reflector.go/db/db.go
Niko Storni c4084eeb68 improve disk cleanup
add index to is_stored
fix test
replace LRU cache
2021-04-29 03:41:18 +02:00

859 lines
21 KiB
Go

package db
import (
"context"
"database/sql"
"fmt"
"runtime"
"strings"
"time"
"github.com/lbryio/lbry.go/v2/dht/bits"
"github.com/lbryio/lbry.go/v2/extras/errors"
qt "github.com/lbryio/lbry.go/v2/extras/query"
"github.com/lbryio/lbry.go/v2/extras/stop"
"github.com/lbryio/lbry.go/v2/stream"
"github.com/go-sql-driver/mysql"
_ "github.com/go-sql-driver/mysql" // blank import for db driver ensures its imported even if its not used
log "github.com/sirupsen/logrus"
"github.com/volatiletech/null"
"go.uber.org/atomic"
)
// SdBlob is a special blob that contains information on the rest of the blobs in the stream
type SdBlob struct {
StreamName string `json:"stream_name"`
Blobs []struct {
Length int `json:"length"`
BlobNum int `json:"blob_num"`
BlobHash string `json:"blob_hash,omitempty"`
IV string `json:"iv"`
} `json:"blobs"`
StreamType string `json:"stream_type"`
Key string `json:"key"`
SuggestedFileName string `json:"suggested_file_name"`
StreamHash string `json:"stream_hash"`
}
type trackAccess int
const (
// Don't track accesses
TrackAccessNone trackAccess = iota
// Track accesses at the stream level
TrackAccessStreams
// Track accesses at the blob level
TrackAccessBlobs
)
// SQL implements the DB interface
type SQL struct {
conn *sql.DB
// Track the approx last time a blob or stream was accessed
TrackAccess trackAccess
// Instead of deleting a blob, marked it as not stored in the db
SoftDelete bool
// Log executed queries. qt.InterpolateParams is cpu-heavy. This avoids that call if not needed.
LogQueries bool
}
func (s SQL) logQuery(query string, args ...interface{}) {
if !s.LogQueries {
return
}
qStr, err := qt.InterpolateParams(query, args...)
if err != nil {
log.Errorln(err)
} else {
log.Debugln(qStr)
}
}
// Connect will create a connection to the database
func (s *SQL) Connect(dsn string) error {
var err error
// interpolateParams is necessary. otherwise uploading a stream with thousands of blobs
// will hit MySQL's max_prepared_stmt_count limit because the prepared statements are all
// opened inside a transaction. closing them manually doesn't seem to help
dsn += "?parseTime=1&collation=utf8mb4_unicode_ci&interpolateParams=1"
s.conn, err = sql.Open("mysql", dsn)
if err != nil {
return errors.Err(err)
}
s.conn.SetMaxIdleConns(12)
return errors.Err(s.conn.Ping())
}
// AddBlob adds a blob to the database.
func (s *SQL) AddBlob(hash string, length int, isStored bool) error {
if s.conn == nil {
return errors.Err("not connected")
}
_, err := s.insertBlob(hash, length, isStored)
return err
}
// AddBlob adds a blob to the database.
func (s *SQL) AddBlobs(hash []string) error {
if s.conn == nil {
return errors.Err("not connected")
}
// Split the slice into batches of 20 items.
batch := 10000
totalBlobs := int64(len(hash))
work := make(chan []string, 1000)
stopper := stop.New()
var totalInserted atomic.Int64
start := time.Now()
go func() {
for i := 0; i < len(hash); i += batch {
j := i + batch
if j > len(hash) {
j = len(hash)
}
work <- hash[i:j]
}
log.Infof("done loading %d hashes in the work queue", len(hash))
close(work)
}()
for i := 0; i < runtime.NumCPU(); i++ {
stopper.Add(1)
go func(worker int) {
log.Infof("starting worker %d", worker)
defer stopper.Done()
for hashes := range work {
inserted := totalInserted.Load()
remaining := totalBlobs - inserted
if inserted > 0 {
timePerBlob := time.Since(start).Microseconds() / inserted
remainingTime := time.Duration(remaining*timePerBlob) * time.Microsecond
log.Infof("[T%d] processing batch of %d items. ETA: %s", worker, len(hashes), remainingTime.String())
}
err := s.insertBlobs(hashes) // Process the batch.
if err != nil {
log.Errorf("error while inserting batch: %s", errors.FullTrace(err))
}
totalInserted.Add(int64(len(hashes)))
}
}(i)
}
stopper.Wait()
return nil
}
func (s *SQL) insertBlobs(hashes []string) error {
var (
q string
//args []interface{}
)
dayAgo := time.Now().AddDate(0, 0, -1).Format("2006-01-02 15:04:05")
q = "insert into blob_ (hash, is_stored, length, last_accessed_at) values "
for _, hash := range hashes {
// prepared statements slow everything down by a lot due to reflection
// for this specific instance we'll go ahead and hardcode the query to make it go faster
q += fmt.Sprintf("('%s',1,%d,'%s'),", hash, stream.MaxBlobSize, dayAgo)
//args = append(args, hash, true, stream.MaxBlobSize, dayAgo)
}
q = strings.TrimSuffix(q, ",")
_, err := s.exec(q)
if err != nil {
return err
}
return nil
}
func (s *SQL) insertBlob(hash string, length int, isStored bool) (int64, error) {
if length <= 0 {
return 0, errors.Err("length must be positive")
}
var (
q string
args []interface{}
)
if s.TrackAccess == TrackAccessBlobs {
args = []interface{}{hash, isStored, length, time.Now()}
q = "INSERT INTO blob_ (hash, is_stored, length, last_accessed_at) VALUES (" + qt.Qs(len(args)) + ") ON DUPLICATE KEY UPDATE is_stored = (is_stored or VALUES(is_stored)), last_accessed_at = VALUES(last_accessed_at)"
} else {
args = []interface{}{hash, isStored, length}
q = "INSERT INTO blob_ (hash, is_stored, length) VALUES (" + qt.Qs(len(args)) + ") ON DUPLICATE KEY UPDATE is_stored = (is_stored or VALUES(is_stored))"
}
blobID, err := s.exec(q, args...)
if err != nil {
return 0, err
}
if blobID == 0 {
err = s.conn.QueryRow("SELECT id FROM blob_ WHERE hash = ?", hash).Scan(&blobID)
if err != nil {
return 0, errors.Err(err)
}
if blobID == 0 {
return 0, errors.Err("blob ID is 0 even after INSERTing and SELECTing")
}
if s.TrackAccess == TrackAccessBlobs {
err := s.touchBlobs([]uint64{uint64(blobID)})
if err != nil {
return 0, errors.Err(err)
}
}
}
return blobID, nil
}
func (s *SQL) insertStream(hash string, sdBlobID int64) (int64, error) {
var (
q string
args []interface{}
)
if s.TrackAccess == TrackAccessStreams {
args = []interface{}{hash, sdBlobID, time.Now()}
q = "INSERT IGNORE INTO stream (hash, sd_blob_id, last_accessed_at) VALUES (" + qt.Qs(len(args)) + ")"
} else {
args = []interface{}{hash, sdBlobID}
q = "INSERT IGNORE INTO stream (hash, sd_blob_id) VALUES (" + qt.Qs(len(args)) + ")"
}
streamID, err := s.exec(q, args...)
if err != nil {
return 0, errors.Err(err)
}
if streamID == 0 {
err = s.conn.QueryRow("SELECT id FROM stream WHERE sd_blob_id = ?", sdBlobID).Scan(&streamID)
if err != nil {
return 0, errors.Err(err)
}
if streamID == 0 {
return 0, errors.Err("stream ID is 0 even after INSERTing and SELECTing")
}
if s.TrackAccess == TrackAccessStreams {
err := s.touchStreams([]uint64{uint64(streamID)})
if err != nil {
return 0, errors.Err(err)
}
}
}
return streamID, nil
}
// HasBlob checks if the database contains the blob information.
func (s *SQL) HasBlob(hash string, touch bool) (bool, error) {
exists, err := s.HasBlobs([]string{hash}, touch)
if err != nil {
return false, err
}
return exists[hash], nil
}
// HasBlobs checks if the database contains the set of blobs and returns a bool map.
func (s *SQL) HasBlobs(hashes []string, touch bool) (map[string]bool, error) {
exists, idsNeedingTouch, err := s.hasBlobs(hashes)
if touch {
if s.TrackAccess == TrackAccessBlobs {
s.touchBlobs(idsNeedingTouch)
} else if s.TrackAccess == TrackAccessStreams {
s.touchStreams(idsNeedingTouch)
}
}
return exists, err
}
func (s *SQL) touchBlobs(blobIDs []uint64) error {
if len(blobIDs) == 0 {
return nil
}
query := "UPDATE blob_ SET last_accessed_at = ? WHERE id IN (" + qt.Qs(len(blobIDs)) + ")"
args := make([]interface{}, len(blobIDs)+1)
args[0] = time.Now()
for i := range blobIDs {
args[i+1] = blobIDs[i]
}
startTime := time.Now()
_, err := s.exec(query, args...)
log.Debugf("touched %d blobs and took %s", len(blobIDs), time.Since(startTime))
return errors.Err(err)
}
func (s *SQL) touchStreams(streamIDs []uint64) error {
if len(streamIDs) == 0 {
return nil
}
query := "UPDATE stream SET last_accessed_at = ? WHERE id IN (" + qt.Qs(len(streamIDs)) + ")"
args := make([]interface{}, len(streamIDs)+1)
args[0] = time.Now()
for i := range streamIDs {
args[i+1] = streamIDs[i]
}
startTime := time.Now()
_, err := s.exec(query, args...)
log.Debugf("touched %d streams and took %s", len(streamIDs), time.Since(startTime))
return errors.Err(err)
}
func (s *SQL) hasBlobs(hashes []string) (map[string]bool, []uint64, error) {
if s.conn == nil {
return nil, nil, errors.Err("not connected")
}
var (
hash string
blobID uint64
streamID null.Uint64
lastAccessedAt null.Time
)
var needsTouch []uint64
exists := make(map[string]bool)
touchDeadline := time.Now().Add(-6 * time.Hour) // touch blob if last accessed before this time
maxBatchSize := 10000
doneIndex := 0
for len(hashes) > doneIndex {
sliceEnd := doneIndex + maxBatchSize
if sliceEnd > len(hashes) {
sliceEnd = len(hashes)
}
log.Debugf("getting hashes[%d:%d] of %d", doneIndex, sliceEnd, len(hashes))
batch := hashes[doneIndex:sliceEnd]
var query string
if s.TrackAccess == TrackAccessBlobs {
query = `SELECT b.hash, b.id, NULL, b.last_accessed_at
FROM blob_ b
WHERE b.is_stored = 1 and b.hash IN (` + qt.Qs(len(batch)) + `)`
} else if s.TrackAccess == TrackAccessStreams {
query = `SELECT b.hash, b.id, s.id, s.last_accessed_at
FROM blob_ b
LEFT JOIN stream_blob sb ON b.id = sb.blob_id
INNER JOIN stream s on (sb.stream_id = s.id or s.sd_blob_id = b.id)
WHERE b.is_stored = 1 and b.hash IN (` + qt.Qs(len(batch)) + `)`
} else {
query = `SELECT b.hash, b.id, NULL, NULL
FROM blob_ b
WHERE b.is_stored = 1 and b.hash IN (` + qt.Qs(len(batch)) + `)`
}
args := make([]interface{}, len(batch))
for i := range batch {
args[i] = batch[i]
}
s.logQuery(query, args...)
err := func() error {
startTime := time.Now()
rows, err := s.conn.Query(query, args...)
log.Debugf("hashes query took %s", time.Since(startTime))
if err != nil {
return errors.Err(err)
}
defer closeRows(rows)
for rows.Next() {
err := rows.Scan(&hash, &blobID, &streamID, &lastAccessedAt)
if err != nil {
return errors.Err(err)
}
exists[hash] = true
if !lastAccessedAt.Valid || lastAccessedAt.Time.Before(touchDeadline) {
if s.TrackAccess == TrackAccessBlobs {
needsTouch = append(needsTouch, blobID)
} else if s.TrackAccess == TrackAccessStreams && !streamID.IsZero() {
needsTouch = append(needsTouch, streamID.Uint64)
}
}
}
err = rows.Err()
if err != nil {
return errors.Err(err)
}
doneIndex += len(batch)
return nil
}()
if err != nil {
return nil, nil, err
}
}
return exists, needsTouch, nil
}
// Delete will remove (or soft-delete) the blob from the db
// NOTE: If SoftDelete is enabled, streams will never be deleted
func (s *SQL) Delete(hash string) error {
if s.SoftDelete {
_, err := s.exec("UPDATE blob_ SET is_stored = 0 WHERE hash = ?", hash)
return errors.Err(err)
}
_, err := s.exec("DELETE FROM stream WHERE sd_blob_id = (SELECT id FROM blob_ WHERE hash = ?)", hash)
if err != nil {
return errors.Err(err)
}
_, err = s.exec("DELETE FROM blob_ WHERE hash = ?", hash)
return errors.Err(err)
}
// GetHashRange gets the smallest and biggest hashes in the db
func (s *SQL) LeastRecentlyAccessedHashes(maxBlobs int) ([]string, error) {
if s.conn == nil {
return nil, errors.Err("not connected")
}
if s.TrackAccess != TrackAccessBlobs {
return nil, errors.Err("blob access tracking is disabled")
}
query := "SELECT hash from blob_ where is_stored = 1 order by last_accessed_at limit ?"
s.logQuery(query, maxBlobs)
rows, err := s.conn.Query(query, maxBlobs)
if err != nil {
return nil, errors.Err(err)
}
defer closeRows(rows)
blobs := make([]string, 0, maxBlobs)
for rows.Next() {
var hash string
err := rows.Scan(&hash)
if err != nil {
return nil, errors.Err(err)
}
blobs = append(blobs, hash)
}
return blobs, nil
}
// AllHashes writes all hashes from the db into the channel.
// It does not close the channel when it finishes.
//func (s *SQL) AllHashes(ch chan<- string) error {
// if s.conn == nil {
// return errors.Err("not connected")
// }
//
// query := "SELECT hash from blob_"
// if s.SoftDelete {
// query += " where is_stored = 1"
// }
// s.logQuery(query)
//
// rows, err := s.conn.Query(query)
// if err != nil {
// return errors.Err(err)
// }
// defer closeRows(rows)
//
// for rows.Next() {
// var hash string
// err := rows.Scan(&hash)
// if err != nil {
// return errors.Err(err)
// }
// ch <- hash
// // TODO: this needs testing
// // TODO: need a way to cancel this early (e.g. in case of shutdown)
// }
//
// close(ch)
// return nil
//}
func (s *SQL) Count() (int, error) {
if s.conn == nil {
return 0, errors.Err("not connected")
}
query := "SELECT count(id) from blob_"
if s.SoftDelete {
query += " where is_stored = 1"
}
s.logQuery(query)
var count int
err := s.conn.QueryRow(query).Scan(&count)
return count, errors.Err(err)
}
// Block will mark a blob as blocked
func (s *SQL) Block(hash string) error {
query := "INSERT IGNORE INTO blocked SET hash = ?"
args := []interface{}{hash}
s.logQuery(query, args...)
_, err := s.conn.Exec(query, args...)
return errors.Err(err)
}
// GetBlocked will return a list of blocked hashes
func (s *SQL) GetBlocked() (map[string]bool, error) {
query := "SELECT hash FROM blocked"
s.logQuery(query)
rows, err := s.conn.Query(query)
if err != nil {
return nil, errors.Err(err)
}
defer closeRows(rows)
blocked := make(map[string]bool)
var hash string
for rows.Next() {
err := rows.Scan(&hash)
if err != nil {
return nil, errors.Err(err)
}
blocked[hash] = true
}
err = rows.Err()
if err != nil {
return nil, errors.Err(err)
}
return blocked, nil
}
// MissingBlobsForKnownStream returns missing blobs for an existing stream
// WARNING: if the stream does NOT exist, no blob hashes will be returned, which looks
// like no blobs are missing
func (s *SQL) MissingBlobsForKnownStream(sdHash string) ([]string, error) {
if s.conn == nil {
return nil, errors.Err("not connected")
}
query := `
SELECT b.hash FROM blob_ b
INNER JOIN stream_blob sb ON b.id = sb.blob_id
INNER JOIN stream s ON s.id = sb.stream_id
INNER JOIN blob_ sdb ON sdb.id = s.sd_blob_id AND sdb.hash = ?
WHERE b.is_stored = 0
`
args := []interface{}{sdHash}
s.logQuery(query, args...)
rows, err := s.conn.Query(query, args...)
if err != nil {
return nil, errors.Err(err)
}
defer closeRows(rows)
var missingBlobs []string
var hash string
for rows.Next() {
err := rows.Scan(&hash)
if err != nil {
return nil, errors.Err(err)
}
missingBlobs = append(missingBlobs, hash)
}
err = rows.Err()
if err != nil {
return nil, errors.Err(err)
}
return missingBlobs, errors.Err(err)
}
// AddSDBlob insert the SD blob and all the content blobs. The content blobs are marked as "not stored",
// but they are tracked so reflector knows what it is missing.
func (s *SQL) AddSDBlob(sdHash string, sdBlobLength int, sdBlob SdBlob) error {
if s.conn == nil {
return errors.Err("not connected")
}
sdBlobID, err := s.insertBlob(sdHash, sdBlobLength, true)
if err != nil {
return err
}
streamID, err := s.insertStream(sdBlob.StreamHash, sdBlobID)
if err != nil {
return err
}
// insert content blobs and connect them to stream
for _, contentBlob := range sdBlob.Blobs {
if contentBlob.BlobHash == "" {
// null terminator blob
continue
}
blobID, err := s.insertBlob(contentBlob.BlobHash, contentBlob.Length, false)
if err != nil {
return err
}
args := []interface{}{streamID, blobID, contentBlob.BlobNum}
_, err = s.exec(
"INSERT IGNORE INTO stream_blob (stream_id, blob_id, num) VALUES ("+qt.Qs(len(args))+")",
args...,
)
if err != nil {
return errors.Err(err)
}
}
return nil
}
// GetHashRange gets the smallest and biggest hashes in the db
func (s *SQL) GetHashRange() (string, string, error) {
var min string
var max string
if s.conn == nil {
return "", "", errors.Err("not connected")
}
query := "SELECT MIN(hash), MAX(hash) from blob_"
s.logQuery(query)
err := s.conn.QueryRow(query).Scan(&min, &max)
return min, max, err
}
// GetStoredHashesInRange gets stored blobs with hashes in a given range, and sends the hashes into a channel
func (s *SQL) GetStoredHashesInRange(ctx context.Context, start, end bits.Bitmap) (ch chan bits.Bitmap, ech chan error) {
ch = make(chan bits.Bitmap)
ech = make(chan error)
// TODO: needs waitgroup?
go func() {
defer close(ch)
defer close(ech)
if s.conn == nil {
ech <- errors.Err("not connected")
return
}
query := "SELECT hash FROM blob_ WHERE hash >= ? AND hash <= ? AND is_stored = 1"
args := []interface{}{start.Hex(), end.Hex()}
s.logQuery(query, args...)
rows, err := s.conn.Query(query, args...)
defer closeRows(rows)
if err != nil {
ech <- err
return
}
var hash string
ScanLoop:
for rows.Next() {
err := rows.Scan(&hash)
if err != nil {
ech <- err
return
}
select {
case <-ctx.Done():
break ScanLoop
case ch <- bits.FromHexP(hash):
}
}
err = rows.Err()
if err != nil {
ech <- err
return
}
}()
return
}
// txFunc is a function that can be wrapped in a transaction
type txFunc func(tx *sql.Tx) error
// withTx wraps a function in an sql transaction. the transaction is committed if there's no error, or rolled back if there is one.
// if dbOrTx is an sql.DB, a new transaction is started
func withTx(dbOrTx interface{}, f txFunc) (err error) {
var tx *sql.Tx
switch t := dbOrTx.(type) {
case *sql.Tx:
tx = t
case *sql.DB:
tx, err = t.Begin()
if err != nil {
return err
}
defer func() {
if p := recover(); p != nil {
if rollBackError := tx.Rollback(); rollBackError != nil {
log.Error("failed to rollback tx on panic - ", rollBackError)
}
panic(p)
} else if err != nil {
if rollBackError := tx.Rollback(); rollBackError != nil {
log.Error("failed to rollback tx on panic - ", rollBackError)
}
} else {
err = errors.Err(tx.Commit())
}
}()
default:
return errors.Err("db or tx required")
}
return f(tx)
}
func closeRows(rows *sql.Rows) {
if rows != nil {
err := rows.Close()
if err != nil {
log.Error("error closing rows: ", err)
}
}
}
func (s *SQL) exec(query string, args ...interface{}) (int64, error) {
s.logQuery(query, args...)
attempt, maxAttempts := 0, 3
Retry:
attempt++
result, err := s.conn.Exec(query, args...)
if isLockTimeoutError(err) {
if attempt <= maxAttempts {
//Error 1205: Lock wait timeout exceeded; try restarting transaction
goto Retry
}
err = errors.Prefix("Lock timeout for query "+query, err)
}
if err != nil {
return 0, errors.Err(err)
}
lastID, err := result.LastInsertId()
return lastID, errors.Err(err)
}
func isLockTimeoutError(err error) bool {
e, ok := err.(*mysql.MySQLError)
return ok && e != nil && e.Number == 1205
}
/* SQL schema
in prod, set tx_isolation to READ-COMMITTED to improve db performance
make sure you use latin1 or utf8 charset, NOT utf8mb4. that's a waste of space.
todo: could add UNIQUE KEY (stream_hash, num) to stream_blob ...
CREATE TABLE blob_ (
id BIGINT UNSIGNED NOT NULL AUTO_INCREMENT UNIQUE,
hash char(96) NOT NULL,
is_stored TINYINT(1) NOT NULL DEFAULT 0,
length bigint(20) unsigned DEFAULT NULL,
last_accessed_at TIMESTAMP NULL DEFAULT NULL,
PRIMARY KEY (id),
UNIQUE KEY blob_hash_idx (hash),
KEY `blob_last_accessed_idx` (`last_accessed_at`),
KEY `is_stored_idx` (`is_stored`)
);
CREATE TABLE stream (
id BIGINT UNSIGNED NOT NULL AUTO_INCREMENT UNIQUE,
hash char(96) NOT NULL,
sd_blob_id BIGINT UNSIGNED NOT NULL,
last_accessed_at TIMESTAMP NULL DEFAULT NULL,
PRIMARY KEY (id),
UNIQUE KEY stream_hash_idx (hash),
KEY stream_sd_blob_id_idx (sd_blob_id),
KEY last_accessed_at_idx (last_accessed_at),
FOREIGN KEY (sd_blob_id) REFERENCES blob_ (id) ON DELETE RESTRICT ON UPDATE CASCADE
);
CREATE TABLE stream_blob (
stream_id BIGINT UNSIGNED NOT NULL,
blob_id BIGINT UNSIGNED NOT NULL,
num int NOT NULL,
PRIMARY KEY (stream_id, blob_id),
KEY stream_blob_blob_id_idx (blob_id),
FOREIGN KEY (stream_id) REFERENCES stream (id) ON DELETE CASCADE ON UPDATE CASCADE,
FOREIGN KEY (blob_id) REFERENCES blob_ (id) ON DELETE CASCADE ON UPDATE CASCADE
);
CREATE TABLE blocked (
hash char(96) NOT NULL,
PRIMARY KEY (hash)
);
*/
//func (d *LiteDBBackedStore) selfClean() {
// d.stopper.Add(1)
// defer d.stopper.Done()
// lastCleanup := time.Now()
// const cleanupInterval = 10 * time.Second
// for {
// select {
// case <-d.stopper.Ch():
// log.Infoln("stopping self cleanup")
// return
// default:
// time.Sleep(1 * time.Second)
// }
// if time.Since(lastCleanup) < cleanupInterval {
// continue
//
// blobsCount, err := d.db.BlobsCount()
// if err != nil {
// log.Errorf(errors.FullTrace(err))
// }
// if blobsCount >= d.maxItems {
// itemsToDelete := blobsCount / 100 * 10
// blobs, err := d.db.GetLRUBlobs(itemsToDelete)
// if err != nil {
// log.Errorf(errors.FullTrace(err))
// }
// for _, hash := range blobs {
// select {
// case <-d.stopper.Ch():
// return
// default:
//
// }
// err = d.Delete(hash)
// if err != nil {
// log.Errorf(errors.FullTrace(err))
// }
// metrics.CacheLRUEvictCount.With(metrics.CacheLabels(d.Name(), d.component)).Inc()
// }
// }
// lastCleanup = time.Now()
// }
//}