There is a danger of deadlock wherever we have database operations inside a database iterator. The iterator itself pins a connection, so the nested operations need another connection; if we've reached maxOpenConns then it blocks until a connection is released. If all connections are consumed by such iterators then none can make progress. Luckily, we have limited concurrency for most such iterator loops. They are part of scanning, pulling, reverting, etc where there is only ever one such routine per folder. The exception is block reuse in the copier routine, which is limited by the `Copiers` setting per folder. This could in practice deadlock since you could set copiers to eight and end up with six `AllLocalBlocksWithHash` iterators when `maxDBConns=6`, all of which need to make additional database calls inside the loop. This PR fixes the problem twice; - The problematic loop does not need to be reentrant. The set of blocks that may be returned by the iterator is finite so we can easily just collect them to a slice before we start processing them. This avoids the problem entirely. - We do not need to limit database connections as strictly as we currently do. Increase the maximum allowed, while reducing the number of held-open idle connections slightly. This is not an exact science, but ideally we want "most" operations to be able to use the pinned connections to avoid cache churn. Most operations are short lived queries, or single-goroutine iterators with short lived queries inside, so four connections seems like it should usually be enough. 🤷 - Additionally, set a cap on Copiers. Currently you could set it to an arbitrarily large number, which is not advantageous. Limit it to 2*NumCPU which scales somewhat with system performance. Signed-off-by: Jakob Borg <jakob@kastelo.net>
186 lines
4.8 KiB
Go
186 lines
4.8 KiB
Go
// Copyright (C) 2025 The Syncthing Authors.
|
|
//
|
|
// This Source Code Form is subject to the terms of the Mozilla Public
|
|
// License, v. 2.0. If a copy of the MPL was not distributed with this file,
|
|
// You can obtain one at https://mozilla.org/MPL/2.0/.
|
|
|
|
package sqlite
|
|
|
|
import (
|
|
"fmt"
|
|
"log/slog"
|
|
"os"
|
|
"path/filepath"
|
|
"runtime"
|
|
"sync"
|
|
"time"
|
|
|
|
"github.com/syncthing/syncthing/internal/db"
|
|
"github.com/syncthing/syncthing/internal/slogutil"
|
|
"github.com/syncthing/syncthing/lib/build"
|
|
)
|
|
|
|
const (
|
|
// maxIdleConns is set so that most db operations will usually fit
|
|
// within one of those connections and hence use their persistent page
|
|
// cache. Additional connections on top of these will allocate their own
|
|
// page cache (same as any other), but it will be deallocated on
|
|
// connection close.
|
|
maxIdleConns = 4
|
|
|
|
minDeleteRetention = 24 * time.Hour
|
|
)
|
|
|
|
// maxOpenConns is sized for handling spikes. The primary driver is the
|
|
// Copiers folder option which may result in up to 2*NumCPU concurrent
|
|
// iterations. We reserve additional space on top of this to serve
|
|
// additional operations, some of which may be reentrant (queries within
|
|
// iterators) without deadlock.
|
|
var maxOpenConns = max(16, 4*runtime.NumCPU())
|
|
|
|
type DB struct {
|
|
*baseDB
|
|
|
|
pathBase string
|
|
deleteRetention time.Duration
|
|
|
|
folderDBsMut sync.RWMutex
|
|
folderDBs map[string]*folderDB
|
|
folderDBOpener func(folder, path string, deleteRetention time.Duration) (*folderDB, error)
|
|
}
|
|
|
|
var _ db.DB = (*DB)(nil)
|
|
|
|
type Option func(*DB)
|
|
|
|
func WithDeleteRetention(d time.Duration) Option {
|
|
return func(s *DB) {
|
|
if d <= 0 {
|
|
s.deleteRetention = 0
|
|
} else {
|
|
s.deleteRetention = max(d, minDeleteRetention)
|
|
}
|
|
}
|
|
}
|
|
|
|
func Open(path string, opts ...Option) (*DB, error) {
|
|
pragmas := []string{
|
|
"journal_mode = WAL",
|
|
"optimize = 0x10002",
|
|
"auto_vacuum = INCREMENTAL",
|
|
fmt.Sprintf("application_id = %d", applicationIDMain),
|
|
}
|
|
schemas := []string{
|
|
"sql/schema/common/*",
|
|
"sql/schema/main/*",
|
|
}
|
|
migrations := []string{
|
|
"sql/migrations/common/*",
|
|
"sql/migrations/main/*",
|
|
}
|
|
|
|
_ = os.MkdirAll(path, 0o700)
|
|
initTmpDir(path)
|
|
|
|
mainPath := filepath.Join(path, "main.db")
|
|
mainBase, err := openBase(mainPath, maxOpenConns, maxIdleConns, pragmas, schemas, migrations)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
db := &DB{
|
|
pathBase: path,
|
|
baseDB: mainBase,
|
|
folderDBs: make(map[string]*folderDB),
|
|
folderDBOpener: openFolderDB,
|
|
}
|
|
|
|
for _, opt := range opts {
|
|
opt(db)
|
|
}
|
|
|
|
if err := db.cleanDroppedFolders(); err != nil {
|
|
slog.Warn("Failed to clean dropped folders", slogutil.Error(err))
|
|
}
|
|
|
|
if err := db.startFolderDatabases(); err != nil {
|
|
return nil, wrap(err)
|
|
}
|
|
|
|
return db, nil
|
|
}
|
|
|
|
// Open the database with options suitable for the migration inserts. This
|
|
// is not a safe mode of operation for normal processing, use only for bulk
|
|
// inserts with a close afterwards.
|
|
func OpenForMigration(path string) (*DB, error) {
|
|
pragmas := []string{
|
|
"journal_mode = OFF",
|
|
"foreign_keys = 0",
|
|
"synchronous = 0",
|
|
"locking_mode = EXCLUSIVE",
|
|
fmt.Sprintf("application_id = %d", applicationIDMain),
|
|
}
|
|
schemas := []string{
|
|
"sql/schema/common/*",
|
|
"sql/schema/main/*",
|
|
}
|
|
migrations := []string{
|
|
"sql/migrations/common/*",
|
|
"sql/migrations/main/*",
|
|
}
|
|
|
|
_ = os.MkdirAll(path, 0o700)
|
|
initTmpDir(path)
|
|
|
|
mainPath := filepath.Join(path, "main.db")
|
|
mainBase, err := openBase(mainPath, 1, 1, pragmas, schemas, migrations)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
db := &DB{
|
|
pathBase: path,
|
|
baseDB: mainBase,
|
|
folderDBs: make(map[string]*folderDB),
|
|
folderDBOpener: openFolderDBForMigration,
|
|
}
|
|
|
|
if err := db.cleanDroppedFolders(); err != nil {
|
|
slog.Warn("Failed to clean dropped folders", slogutil.Error(err))
|
|
}
|
|
|
|
return db, nil
|
|
}
|
|
|
|
func (s *DB) Close() error {
|
|
s.folderDBsMut.Lock()
|
|
defer s.folderDBsMut.Unlock()
|
|
for folder, fdb := range s.folderDBs {
|
|
fdb.Close()
|
|
delete(s.folderDBs, folder)
|
|
}
|
|
return wrap(s.baseDB.Close())
|
|
}
|
|
|
|
func initTmpDir(path string) {
|
|
if build.IsWindows || build.IsDarwin || os.Getenv("SQLITE_TMPDIR") != "" {
|
|
// Doesn't use SQLITE_TMPDIR, isn't likely to have a tiny
|
|
// ram-backed temp directory, or already set to something.
|
|
return
|
|
}
|
|
|
|
// Attempt to override the SQLite temporary directory by setting the
|
|
// env var prior to the (first) database being opened and hence
|
|
// SQLite becoming initialized. We set the temp dir to the same
|
|
// place we store the database, in the hope that there will be
|
|
// enough space there for the operations it needs to perform, as
|
|
// opposed to /tmp and similar, on some systems.
|
|
dbTmpDir := filepath.Join(path, ".tmp")
|
|
if err := os.MkdirAll(dbTmpDir, 0o700); err == nil {
|
|
os.Setenv("SQLITE_TMPDIR", dbTmpDir)
|
|
} else {
|
|
slog.Warn("Failed to create temp directory for SQLite", slogutil.FilePath(dbTmpDir), slogutil.Error(err))
|
|
}
|
|
}
|