[#1367] writecache: Drop bbolt DB

Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
This commit is contained in:
Dmitrii Stepanov 2024-09-09 18:37:06 +03:00
parent 66e17f4b8e
commit 5f6c7cbdb1
12 changed files with 82 additions and 415 deletions

View file

@ -542,6 +542,6 @@ const (
StartedWritecacheSealAsync = "started writecache seal async" StartedWritecacheSealAsync = "started writecache seal async"
WritecacheSealCompletedAsync = "writecache seal completed successfully" WritecacheSealCompletedAsync = "writecache seal completed successfully"
FailedToSealWritecacheAsync = "failed to seal writecache async" FailedToSealWritecacheAsync = "failed to seal writecache async"
WritecacheShrinkSkippedNotEmpty = "writecache shrink skipped: database is not empty" WritecacheShrinkSkippedNotEmpty = "writecache shrink skipped: not empty"
BlobovniczatreeFailedToRemoveRebuildTempFile = "failed to remove rebuild temp file" BlobovniczatreeFailedToRemoveRebuildTempFile = "failed to remove rebuild temp file"
) )

View file

@ -2,7 +2,7 @@ package writecache
import ( import (
"context" "context"
"os" "fmt"
"sync" "sync"
"sync/atomic" "sync/atomic"
@ -27,8 +27,6 @@ type cache struct {
cancel atomic.Value cancel atomic.Value
// wg is a wait group for flush workers. // wg is a wait group for flush workers.
wg sync.WaitGroup wg sync.WaitGroup
// store contains underlying database.
store
// fsTree contains big files stored directly on file-system. // fsTree contains big files stored directly on file-system.
fsTree *fstree.FSTree fsTree *fstree.FSTree
} }
@ -67,7 +65,6 @@ func New(opts ...Option) Cache {
maxCacheSize: defaultMaxCacheSize, maxCacheSize: defaultMaxCacheSize,
maxBatchSize: bbolt.DefaultMaxBatchSize, maxBatchSize: bbolt.DefaultMaxBatchSize,
maxBatchDelay: bbolt.DefaultMaxBatchDelay, maxBatchDelay: bbolt.DefaultMaxBatchDelay,
openFile: os.OpenFile,
metrics: DefaultMetrics(), metrics: DefaultMetrics(),
}, },
} }
@ -102,13 +99,15 @@ func (c *cache) Open(_ context.Context, mod mode.Mode) error {
if err != nil { if err != nil {
return metaerr.Wrap(err) return metaerr.Wrap(err)
} }
return metaerr.Wrap(c.initCounters()) return metaerr.Wrap(c.initCounters())
} }
// Init runs necessary services. // Init runs necessary services.
func (c *cache) Init() error { func (c *cache) Init() error {
c.metrics.SetMode(mode.ConvertToComponentModeDegraded(c.mode)) c.metrics.SetMode(mode.ConvertToComponentModeDegraded(c.mode))
if err := c.flushAndDropBBoltDB(context.Background()); err != nil {
return fmt.Errorf("flush previous version write-cache database: %w", err)
}
ctx, cancel := context.WithCancel(context.Background()) ctx, cancel := context.WithCancel(context.Background())
c.cancel.Store(cancel) c.cancel.Store(cancel)
c.runFlushLoop(ctx) c.runFlushLoop(ctx)
@ -132,10 +131,10 @@ func (c *cache) Close() error {
defer c.modeMtx.Unlock() defer c.modeMtx.Unlock()
var err error var err error
if c.db != nil { if c.fsTree != nil {
err = c.db.Close() err = c.fsTree.Close()
if err != nil { if err != nil {
c.db = nil c.fsTree = nil
} }
} }
c.metrics.Close() c.metrics.Close()

View file

@ -2,7 +2,6 @@ package writecache
import ( import (
"context" "context"
"math"
"time" "time"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/blobstor/common" "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/blobstor/common"
@ -10,7 +9,6 @@ import (
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/internal/metaerr" "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/internal/metaerr"
"git.frostfs.info/TrueCloudLab/frostfs-observability/tracing" "git.frostfs.info/TrueCloudLab/frostfs-observability/tracing"
oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id" oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id"
"go.etcd.io/bbolt"
"go.opentelemetry.io/otel/attribute" "go.opentelemetry.io/otel/attribute"
"go.opentelemetry.io/otel/trace" "go.opentelemetry.io/otel/trace"
) )
@ -45,46 +43,11 @@ func (c *cache) Delete(ctx context.Context, addr oid.Address) error {
return ErrDegraded return ErrDegraded
} }
saddr := addr.EncodeToString()
var dataSize int
_ = c.db.View(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
dataSize = len(b.Get([]byte(saddr)))
return nil
})
if dataSize > 0 {
storageType = StorageTypeDB
var recordDeleted bool
err := c.db.Update(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
key := []byte(saddr)
recordDeleted = b.Get(key) != nil
err := b.Delete(key)
return err
})
if err != nil {
return err
}
storagelog.Write(c.log,
storagelog.AddressField(saddr),
storagelog.StorageTypeField(wcStorageType),
storagelog.OpField("db DELETE"),
)
if recordDeleted {
c.objCounters.cDB.Add(math.MaxUint64)
c.estimateCacheSize()
}
deleted = true
return nil
}
storageType = StorageTypeFSTree storageType = StorageTypeFSTree
_, err := c.fsTree.Delete(ctx, common.DeletePrm{Address: addr}) _, err := c.fsTree.Delete(ctx, common.DeletePrm{Address: addr})
if err == nil { if err == nil {
storagelog.Write(c.log, storagelog.Write(c.log,
storagelog.AddressField(saddr), storagelog.AddressField(addr.EncodeToString()),
storagelog.StorageTypeField(wcStorageType), storagelog.StorageTypeField(wcStorageType),
storagelog.OpField("fstree DELETE"), storagelog.OpField("fstree DELETE"),
) )

View file

@ -4,6 +4,9 @@ import (
"bytes" "bytes"
"context" "context"
"errors" "errors"
"fmt"
"os"
"path/filepath"
"time" "time"
"git.frostfs.info/TrueCloudLab/frostfs-node/internal/logs" "git.frostfs.info/TrueCloudLab/frostfs-node/internal/logs"
@ -16,7 +19,6 @@ import (
"git.frostfs.info/TrueCloudLab/frostfs-observability/tracing" "git.frostfs.info/TrueCloudLab/frostfs-observability/tracing"
objectSDK "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object" objectSDK "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object"
oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id" oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id"
"github.com/mr-tron/base58"
"go.etcd.io/bbolt" "go.etcd.io/bbolt"
"go.opentelemetry.io/otel/attribute" "go.opentelemetry.io/otel/attribute"
"go.opentelemetry.io/otel/trace" "go.opentelemetry.io/otel/trace"
@ -24,10 +26,6 @@ import (
) )
const ( const (
// flushBatchSize is amount of keys which will be read from cache to be flushed
// to the main storage. It is used to reduce contention between cache put
// and cache persist.
flushBatchSize = 512
// defaultFlushWorkersCount is number of workers for putting objects in main storage. // defaultFlushWorkersCount is number of workers for putting objects in main storage.
defaultFlushWorkersCount = 20 defaultFlushWorkersCount = 20
// defaultFlushInterval is default time interval between successive flushes. // defaultFlushInterval is default time interval between successive flushes.
@ -41,112 +39,11 @@ func (c *cache) runFlushLoop(ctx context.Context) {
if c.disableBackgroundFlush { if c.disableBackgroundFlush {
return return
} }
for range c.workersCount {
c.wg.Add(1)
go c.workerFlushSmall(ctx)
}
c.wg.Add(1) c.wg.Add(1)
go func() { go func() {
c.workerFlushBig(ctx) c.workerFlushBig(ctx)
c.wg.Done() c.wg.Done()
}() }()
c.wg.Add(1)
go func() {
defer c.wg.Done()
tt := time.NewTimer(defaultFlushInterval)
defer tt.Stop()
for {
select {
case <-tt.C:
c.flushSmallObjects(ctx)
tt.Reset(defaultFlushInterval)
c.estimateCacheSize()
case <-ctx.Done():
return
}
}
}()
}
func (c *cache) flushSmallObjects(ctx context.Context) {
var lastKey []byte
for {
select {
case <-ctx.Done():
return
default:
}
var m []objectInfo
c.modeMtx.RLock()
if c.readOnly() {
c.modeMtx.RUnlock()
time.Sleep(time.Second)
continue
}
// We put objects in batches of fixed size to not interfere with main put cycle a lot.
_ = c.db.View(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
cs := b.Cursor()
var k, v []byte
if len(lastKey) == 0 {
k, v = cs.First()
} else {
k, v = cs.Seek(lastKey)
if bytes.Equal(k, lastKey) {
k, v = cs.Next()
}
}
for ; k != nil && len(m) < flushBatchSize; k, v = cs.Next() {
if len(lastKey) == len(k) {
copy(lastKey, k)
} else {
lastKey = bytes.Clone(k)
}
m = append(m, objectInfo{
addr: string(k),
data: bytes.Clone(v),
})
}
return nil
})
var count int
for i := range m {
obj := objectSDK.New()
if err := obj.Unmarshal(m[i].data); err != nil {
continue
}
m[i].obj = obj
count++
select {
case c.flushCh <- m[i]:
case <-ctx.Done():
c.modeMtx.RUnlock()
return
}
}
c.modeMtx.RUnlock()
if count == 0 {
break
}
c.log.Debug(logs.WritecacheTriedToFlushItemsFromWritecache,
zap.Int("count", count),
zap.String("start", base58.Encode(lastKey)))
}
} }
func (c *cache) workerFlushBig(ctx context.Context) { func (c *cache) workerFlushBig(ctx context.Context) {
@ -197,9 +94,6 @@ func (c *cache) flushFSTree(ctx context.Context, ignoreErrors bool) error {
err = c.flushObject(ctx, &obj, e.ObjectData, StorageTypeFSTree) err = c.flushObject(ctx, &obj, e.ObjectData, StorageTypeFSTree)
if err != nil { if err != nil {
if ignoreErrors {
return nil
}
return err return err
} }
@ -211,29 +105,6 @@ func (c *cache) flushFSTree(ctx context.Context, ignoreErrors bool) error {
return err return err
} }
// workerFlushSmall writes small objects to the main storage.
func (c *cache) workerFlushSmall(ctx context.Context) {
defer c.wg.Done()
var objInfo objectInfo
for {
// Give priority to direct put.
select {
case objInfo = <-c.flushCh:
case <-ctx.Done():
return
}
err := c.flushObject(ctx, objInfo.obj, objInfo.data, StorageTypeDB)
if err != nil {
// Error is handled in flushObject.
continue
}
c.deleteFromDB(objInfo.addr, true)
}
}
// flushObject is used to write object directly to the main storage. // flushObject is used to write object directly to the main storage.
func (c *cache) flushObject(ctx context.Context, obj *objectSDK.Object, data []byte, st StorageType) error { func (c *cache) flushObject(ctx context.Context, obj *objectSDK.Object, data []byte, st StorageType) error {
var err error var err error
@ -300,13 +171,33 @@ func (c *cache) Flush(ctx context.Context, ignoreErrors, seal bool) error {
} }
func (c *cache) flush(ctx context.Context, ignoreErrors bool) error { func (c *cache) flush(ctx context.Context, ignoreErrors bool) error {
if err := c.flushFSTree(ctx, ignoreErrors); err != nil { return c.flushFSTree(ctx, ignoreErrors)
return err
} }
type batchItem struct {
data []byte
address string
}
func (c *cache) flushAndDropBBoltDB(ctx context.Context) error {
_, err := os.Stat(filepath.Join(c.path, dbName))
if err != nil && os.IsNotExist(err) {
return nil
}
if err != nil {
return fmt.Errorf("could not check write-cache database existence: %w", err)
}
db, err := OpenDB(c.path, true, os.OpenFile, c.pageSize)
if err != nil {
return fmt.Errorf("could not open write-cache database: %w", err)
}
defer func() {
_ = db.Close()
}()
var last string var last string
for { for {
batch, err := c.readNextDBBatch(ignoreErrors, last) batch, err := c.readNextDBBatch(db, last)
if err != nil { if err != nil {
return err return err
} }
@ -316,32 +207,27 @@ func (c *cache) flush(ctx context.Context, ignoreErrors bool) error {
for _, item := range batch { for _, item := range batch {
var obj objectSDK.Object var obj objectSDK.Object
if err := obj.Unmarshal(item.data); err != nil { if err := obj.Unmarshal(item.data); err != nil {
c.reportFlushError(logs.FSTreeCantDecodeDBObjectAddress, item.address, metaerr.Wrap(err)) return fmt.Errorf("unmarshal object from database: %w", err)
if ignoreErrors {
continue
} }
return err
}
if err := c.flushObject(ctx, &obj, item.data, StorageTypeDB); err != nil { if err := c.flushObject(ctx, &obj, item.data, StorageTypeDB); err != nil {
return err return fmt.Errorf("flush object from database: %w", err)
} }
c.deleteFromDB(item.address, false)
} }
last = batch[len(batch)-1].address last = batch[len(batch)-1].address
} }
if err := db.Close(); err != nil {
return fmt.Errorf("close write-cache database: %w", err)
}
if err := os.Remove(filepath.Join(c.path, dbName)); err != nil {
return fmt.Errorf("remove write-cache database: %w", err)
}
return nil return nil
} }
type batchItem struct { func (c *cache) readNextDBBatch(db *bbolt.DB, last string) ([]batchItem, error) {
data []byte
address string
}
func (c *cache) readNextDBBatch(ignoreErrors bool, last string) ([]batchItem, error) {
const batchSize = 100 const batchSize = 100
var batch []batchItem var batch []batchItem
err := c.db.View(func(tx *bbolt.Tx) error { err := db.View(func(tx *bbolt.Tx) error {
var addr oid.Address var addr oid.Address
b := tx.Bucket(defaultBucket) b := tx.Bucket(defaultBucket)
@ -352,11 +238,7 @@ func (c *cache) readNextDBBatch(ignoreErrors bool, last string) ([]batchItem, er
continue continue
} }
if err := addr.DecodeString(sa); err != nil { if err := addr.DecodeString(sa); err != nil {
c.reportFlushError(logs.FSTreeCantDecodeDBObjectAddress, sa, metaerr.Wrap(err)) return fmt.Errorf("decode address from database: %w", err)
if ignoreErrors {
continue
}
return err
} }
batch = append(batch, batchItem{data: bytes.Clone(data), address: sa}) batch = append(batch, batchItem{data: bytes.Clone(data), address: sa})

View file

@ -19,7 +19,6 @@ import (
oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id" oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id"
oidtest "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id/test" oidtest "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id/test"
"github.com/stretchr/testify/require" "github.com/stretchr/testify/require"
"go.etcd.io/bbolt"
"go.uber.org/zap" "go.uber.org/zap"
) )
@ -47,31 +46,6 @@ func TestFlush(t *testing.T) {
} }
failures := []TestFailureInjector[Option]{ failures := []TestFailureInjector[Option]{
{
Desc: "db, invalid address",
InjectFn: func(t *testing.T, wc Cache) {
c := wc.(*cache)
obj := testutil.GenerateObject()
data, err := obj.Marshal()
require.NoError(t, err)
require.NoError(t, c.db.Batch(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
return b.Put([]byte{1, 2, 3}, data)
}))
},
},
{
Desc: "db, invalid object",
InjectFn: func(t *testing.T, wc Cache) {
c := wc.(*cache)
require.NoError(t, c.db.Batch(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
k := []byte(oidtest.Address().EncodeToString())
v := []byte{1, 2, 3}
return b.Put(k, v)
}))
},
},
{ {
Desc: "fs, read error", Desc: "fs, read error",
InjectFn: func(t *testing.T, wc Cache) { InjectFn: func(t *testing.T, wc Cache) {
@ -263,7 +237,7 @@ func check(t *testing.T, mb *meta.DB, bs *blobstor.BlobStor, objects []objectPai
prm.StorageID = mRes.StorageID() prm.StorageID = mRes.StorageID()
res, err := bs.Get(context.Background(), prm) res, err := bs.Get(context.Background(), prm)
require.NoError(t, err) require.NoError(t, err, objects[i].addr)
require.Equal(t, objects[i].obj, res.Object) require.Equal(t, objects[i].obj, res.Object)
} }
} }

View file

@ -37,11 +37,11 @@ func (c *cache) Get(ctx context.Context, addr oid.Address) (*objectSDK.Object, e
return nil, ErrDegraded return nil, ErrDegraded
} }
obj, err := c.getInternal(ctx, saddr, addr) obj, err := c.getInternal(ctx, addr)
return obj, metaerr.Wrap(err) return obj, metaerr.Wrap(err)
} }
func (c *cache) getInternal(ctx context.Context, saddr string, addr oid.Address) (*objectSDK.Object, error) { func (c *cache) getInternal(ctx context.Context, addr oid.Address) (*objectSDK.Object, error) {
found := false found := false
storageType := StorageTypeUndefined storageType := StorageTypeUndefined
startedAt := time.Now() startedAt := time.Now()
@ -49,14 +49,6 @@ func (c *cache) getInternal(ctx context.Context, saddr string, addr oid.Address)
c.metrics.Get(time.Since(startedAt), found, storageType) c.metrics.Get(time.Since(startedAt), found, storageType)
}() }()
value, err := Get(c.db, []byte(saddr))
if err == nil {
obj := objectSDK.New()
found = true
storageType = StorageTypeDB
return obj, obj.Unmarshal(value)
}
res, err := c.fsTree.Get(ctx, common.GetPrm{Address: addr}) res, err := c.fsTree.Get(ctx, common.GetPrm{Address: addr})
if err != nil { if err != nil {
return nil, logicerr.Wrap(new(apistatus.ObjectNotFound)) return nil, logicerr.Wrap(new(apistatus.ObjectNotFound))
@ -87,7 +79,7 @@ func (c *cache) Head(ctx context.Context, addr oid.Address) (*objectSDK.Object,
return nil, ErrDegraded return nil, ErrDegraded
} }
obj, err := c.getInternal(ctx, saddr, addr) obj, err := c.getInternal(ctx, addr)
if err != nil { if err != nil {
return nil, metaerr.Wrap(err) return nil, metaerr.Wrap(err)
} }

View file

@ -5,13 +5,12 @@ import (
"errors" "errors"
"fmt" "fmt"
"os" "os"
"path/filepath"
"time" "time"
"git.frostfs.info/TrueCloudLab/frostfs-node/internal/logs" "git.frostfs.info/TrueCloudLab/frostfs-node/internal/logs"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/blobstor/common"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/shard/mode" "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/shard/mode"
"git.frostfs.info/TrueCloudLab/frostfs-observability/tracing" "git.frostfs.info/TrueCloudLab/frostfs-observability/tracing"
"go.etcd.io/bbolt"
"go.opentelemetry.io/otel/attribute" "go.opentelemetry.io/otel/attribute"
"go.opentelemetry.io/otel/trace" "go.opentelemetry.io/otel/trace"
) )
@ -53,7 +52,7 @@ func (c *cache) setMode(ctx context.Context, m mode.Mode, prm setModePrm) error
} }
} }
if err := c.closeDB(prm.shrink); err != nil { if err := c.closeStorage(ctx, prm.shrink); err != nil {
return err return err
} }
@ -78,33 +77,37 @@ func (c *cache) setMode(ctx context.Context, m mode.Mode, prm setModePrm) error
return nil return nil
} }
func (c *cache) closeDB(shrink bool) error { func (c *cache) closeStorage(ctx context.Context, shrink bool) error {
if c.db == nil { if c.fsTree == nil {
return nil return nil
} }
if !shrink { if !shrink {
if err := c.db.Close(); err != nil { if err := c.fsTree.Close(); err != nil {
return fmt.Errorf("can't close write-cache database: %w", err) return fmt.Errorf("can't close write-cache storage: %w", err)
} }
return nil return nil
} }
var empty bool empty := true
err := c.db.View(func(tx *bbolt.Tx) error { _, err := c.fsTree.Iterate(ctx, common.IteratePrm{
b := tx.Bucket(defaultBucket) Handler: func(common.IterationElement) error {
empty = b == nil || b.Stats().KeyN == 0 return errIterationCompleted
return nil },
}) })
if err != nil && !errors.Is(err, bbolt.ErrDatabaseNotOpen) { if err != nil {
return fmt.Errorf("failed to check DB items: %w", err) if errors.Is(err, errIterationCompleted) {
empty = false
} else {
return fmt.Errorf("failed to check write-cache items: %w", err)
} }
if err := c.db.Close(); err != nil { }
return fmt.Errorf("can't close write-cache database: %w", err) if err := c.fsTree.Close(); err != nil {
return fmt.Errorf("can't close write-cache storage: %w", err)
} }
if empty { if empty {
err := os.Remove(filepath.Join(c.path, dbName)) err := os.RemoveAll(c.path)
if err != nil && !os.IsNotExist(err) { if err != nil && !os.IsNotExist(err) {
return fmt.Errorf("failed to remove DB file: %w", err) return fmt.Errorf("failed to remove write-cache files: %w", err)
} }
} else { } else {
c.log.Info(logs.WritecacheShrinkSkippedNotEmpty) c.log.Info(logs.WritecacheShrinkSkippedNotEmpty)

View file

@ -17,14 +17,14 @@ func TestMode(t *testing.T) {
WithPath(t.TempDir())) WithPath(t.TempDir()))
require.NoError(t, wc.Open(context.Background(), mode.DegradedReadOnly)) require.NoError(t, wc.Open(context.Background(), mode.DegradedReadOnly))
require.Nil(t, wc.(*cache).db) require.Nil(t, wc.(*cache).fsTree)
require.NoError(t, wc.Init()) require.NoError(t, wc.Init())
require.Nil(t, wc.(*cache).db) require.Nil(t, wc.(*cache).fsTree)
require.NoError(t, wc.Close()) require.NoError(t, wc.Close())
require.NoError(t, wc.Open(context.Background(), mode.Degraded)) require.NoError(t, wc.Open(context.Background(), mode.Degraded))
require.Nil(t, wc.(*cache).db) require.Nil(t, wc.(*cache).fsTree)
require.NoError(t, wc.Init()) require.NoError(t, wc.Init())
require.Nil(t, wc.(*cache).db) require.Nil(t, wc.(*cache).fsTree)
require.NoError(t, wc.Close()) require.NoError(t, wc.Close())
} }

View file

@ -1,8 +1,6 @@
package writecache package writecache
import ( import (
"io/fs"
"os"
"time" "time"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/util/logger" "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/util/logger"
@ -42,8 +40,6 @@ type options struct {
noSync bool noSync bool
// reportError is the function called when encountering disk errors in background workers. // reportError is the function called when encountering disk errors in background workers.
reportError func(string, error) reportError func(string, error)
// openFile is the function called internally by bbolt to open database files. Useful for hermetic testing.
openFile func(string, int, fs.FileMode) (*os.File, error)
// metrics is metrics implementation // metrics is metrics implementation
metrics Metrics metrics Metrics
// disableBackgroundFlush is for testing purposes only. // disableBackgroundFlush is for testing purposes only.
@ -155,13 +151,6 @@ func WithReportErrorFunc(f func(string, error)) Option {
} }
} }
// WithOpenFile sets the OpenFile function to use internally by bolt. Useful for hermetic testing.
func WithOpenFile(f func(string, int, fs.FileMode) (*os.File, error)) Option {
return func(o *options) {
o.openFile = f
}
}
// WithMetrics sets metrics implementation. // WithMetrics sets metrics implementation.
func WithMetrics(metrics Metrics) Option { func WithMetrics(metrics Metrics) Option {
return func(o *options) { return func(o *options) {

View file

@ -8,7 +8,6 @@ import (
storagelog "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/internal/log" storagelog "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/internal/log"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/internal/metaerr" "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/internal/metaerr"
"git.frostfs.info/TrueCloudLab/frostfs-observability/tracing" "git.frostfs.info/TrueCloudLab/frostfs-observability/tracing"
"go.etcd.io/bbolt"
"go.opentelemetry.io/otel/attribute" "go.opentelemetry.io/otel/attribute"
"go.opentelemetry.io/otel/trace" "go.opentelemetry.io/otel/trace"
) )
@ -50,62 +49,16 @@ func (c *cache) Put(ctx context.Context, prm common.PutPrm) (common.PutRes, erro
return common.PutRes{}, ErrBigObject return common.PutRes{}, ErrBigObject
} }
oi := objectInfo{
addr: prm.Address.EncodeToString(),
obj: prm.Object,
data: prm.RawData,
}
if sz <= c.smallObjectSize {
storageType = StorageTypeDB
err := c.putSmall(oi)
if err == nil {
added = true
}
return common.PutRes{}, err
}
storageType = StorageTypeFSTree storageType = StorageTypeFSTree
err := c.putBig(ctx, oi.addr, prm) err := c.putBig(ctx, prm)
if err == nil { if err == nil {
added = true added = true
} }
return common.PutRes{}, metaerr.Wrap(err) return common.PutRes{}, metaerr.Wrap(err)
} }
// putSmall persists small objects to the write-cache database and
// pushes the to the flush workers queue.
func (c *cache) putSmall(obj objectInfo) error {
if !c.hasEnoughSpaceDB() {
return ErrOutOfSpace
}
var newRecord bool
err := c.db.Batch(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
key := []byte(obj.addr)
newRecord = b.Get(key) == nil
if newRecord {
return b.Put(key, obj.data)
}
return nil
})
if err == nil {
storagelog.Write(c.log,
storagelog.AddressField(obj.addr),
storagelog.StorageTypeField(wcStorageType),
storagelog.OpField("db PUT"),
)
if newRecord {
c.objCounters.cDB.Add(1)
c.estimateCacheSize()
}
}
return err
}
// putBig writes object to FSTree and pushes it to the flush workers queue. // putBig writes object to FSTree and pushes it to the flush workers queue.
func (c *cache) putBig(ctx context.Context, addr string, prm common.PutPrm) error { func (c *cache) putBig(ctx context.Context, prm common.PutPrm) error {
if !c.hasEnoughSpaceFS() { if !c.hasEnoughSpaceFS() {
return ErrOutOfSpace return ErrOutOfSpace
} }
@ -116,7 +69,7 @@ func (c *cache) putBig(ctx context.Context, addr string, prm common.PutPrm) erro
} }
storagelog.Write(c.log, storagelog.Write(c.log,
storagelog.AddressField(addr), storagelog.AddressField(prm.Address.EncodeToString()),
storagelog.StorageTypeField(wcStorageType), storagelog.StorageTypeField(wcStorageType),
storagelog.OpField("fstree PUT"), storagelog.OpField("fstree PUT"),
) )

View file

@ -1,29 +1,18 @@
package writecache package writecache
import ( import (
"fmt"
"math" "math"
"sync/atomic" "sync/atomic"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/blobstor/fstree" "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/blobstor/fstree"
"go.etcd.io/bbolt"
) )
func (c *cache) estimateCacheSize() (uint64, uint64) { func (c *cache) estimateCacheSize() (uint64, uint64) {
dbCount := c.objCounters.DB()
fsCount := c.objCounters.FS() fsCount := c.objCounters.FS()
if fsCount > 0 {
fsCount-- // db file
}
dbSize := dbCount * c.smallObjectSize
fsSize := fsCount * c.maxObjectSize fsSize := fsCount * c.maxObjectSize
c.metrics.SetEstimateSize(dbSize, fsSize) c.metrics.SetEstimateSize(0, fsSize)
c.metrics.SetActualCounters(dbCount, fsCount) c.metrics.SetActualCounters(0, fsCount)
return dbCount + fsCount, dbSize + fsSize return fsCount, fsSize
}
func (c *cache) hasEnoughSpaceDB() bool {
return c.hasEnoughSpace(c.smallObjectSize)
} }
func (c *cache) hasEnoughSpaceFS() bool { func (c *cache) hasEnoughSpaceFS() bool {
@ -41,11 +30,7 @@ func (c *cache) hasEnoughSpace(objectSize uint64) bool {
var _ fstree.FileCounter = &counters{} var _ fstree.FileCounter = &counters{}
type counters struct { type counters struct {
cDB, cFS atomic.Uint64 cFS atomic.Uint64
}
func (x *counters) DB() uint64 {
return x.cDB.Load()
} }
func (x *counters) FS() uint64 { func (x *counters) FS() uint64 {
@ -68,18 +53,6 @@ func (x *counters) Dec() {
} }
func (c *cache) initCounters() error { func (c *cache) initCounters() error {
var inDB uint64
err := c.db.View(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
if b != nil {
inDB = uint64(b.Stats().KeyN)
}
return nil
})
if err != nil {
return fmt.Errorf("could not read write-cache DB counter: %w", err)
}
c.objCounters.cDB.Store(inDB)
c.estimateCacheSize() c.estimateCacheSize()
return nil return nil
} }

View file

@ -3,7 +3,6 @@ package writecache
import ( import (
"context" "context"
"fmt" "fmt"
"math"
"os" "os"
"git.frostfs.info/TrueCloudLab/frostfs-node/internal/logs" "git.frostfs.info/TrueCloudLab/frostfs-node/internal/logs"
@ -14,16 +13,9 @@ import (
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/util" "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/util"
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/client" "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/client"
oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id" oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id"
"go.etcd.io/bbolt"
"go.uber.org/zap" "go.uber.org/zap"
) )
// store represents persistent storage with in-memory LRU cache
// for flushed items on top of it.
type store struct {
db *bbolt.DB
}
const dbName = "small.bolt" const dbName = "small.bolt"
func (c *cache) openStore(mod mode.ComponentMode) error { func (c *cache) openStore(mod mode.ComponentMode) error {
@ -32,24 +24,6 @@ func (c *cache) openStore(mod mode.ComponentMode) error {
return err return err
} }
c.db, err = OpenDB(c.path, mod.ReadOnly(), c.openFile, c.pageSize)
if err != nil {
return fmt.Errorf("could not open database: %w", err)
}
c.db.MaxBatchSize = c.maxBatchSize
c.db.MaxBatchDelay = c.maxBatchDelay
if !mod.ReadOnly() {
err = c.db.Update(func(tx *bbolt.Tx) error {
_, err := tx.CreateBucketIfNotExists(defaultBucket)
return err
})
if err != nil {
return fmt.Errorf("could not create default bucket: %w", err)
}
}
c.fsTree = fstree.New( c.fsTree = fstree.New(
fstree.WithPath(c.path), fstree.WithPath(c.path),
fstree.WithPerm(os.ModePerm), fstree.WithPerm(os.ModePerm),
@ -68,41 +42,6 @@ func (c *cache) openStore(mod mode.ComponentMode) error {
return nil return nil
} }
func (c *cache) deleteFromDB(key string, batched bool) {
var recordDeleted bool
var err error
if batched {
err = c.db.Batch(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
key := []byte(key)
recordDeleted = b.Get(key) != nil
return b.Delete(key)
})
} else {
err = c.db.Update(func(tx *bbolt.Tx) error {
b := tx.Bucket(defaultBucket)
key := []byte(key)
recordDeleted = b.Get(key) != nil
return b.Delete(key)
})
}
if err == nil {
c.metrics.Evict(StorageTypeDB)
storagelog.Write(c.log,
storagelog.AddressField(key),
storagelog.StorageTypeField(wcStorageType),
storagelog.OpField("db DELETE"),
)
if recordDeleted {
c.objCounters.cDB.Add(math.MaxUint64)
c.estimateCacheSize()
}
} else {
c.log.Error(logs.WritecacheCantRemoveObjectsFromTheDatabase, zap.Error(err))
}
}
func (c *cache) deleteFromDisk(ctx context.Context, addr oid.Address) { func (c *cache) deleteFromDisk(ctx context.Context, addr oid.Address) {
_, err := c.fsTree.Delete(ctx, common.DeletePrm{Address: addr}) _, err := c.fsTree.Delete(ctx, common.DeletePrm{Address: addr})
if err != nil && !client.IsErrObjectNotFound(err) { if err != nil && !client.IsErrObjectNotFound(err) {