryptographic functions that it’ll support.

Finally, we create a New() function that allows for the VM we’ve worked on to be instantiated.

// NewWithOptions returns a VM with the specified options
func New(options ...vm.Option) (*vm.VM, error) {
	return defaultvm.New(
		consts.Version,
		genesis.DefaultGenesisFactory{},
		&storage.StateManager{},
		ActionParser,
		AuthParser,
		OutputParser,
		auth.Engines(),
		options...,
	)
}

Note: if you see an error when importing "github.com/ava-labs/hypersdk/vm/defaultvm", running go mod tidy should fix it

At this point, we've implemented the necessary components of MorpheusVM. It's time to move onto the next section, where we will implement options and extend the functionality of MorpheusVM.