f2f46b674d
In this commit, we add a migration that will be used by existing wallets to ensure they can adhere to the new requirement of storing up to MaxReorgDepth entries within the block hash index.
451 lines
13 KiB
Go
451 lines
13 KiB
Go
package waddrmgr
|
|
|
|
import (
|
|
"bytes"
|
|
"encoding/binary"
|
|
"errors"
|
|
"fmt"
|
|
"testing"
|
|
"time"
|
|
|
|
"github.com/btcsuite/btcd/chaincfg"
|
|
"github.com/btcsuite/btcd/chaincfg/chainhash"
|
|
"github.com/btcsuite/btcwallet/walletdb"
|
|
)
|
|
|
|
// applyMigration is a helper function that allows us to assert the state of the
|
|
// top-level bucket before and after a migration. This can be used to ensure
|
|
// the correctness of migrations.
|
|
func applyMigration(t *testing.T, beforeMigration, afterMigration,
|
|
migration func(walletdb.ReadWriteBucket) error, shouldFail bool) {
|
|
|
|
t.Helper()
|
|
|
|
// We'll start by setting up our address manager backed by a database.
|
|
teardown, db, _ := setupManager(t)
|
|
defer teardown()
|
|
|
|
// First, we'll run the beforeMigration closure, which contains the
|
|
// database modifications/assertions needed before proceeding with the
|
|
// migration.
|
|
err := walletdb.Update(db, func(tx walletdb.ReadWriteTx) error {
|
|
ns := tx.ReadWriteBucket(waddrmgrNamespaceKey)
|
|
if ns == nil {
|
|
return errors.New("top-level namespace does not exist")
|
|
}
|
|
return beforeMigration(ns)
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to run beforeMigration func: %v", err)
|
|
}
|
|
|
|
// Then, we'll run the migration itself and fail if it does not match
|
|
// its expected result.
|
|
err = walletdb.Update(db, func(tx walletdb.ReadWriteTx) error {
|
|
ns := tx.ReadWriteBucket(waddrmgrNamespaceKey)
|
|
if ns == nil {
|
|
return errors.New("top-level namespace does not exist")
|
|
}
|
|
return migration(ns)
|
|
})
|
|
if err != nil && !shouldFail {
|
|
t.Fatalf("unable to perform migration: %v", err)
|
|
} else if err == nil && shouldFail {
|
|
t.Fatal("expected migration to fail, but did not")
|
|
}
|
|
|
|
// Finally, we'll run the afterMigration closure, which contains the
|
|
// assertions needed in order to guarantee than the migration was
|
|
// successful.
|
|
err = walletdb.Update(db, func(tx walletdb.ReadWriteTx) error {
|
|
ns := tx.ReadWriteBucket(waddrmgrNamespaceKey)
|
|
if ns == nil {
|
|
return errors.New("top-level namespace does not exist")
|
|
}
|
|
return afterMigration(ns)
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to run afterMigration func: %v", err)
|
|
}
|
|
}
|
|
|
|
// TestMigrationPupulateBirthdayBlock ensures that the migration to populate the
|
|
// wallet's birthday block works as intended.
|
|
func TestMigrationPopulateBirthdayBlock(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
var expectedHeight int32
|
|
beforeMigration := func(ns walletdb.ReadWriteBucket) error {
|
|
// To test this migration, we'll start by writing to disk 10
|
|
// random blocks.
|
|
block := &BlockStamp{}
|
|
for i := int32(1); i <= 10; i++ {
|
|
block.Height = i
|
|
blockHash := bytes.Repeat([]byte(string(i)), 32)
|
|
copy(block.Hash[:], blockHash)
|
|
if err := PutSyncedTo(ns, block); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
// With the blocks inserted, we'll assume that the birthday
|
|
// block corresponds to the 7th block (out of 11) in the chain.
|
|
// To do this, we'll need to set our birthday timestamp to the
|
|
// estimated timestamp of a block that's 6 blocks after genesis.
|
|
genesisTimestamp := chaincfg.MainNetParams.GenesisBlock.Header.Timestamp
|
|
delta := time.Hour
|
|
expectedHeight = int32(delta.Seconds() / 600)
|
|
birthday := genesisTimestamp.Add(delta)
|
|
if err := putBirthday(ns, birthday); err != nil {
|
|
return err
|
|
}
|
|
|
|
// Finally, since the migration has not yet started, we should
|
|
// not be able to find the birthday block within the database.
|
|
_, err := FetchBirthdayBlock(ns)
|
|
if !IsError(err, ErrBirthdayBlockNotSet) {
|
|
return fmt.Errorf("expected ErrBirthdayBlockNotSet, "+
|
|
"got %v", err)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// After the migration has completed, we should see that the birthday
|
|
// block now exists and is set to the correct expected height.
|
|
afterMigration := func(ns walletdb.ReadWriteBucket) error {
|
|
birthdayBlock, err := FetchBirthdayBlock(ns)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
if birthdayBlock.Height != expectedHeight {
|
|
return fmt.Errorf("expected birthday block with "+
|
|
"height %d, got %d", expectedHeight,
|
|
birthdayBlock.Height)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// We can now apply the migration and expect it not to fail.
|
|
applyMigration(
|
|
t, beforeMigration, afterMigration, populateBirthdayBlock,
|
|
false,
|
|
)
|
|
}
|
|
|
|
// TestMigrationPopulateBirthdayBlockEstimateTooFar ensures that the migration
|
|
// can properly detect a height estimate which the chain from our point of view
|
|
// has not yet reached.
|
|
func TestMigrationPopulateBirthdayBlockEstimateTooFar(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
const numBlocks = 1000
|
|
chainParams := chaincfg.MainNetParams
|
|
|
|
var expectedHeight int32
|
|
beforeMigration := func(ns walletdb.ReadWriteBucket) error {
|
|
// To test this migration, we'll start by writing to disk 999
|
|
// random blocks to simulate a synced chain with height 1000.
|
|
block := &BlockStamp{}
|
|
for i := int32(1); i < numBlocks; i++ {
|
|
block.Height = i
|
|
blockHash := bytes.Repeat([]byte(string(i)), 32)
|
|
copy(block.Hash[:], blockHash)
|
|
if err := PutSyncedTo(ns, block); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
// With the blocks inserted, we'll assume that the birthday
|
|
// block corresponds to the 900th block in the chain. To do
|
|
// this, we'd need to set our birthday timestamp to the
|
|
// estimated timestamp of a block that's 899 blocks after
|
|
// genesis. However, this will not work if the average block
|
|
// time is not 10 mins, which can throw off the height estimate
|
|
// with a height longer than the chain in the event of test
|
|
// networks (testnet, regtest, etc. and not fully synced
|
|
// wallets). Instead the migration should be able to handle this
|
|
// by subtracting a days worth of blocks until finding a block
|
|
// that it is aware of.
|
|
//
|
|
// We'll have the migration assume that our birthday is at block
|
|
// 1001 in the chain. Since this block doesn't exist from the
|
|
// database's point of view, a days worth of blocks will be
|
|
// subtracted from the estimate, which should give us a valid
|
|
// block height.
|
|
genesisTimestamp := chainParams.GenesisBlock.Header.Timestamp
|
|
delta := numBlocks * 10 * time.Minute
|
|
expectedHeight = numBlocks - 144
|
|
|
|
birthday := genesisTimestamp.Add(delta)
|
|
if err := putBirthday(ns, birthday); err != nil {
|
|
return err
|
|
}
|
|
|
|
// Finally, since the migration has not yet started, we should
|
|
// not be able to find the birthday block within the database.
|
|
_, err := FetchBirthdayBlock(ns)
|
|
if !IsError(err, ErrBirthdayBlockNotSet) {
|
|
return fmt.Errorf("expected ErrBirthdayBlockNotSet, "+
|
|
"got %v", err)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// After the migration has completed, we should see that the birthday
|
|
// block now exists and is set to the correct expected height.
|
|
afterMigration := func(ns walletdb.ReadWriteBucket) error {
|
|
birthdayBlock, err := FetchBirthdayBlock(ns)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
if birthdayBlock.Height != expectedHeight {
|
|
return fmt.Errorf("expected birthday block height %d, "+
|
|
"got %d", expectedHeight, birthdayBlock.Height)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// We can now apply the migration and expect it not to fail.
|
|
applyMigration(
|
|
t, beforeMigration, afterMigration, populateBirthdayBlock,
|
|
false,
|
|
)
|
|
}
|
|
|
|
// TestMigrationResetSyncedBlockToBirthday ensures that the wallet properly sees
|
|
// its synced to block as the birthday block after resetting it.
|
|
func TestMigrationResetSyncedBlockToBirthday(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
var birthdayBlock BlockStamp
|
|
beforeMigration := func(ns walletdb.ReadWriteBucket) error {
|
|
// To test this migration, we'll assume we're synced to a chain
|
|
// of 100 blocks, with our birthday being the 50th block.
|
|
block := &BlockStamp{}
|
|
for i := int32(1); i < 100; i++ {
|
|
block.Height = i
|
|
blockHash := bytes.Repeat([]byte(string(i)), 32)
|
|
copy(block.Hash[:], blockHash)
|
|
if err := PutSyncedTo(ns, block); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
const birthdayHeight = 50
|
|
birthdayHash, err := fetchBlockHash(ns, birthdayHeight)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
birthdayBlock = BlockStamp{
|
|
Hash: *birthdayHash, Height: birthdayHeight,
|
|
}
|
|
|
|
return putBirthdayBlock(ns, birthdayBlock)
|
|
}
|
|
|
|
afterMigration := func(ns walletdb.ReadWriteBucket) error {
|
|
// After the migration has succeeded, we should see that the
|
|
// database's synced block now reflects the birthday block.
|
|
syncedBlock, err := fetchSyncedTo(ns)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
if syncedBlock.Height != birthdayBlock.Height {
|
|
return fmt.Errorf("expected synced block height %d, "+
|
|
"got %d", birthdayBlock.Height,
|
|
syncedBlock.Height)
|
|
}
|
|
if !syncedBlock.Hash.IsEqual(&birthdayBlock.Hash) {
|
|
return fmt.Errorf("expected synced block height %v, "+
|
|
"got %v", birthdayBlock.Hash, syncedBlock.Hash)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// We can now apply the migration and expect it not to fail.
|
|
applyMigration(
|
|
t, beforeMigration, afterMigration, resetSyncedBlockToBirthday,
|
|
false,
|
|
)
|
|
}
|
|
|
|
// TestMigrationResetSyncedBlockToBirthdayWithNoBirthdayBlock ensures that we
|
|
// cannot reset our synced to block to our birthday block if one isn't
|
|
// available.
|
|
func TestMigrationResetSyncedBlockToBirthdayWithNoBirthdayBlock(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
// To replicate the scenario where the database is not aware of a
|
|
// birthday block, we won't set one. This should cause the migration to
|
|
// fail.
|
|
beforeMigration := func(walletdb.ReadWriteBucket) error {
|
|
return nil
|
|
}
|
|
afterMigration := func(walletdb.ReadWriteBucket) error {
|
|
return nil
|
|
}
|
|
applyMigration(
|
|
t, beforeMigration, afterMigration, resetSyncedBlockToBirthday,
|
|
true,
|
|
)
|
|
}
|
|
|
|
// TestMigrationStoreMaxReorgDepth ensures that the storeMaxReorgDepth migration
|
|
// works as expected under different sync scenarios.
|
|
func TestMigrationStoreMaxReorgDepth(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
testCases := []struct {
|
|
name string
|
|
numBlocks int32
|
|
}{
|
|
{
|
|
name: "genesis only",
|
|
numBlocks: 0,
|
|
},
|
|
{
|
|
name: "below max reorg depth",
|
|
numBlocks: MaxReorgDepth - 1,
|
|
},
|
|
{
|
|
name: "above max reorg depth",
|
|
numBlocks: MaxReorgDepth + 1,
|
|
},
|
|
{
|
|
name: "double max reorg depth",
|
|
numBlocks: MaxReorgDepth * 2,
|
|
},
|
|
}
|
|
|
|
for _, testCase := range testCases {
|
|
success := t.Run(testCase.name, func(t *testing.T) {
|
|
// We'll start the test by creating the number of blocks
|
|
// we'll add to the chain. We start from height 1 as the
|
|
// genesis block (height 0) is already included when the
|
|
// address manager is created.
|
|
blocks := make([]*BlockStamp, 0, testCase.numBlocks)
|
|
for i := int32(1); i <= testCase.numBlocks; i++ {
|
|
var hash chainhash.Hash
|
|
binary.BigEndian.PutUint32(hash[:], uint32(i))
|
|
blocks = append(blocks, &BlockStamp{
|
|
Hash: hash,
|
|
Height: i,
|
|
})
|
|
}
|
|
|
|
// Before the migration, we'll go ahead and add all of
|
|
// the blocks created. This simulates the behavior of an
|
|
// existing synced chain. We won't use PutSyncedTo as
|
|
// that would remove the stale entries on its own.
|
|
beforeMigration := func(ns walletdb.ReadWriteBucket) error {
|
|
if testCase.numBlocks == 0 {
|
|
return nil
|
|
}
|
|
|
|
// Write all the block hash entries.
|
|
for _, block := range blocks {
|
|
err := addBlockHash(
|
|
ns, block.Height, block.Hash,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
err = updateSyncedTo(ns, block)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
// Check to make sure they've been added
|
|
// properly.
|
|
for _, block := range blocks {
|
|
hash, err := fetchBlockHash(
|
|
ns, block.Height,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if *hash != block.Hash {
|
|
return fmt.Errorf("expected "+
|
|
"hash %v for height "+
|
|
"%v, got %v",
|
|
block.Hash,
|
|
block.Height, hash)
|
|
}
|
|
}
|
|
block, err := fetchSyncedTo(ns)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
expectedBlock := blocks[len(blocks)-1]
|
|
if block.Height != block.Height {
|
|
return fmt.Errorf("expected synced to "+
|
|
"block height %v, got %v",
|
|
expectedBlock.Height,
|
|
block.Height)
|
|
}
|
|
if block.Hash != block.Hash {
|
|
return fmt.Errorf("expected synced to "+
|
|
"block hash %v, got %v",
|
|
expectedBlock.Hash,
|
|
block.Hash)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// After the migration, we'll ensure we're unable to
|
|
// find all the block hashes that should have been
|
|
// removed.
|
|
afterMigration := func(ns walletdb.ReadWriteBucket) error {
|
|
maxStaleHeight := staleHeight(testCase.numBlocks)
|
|
for _, block := range blocks {
|
|
if block.Height <= maxStaleHeight {
|
|
_, err := fetchBlockHash(
|
|
ns, block.Height,
|
|
)
|
|
if IsError(err, ErrBlockNotFound) {
|
|
continue
|
|
}
|
|
return fmt.Errorf("expected "+
|
|
"ErrBlockNotFound for "+
|
|
"height %v, got %v",
|
|
block.Height, err)
|
|
}
|
|
|
|
hash, err := fetchBlockHash(
|
|
ns, block.Height,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if *hash != block.Hash {
|
|
return fmt.Errorf("expected "+
|
|
"hash %v for height "+
|
|
"%v, got %v",
|
|
block.Hash,
|
|
block.Height, hash)
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
applyMigration(
|
|
t, beforeMigration, afterMigration,
|
|
storeMaxReorgDepth, false,
|
|
)
|
|
})
|
|
if !success {
|
|
return
|
|
}
|
|
}
|
|
}
|