smallstep-certificates/scripts/badger-migration/main.go
Mariano Cano d9d7c52997
Add option to dry-run the migration
This commit adds an option that runs the migration on a virtual database
that doesn't do anything. This option can be used to see how many rows
there are.
2023-07-24 16:35:22 -07:00

353 lines
9.5 KiB
Go

package main
import (
"bytes"
"encoding/base64"
"encoding/binary"
"errors"
"flag"
"fmt"
"os"
"path/filepath"
badgerv1 "github.com/dgraph-io/badger"
badgerv2 "github.com/dgraph-io/badger/v2"
"github.com/smallstep/nosql"
)
var (
authorityTables = []string{
"x509_certs",
"x509_certs_data",
"revoked_x509_certs",
"x509_crl",
"revoked_ssh_certs",
"used_ott",
"ssh_certs",
"ssh_hosts",
"ssh_users",
"ssh_host_principals",
}
acmeTables = []string{
"acme_accounts",
"acme_keyID_accountID_index",
"acme_authzs",
"acme_challenges",
"nonces",
"acme_orders",
"acme_account_orders_index",
"acme_certs",
"acme_serial_certs_index",
"acme_external_account_keys",
"acme_external_account_keyID_reference_index",
"acme_external_account_keyID_provisionerID_index",
}
adminTables = []string{
"admins",
"provisioners",
"authority_policies",
}
)
type DB interface {
CreateTable([]byte) error
Set(bucket, key, value []byte) error
}
type dryRunDB struct{}
func (*dryRunDB) CreateTable([]byte) error { return nil }
func (*dryRunDB) Set(bucket, key, value []byte) error { return nil }
func usage(fs *flag.FlagSet) {
name := filepath.Base(os.Args[0])
fmt.Fprintf(os.Stderr, "%s is a tool to migrate data from BadgerDB to MySQL or PostgreSQL.\n", name)
fmt.Fprintln(os.Stderr, "\nUsage:")
fmt.Fprintf(os.Stderr, " %s [-v1|-v2] -dir=<path> [-value-dir=<path>] -type=type -database=<source>\n", name)
fmt.Fprintln(os.Stderr, "\nExamples:")
fmt.Fprintf(os.Stderr, " %s -v1 -dir /var/lib/step-ca/db -type=mysql -database \"user@unix/step_ca\"\n", name)
fmt.Fprintf(os.Stderr, " %s -v1 -dir /var/lib/step-ca/db -type=mysql -database \"user:password@tcp(localhost:3306)/step_ca\"\n", name)
fmt.Fprintf(os.Stderr, " %s -v2 -dir /var/lib/step-ca/db -type=postgresql -database \"user=postgres dbname=step_ca\"\n", name)
fmt.Fprintf(os.Stderr, " %s -v2 -dir /var/lib/step-ca/db -dry-run\"\n", name)
fmt.Fprintln(os.Stderr, "\nOptions:")
fs.PrintDefaults()
}
func main() {
var v1, v2, dryRun bool
var dir, valueDir string
var typ, database string
var key string
fs := flag.NewFlagSet(os.Args[0], flag.ExitOnError)
fs.BoolVar(&v1, "v1", false, "use badger v1 as the source database")
fs.BoolVar(&v2, "v2", false, "use badger v2 as the source database")
fs.StringVar(&dir, "dir", "", "badger database directory")
fs.StringVar(&valueDir, "value-dir", "", "badger database value directory")
fs.StringVar(&typ, "type", "", "the destination database type to use")
fs.StringVar(&database, "database", "", "the destination driver-specific data source name")
fs.StringVar(&key, "key", "", "the key used to resume the migration")
fs.BoolVar(&dryRun, "dry-run", false, "runs the migration scripts without writing anything")
fs.Usage = func() { usage(fs) }
fs.Parse(os.Args[1:])
switch {
case v1 == v2:
fatal("flag -v1 or -v2 are required")
case dir == "":
fatal("flag -dir is required")
case typ != "postgresql" && typ != "mysql" && !dryRun:
fatal(`flag -type must be "postgresql" or "mysql"`)
case database == "" && !dryRun:
fatal("flag --database required")
}
var (
err error
v1DB *badgerv1.DB
v2DB *badgerv2.DB
lastKey []byte
)
if key != "" {
if lastKey, err = base64.StdEncoding.DecodeString(key); err != nil {
fatal("error decoding key: %v", err)
}
}
if v1 {
if v1DB, err = badgerV1Open(dir, valueDir); err != nil {
fatal("error opening badger v1 database: %v", err)
}
} else {
if v2DB, err = badgerV2Open(dir, valueDir); err != nil {
fatal("error opening badger v2 database: %v", err)
}
}
var db DB
if dryRun {
db = &dryRunDB{}
} else {
db, err = nosql.New(typ, database)
if err != nil {
fatal("error opening %s database: %v", typ, err)
}
}
allTables := append([]string{}, authorityTables...)
allTables = append(allTables, acmeTables...)
allTables = append(allTables, adminTables...)
// Convert prefix names to badger key prefixes
badgerKeys := make([][]byte, len(allTables))
for i, name := range allTables {
badgerKeys[i], err = badgerEncode([]byte(name))
if err != nil {
fatal("error encoding table %s: %v", name, err)
}
}
for i, prefix := range badgerKeys {
table := allTables[i]
// With a key flag, resume from that table and prefix
if lastKey != nil {
bucket, _ := parseBadgerEncode(lastKey)
if table != string(bucket) {
fmt.Printf("skipping table %s\n", table)
continue
}
// Continue with a new prefix
prefix = lastKey
lastKey = nil
}
var n int64
fmt.Printf("migrating %s ...", table)
if err := db.CreateTable([]byte(table)); err != nil {
fatal("error creating table %s: %v", table, err)
}
if v1 {
if badgerKey, err := badgerV1Iterate(v1DB, prefix, func(bucket, key, value []byte) error {
n++
return db.Set(bucket, key, value)
}); err != nil {
fmt.Println()
fatal("error inserting into %s: %v\nLast key: %s", table, err, base64.StdEncoding.EncodeToString(badgerKey))
}
} else {
if badgerKey, err := badgerV2Iterate(v2DB, prefix, func(bucket, key, value []byte) error {
n++
return db.Set(bucket, key, value)
}); err != nil {
fmt.Println()
fatal("error inserting into %s: %v\nLast key: %s", table, err, base64.StdEncoding.EncodeToString(badgerKey))
}
}
fmt.Printf(" %d rows\n", n)
}
}
func fatal(format string, args ...any) {
fmt.Fprintf(os.Stderr, format, args...)
fmt.Fprintln(os.Stderr)
os.Exit(1)
}
func badgerV1Open(dir, valueDir string) (*badgerv1.DB, error) {
opts := badgerv1.DefaultOptions(dir)
if valueDir != "" {
opts.ValueDir = valueDir
}
return badgerv1.Open(opts)
}
func badgerV2Open(dir, valueDir string) (*badgerv2.DB, error) {
opts := badgerv2.DefaultOptions(dir)
if valueDir != "" {
opts.ValueDir = valueDir
}
return badgerv2.Open(opts)
}
type Iterator interface {
Seek([]byte)
ValidForPrefix([]byte) bool
Next()
}
type Item interface {
KeyCopy([]byte) []byte
ValueCopy([]byte) ([]byte, error)
}
func badgerV1Iterate(db *badgerv1.DB, prefix []byte, fn func(bucket, key, value []byte) error) (badgerKey []byte, err error) {
err = db.View(func(txn *badgerv1.Txn) error {
it := txn.NewIterator(badgerv1.DefaultIteratorOptions)
defer it.Close()
badgerKey, err = badgerIterate(it, prefix, fn)
return err
})
return
}
func badgerV2Iterate(db *badgerv2.DB, prefix []byte, fn func(bucket, key, value []byte) error) (badgerKey []byte, err error) {
err = db.View(func(txn *badgerv2.Txn) error {
it := txn.NewIterator(badgerv2.DefaultIteratorOptions)
defer it.Close()
badgerKey, err = badgerIterate(it, prefix, fn)
return err
})
return
}
func badgerIterate(it Iterator, prefix []byte, fn func(bucket, key, value []byte) error) ([]byte, error) {
var badgerKey []byte
for it.Seek(prefix); it.ValidForPrefix(prefix); it.Next() {
var item Item
switch itt := it.(type) {
case *badgerv1.Iterator:
item = itt.Item()
case *badgerv2.Iterator:
item = itt.Item()
default:
return badgerKey, fmt.Errorf("unexpected iterator type %T", it)
}
badgerKey = item.KeyCopy(nil)
if isBadgerTable(badgerKey) {
continue
}
bucket, key, err := fromBadgerKey(badgerKey)
if err != nil {
return badgerKey, fmt.Errorf("error converting from badger key %s", badgerKey)
}
value, err := item.ValueCopy(nil)
if err != nil {
return badgerKey, fmt.Errorf("error retrieving contents from database value: %w", err)
}
if err := fn(bucket, key, value); err != nil {
return badgerKey, fmt.Errorf("error exporting %s[%s]=%x", bucket, key, value)
}
}
return badgerKey, nil
}
// badgerEncode encodes a byte slice into a section of a BadgerKey. See
// documentation for toBadgerKey.
func badgerEncode(val []byte) ([]byte, error) {
l := len(val)
switch {
case l == 0:
return nil, errors.New("input cannot be empty")
case l > 65535:
return nil, errors.New("length of input cannot be greater than 65535")
default:
lb := new(bytes.Buffer)
if err := binary.Write(lb, binary.LittleEndian, uint16(l)); err != nil {
return nil, fmt.Errorf("error doing binary Write: %w", err)
}
return append(lb.Bytes(), val...), nil
}
}
// parseBadgerEncode decodes the badger key and returns the bucket and the rest.
func parseBadgerEncode(bk []byte) (value, rest []byte) {
var (
keyLen uint16
start = uint16(2)
length = uint16(len(bk))
)
if uint16(len(bk)) < start {
return nil, bk
}
// First 2 bytes stores the length of the value.
if err := binary.Read(bytes.NewReader(bk[:2]), binary.LittleEndian, &keyLen); err != nil {
return nil, bk
}
end := start + keyLen
switch {
case length < end:
return nil, bk
case length == end:
return bk[start:end], nil
default:
return bk[start:end], bk[end:]
}
}
// isBadgerTable returns True if the slice is a badgerTable token, false
// otherwise. badgerTable means that the slice contains only the [size|value] of
// one section of a badgerKey and no remainder. A badgerKey is [bucket|key],
// while a badgerTable is only the bucket section.
func isBadgerTable(bk []byte) bool {
if k, rest := parseBadgerEncode(bk); len(k) > 0 && len(rest) == 0 {
return true
}
return false
}
// fromBadgerKey returns the bucket and key encoded in a BadgerKey. See
// documentation for toBadgerKey.
func fromBadgerKey(bk []byte) ([]byte, []byte, error) {
bucket, rest := parseBadgerEncode(bk)
if len(bucket) == 0 || len(rest) == 0 {
return nil, nil, fmt.Errorf("invalid badger key: %v", bk)
}
key, rest2 := parseBadgerEncode(rest)
if len(key) == 0 || len(rest2) != 0 {
return nil, nil, fmt.Errorf("invalid badger key: %v", bk)
}
return bucket, key, nil
}