Support faster Get, retry flushes on error

This commit is contained in:
Aayush Rajasekaran 2022-01-11 20:29:26 -05:00 committed by Jennifer Wang
parent ce6e328d52
commit dafd1f77fa
2 changed files with 75 additions and 44 deletions

View File

@ -3,6 +3,7 @@ package blockstore
import (
"context"
"sync"
"time"
"golang.org/x/xerrors"
@ -15,24 +16,39 @@ import (
var autolog = log.Named("auto")
type AutobatchBlockstore struct {
bufferedBlks []block.Block
addedCids map[cid.Cid]struct{}
bufferedBlksLk sync.Mutex
flushCh chan struct{}
flushErr error
shutdownCh chan struct{}
flushCtx context.Context
backingBs Blockstore
// TODO: drop if memory consumption is too high
addedCids map[cid.Cid]struct{}
bufferedLk sync.Mutex
bufferedBlksOrdered []block.Block
bufferedBlksMap map[cid.Cid]block.Block
flushingLk sync.Mutex
flushingBlksMap map[cid.Cid]block.Block
flushCh chan struct{}
flushErr error
flushRetryDelay time.Duration
flushCtx context.Context
shutdownCh chan struct{}
backingBs Blockstore
bufferCapacity int
bufferSize int
}
func NewAutobatch(ctx context.Context, backingBs Blockstore, bufferCapacity int) *AutobatchBlockstore {
bs := &AutobatchBlockstore{
backingBs: backingBs,
bufferCapacity: bufferCapacity,
addedCids: make(map[cid.Cid]struct{}),
flushCtx: ctx,
addedCids: make(map[cid.Cid]struct{}),
backingBs: backingBs,
bufferCapacity: bufferCapacity,
bufferedBlksMap: make(map[cid.Cid]block.Block),
flushingBlksMap: make(map[cid.Cid]block.Block),
flushCtx: ctx,
flushCh: make(chan struct{}, 1),
// could be made configable
flushRetryDelay: time.Second * 5,
}
go bs.flushWorker()
@ -41,11 +57,12 @@ func NewAutobatch(ctx context.Context, backingBs Blockstore, bufferCapacity int)
}
func (bs *AutobatchBlockstore) Put(ctx context.Context, blk block.Block) error {
bs.bufferedBlksLk.Lock()
bs.bufferedLk.Lock()
_, ok := bs.addedCids[blk.Cid()]
if !ok {
bs.bufferedBlks = append(bs.bufferedBlks, blk)
bs.addedCids[blk.Cid()] = struct{}{}
bs.bufferedBlksOrdered = append(bs.bufferedBlksOrdered, blk)
bs.bufferedBlksMap[blk.Cid()] = blk
bs.bufferSize += len(blk.RawData())
if bs.bufferSize >= bs.bufferCapacity {
// signal that a flush is appropriate, may be ignored
@ -56,7 +73,7 @@ func (bs *AutobatchBlockstore) Put(ctx context.Context, blk block.Block) error {
}
}
}
bs.bufferedBlksLk.Unlock()
bs.bufferedLk.Unlock()
return nil
}
@ -65,9 +82,16 @@ func (bs *AutobatchBlockstore) flushWorker() {
select {
case <-bs.flushCh:
putErr := bs.doFlush(bs.flushCtx)
if putErr != nil {
autolog.Errorf("FLUSH ERRORED: %w", putErr)
bs.flushErr = xerrors.Errorf("%w, put error: %w", bs.flushErr, putErr)
for putErr != nil {
select {
case <-bs.shutdownCh:
bs.flushErr = putErr
return
default:
autolog.Errorf("FLUSH ERRORED: %w, retrying in %v", putErr, bs.flushRetryDelay)
time.Sleep(bs.flushRetryDelay)
putErr = bs.doFlush(bs.flushCtx)
}
}
case <-bs.shutdownCh:
return
@ -76,20 +100,24 @@ func (bs *AutobatchBlockstore) flushWorker() {
}
func (bs *AutobatchBlockstore) doFlush(ctx context.Context) error {
bs.bufferedBlksLk.Lock()
bs.bufferedLk.Lock()
bs.flushingLk.Lock()
// We do NOT clear addedCids here, because its purpose is to expedite Puts
toFlush := bs.bufferedBlks
bs.bufferedBlks = []block.Block{}
bs.bufferedBlksLk.Unlock()
return bs.backingBs.PutMany(ctx, toFlush)
}
func (bs *AutobatchBlockstore) Flush(ctx context.Context) error {
return bs.doFlush(ctx)
flushingBlksOrdered := bs.bufferedBlksOrdered
bs.flushingBlksMap = bs.bufferedBlksMap
bs.bufferedBlksOrdered = []block.Block{}
bs.bufferedBlksMap = make(map[cid.Cid]block.Block)
bs.bufferedLk.Unlock()
bs.flushingLk.Unlock()
return bs.backingBs.PutMany(ctx, flushingBlksOrdered)
}
func (bs *AutobatchBlockstore) Shutdown(ctx context.Context) error {
// request one last flush of the worker
bs.flushCh <- struct{}{}
// shutdown the flush worker
bs.shutdownCh <- struct{}{}
// if it ever errored, this method fails
if bs.flushErr != nil {
return xerrors.Errorf("flushWorker errored: %w", bs.flushErr)
}
@ -98,8 +126,8 @@ func (bs *AutobatchBlockstore) Shutdown(ctx context.Context) error {
return bs.doFlush(ctx)
}
// May be very slow if the cid queried wasn't in the backingBs at the time of creation of this AutobatchBlockstore
func (bs *AutobatchBlockstore) Get(ctx context.Context, c cid.Cid) (block.Block, error) {
// may seem backward to check the backingBs first, but that is the likeliest case
blk, err := bs.backingBs.Get(ctx, c)
if err == nil {
return blk, nil
@ -109,16 +137,24 @@ func (bs *AutobatchBlockstore) Get(ctx context.Context, c cid.Cid) (block.Block,
return blk, err
}
bs.Flush(ctx)
return bs.backingBs.Get(ctx, c)
bs.flushingLk.Lock()
v, ok := bs.flushingBlksMap[c]
bs.flushingLk.Unlock()
if ok {
return v, nil
}
bs.bufferedLk.Lock()
v, ok = bs.bufferedBlksMap[c]
bs.bufferedLk.Unlock()
if ok {
return v, nil
}
return nil, ErrNotFound
}
func (bs *AutobatchBlockstore) DeleteBlock(context.Context, cid.Cid) error {
// if we wanted to support this, we would have to:
// - flush
// - delete from the backingBs (if present)
// - remove from addedCids (if present)
// - if present in addedCids, also walk bufferedBlks and remove if present
return xerrors.New("deletion is unsupported")
}
@ -159,8 +195,8 @@ func (bs *AutobatchBlockstore) PutMany(ctx context.Context, blks []block.Block)
}
func (bs *AutobatchBlockstore) AllKeysChan(ctx context.Context) (<-chan cid.Cid, error) {
bs.Flush(ctx)
return bs.backingBs.AllKeysChan(ctx)
return nil, xerrors.New("unsupported")
}
func (bs *AutobatchBlockstore) HashOnRead(enabled bool) {
@ -168,6 +204,5 @@ func (bs *AutobatchBlockstore) HashOnRead(enabled bool) {
}
func (bs *AutobatchBlockstore) View(ctx context.Context, cid cid.Cid, callback func([]byte) error) error {
bs.Flush(ctx)
return bs.backingBs.View(ctx, cid, callback)
return xerrors.New("unsupported")
}

View File

@ -1297,11 +1297,7 @@ func upgradeActorsV7Common(
return cid.Undef, xerrors.Errorf("failed to persist new state root: %w", err)
}
// Persist the new tree. Blocks until the entire writeStore is in the state blockstore.
if err := writeStore.Flush(ctx); err != nil {
return cid.Undef, xerrors.Errorf("failed to flush writestore: %w", err)
}
// Persists the new tree and shuts down the flush worker
if err := writeStore.Shutdown(ctx); err != nil {
return cid.Undef, xerrors.Errorf("writeStore failed: %w", err)
}