storj/satellite/metainfo/piecedeletion/service.go
Egon Elbre edb8d656de satellite/metainfo: adjust piecedeletion timeouts
Currently slower storagenodes can slow down deletion queue.
To make piece deletion faster reduce the maximum time spent in
either dialing or piece deletion requests.

With this change:
* dial timeout is 3s
* request timeout is 15s
* fail threshold is set to 10min

Similarly, we'll mark storage node as failed when the timeout occurs.
The timeout usually indicates that the storagenode is overwhelmed.
Garbage collection will ensure that the pieces get deleted eventually.

Change-Id: Iec5de699f5917905f5807140e2c3252088c6399b
2021-10-28 13:37:01 +03:00

247 lines
7.0 KiB
Go

// Copyright (C) 2020 Storj Labs, Inc.
// See LICENSE for copying information.
package piecedeletion
import (
"context"
"time"
"github.com/zeebo/errs"
"go.uber.org/zap"
"golang.org/x/sync/semaphore"
"storj.io/common/pb"
"storj.io/common/rpc"
"storj.io/common/storj"
"storj.io/common/sync2"
)
// Config defines configuration options for Service.
type Config struct {
MaxConcurrency int `help:"maximum number of concurrent requests to storage nodes" default:"100"`
MaxConcurrentPieces int `help:"maximum number of concurrent pieces can be processed" default:"1000000" testDefault:"1000"`
MaxPiecesPerBatch int `help:"maximum number of pieces per batch" default:"5000" testDefault:"4000"`
MaxPiecesPerRequest int `help:"maximum number pieces per single request" default:"1000" testDefault:"2000"`
DialTimeout time.Duration `help:"timeout for dialing nodes (0 means satellite default)" default:"3s" testDefault:"2s"`
FailThreshold time.Duration `help:"threshold for retrying a failed node" releaseDefault:"10m" devDefault:"2s"`
RequestTimeout time.Duration `help:"timeout for a single delete request" releaseDefault:"15s" devDefault:"2s"`
}
const (
minTimeout = 5 * time.Millisecond
maxTimeout = 5 * time.Minute
)
// Verify verifies configuration sanity.
func (config *Config) Verify() errs.Group {
var errlist errs.Group
if config.MaxConcurrency <= 0 {
errlist.Add(Error.New("concurrency %d must be greater than 0", config.MaxConcurrency))
}
if config.MaxConcurrentPieces <= 0 {
errlist.Add(Error.New("max concurrent pieces %d must be greater than 0", config.MaxConcurrentPieces))
}
if config.MaxPiecesPerBatch < config.MaxPiecesPerRequest {
errlist.Add(Error.New("max pieces per batch %d should be larger than max pieces per request %d", config.MaxPiecesPerBatch, config.MaxPiecesPerRequest))
}
if config.MaxPiecesPerBatch <= 0 {
errlist.Add(Error.New("max pieces per batch %d must be greater than 0", config.MaxPiecesPerBatch))
}
if config.MaxPiecesPerRequest <= 0 {
errlist.Add(Error.New("max pieces per request %d must be greater than 0", config.MaxPiecesPerRequest))
}
if config.DialTimeout != 0 && (config.DialTimeout <= minTimeout || maxTimeout <= config.DialTimeout) {
errlist.Add(Error.New("dial timeout %v must be between %v and %v", config.DialTimeout, minTimeout, maxTimeout))
}
if config.RequestTimeout < minTimeout || maxTimeout < config.RequestTimeout {
errlist.Add(Error.New("request timeout %v should be between %v and %v", config.RequestTimeout, minTimeout, maxTimeout))
}
return errlist
}
// Nodes stores reliable nodes information.
type Nodes interface {
KnownReliable(ctx context.Context, nodeIDs storj.NodeIDList) ([]*pb.Node, error)
}
// Service handles combining piece deletion requests.
//
// architecture: Service
type Service struct {
log *zap.Logger
config Config
concurrentRequests *semaphore.Weighted
rpcDialer rpc.Dialer
nodesDB Nodes
running sync2.Fence
combiner *Combiner
dialer *Dialer
limited *LimitedHandler
}
// NewService creates a new service.
func NewService(log *zap.Logger, dialer rpc.Dialer, nodesDB Nodes, config Config) (*Service, error) {
var errlist errs.Group
if log == nil {
errlist.Add(Error.New("log is nil"))
}
if dialer == (rpc.Dialer{}) {
errlist.Add(Error.New("dialer is zero"))
}
if nodesDB == nil {
errlist.Add(Error.New("nodesDB is nil"))
}
if errs := config.Verify(); len(errs) > 0 {
errlist.Add(errs...)
}
if err := errlist.Err(); err != nil {
return nil, Error.Wrap(err)
}
dialerClone := dialer
if config.DialTimeout > 0 {
dialerClone.DialTimeout = config.DialTimeout
}
if dialerClone.Pool == nil {
dialerClone.Pool = rpc.NewDefaultConnectionPool()
}
return &Service{
log: log,
config: config,
concurrentRequests: semaphore.NewWeighted(int64(config.MaxConcurrentPieces)),
rpcDialer: dialerClone,
nodesDB: nodesDB,
}, nil
}
// newQueue creates the configured queue.
func (service *Service) newQueue() Queue {
return NewLimitedJobs(service.config.MaxPiecesPerBatch)
}
// Run initializes the service.
func (service *Service) Run(ctx context.Context) error {
defer service.running.Release()
config := service.config
service.dialer = NewDialer(service.log.Named("dialer"), service.rpcDialer, config.RequestTimeout, config.FailThreshold, config.MaxPiecesPerRequest)
service.limited = NewLimitedHandler(service.dialer, config.MaxConcurrency)
service.combiner = NewCombiner(ctx, service.limited, service.newQueue)
return nil
}
// Close shuts down the service.
func (service *Service) Close() error {
if service.combiner != nil {
service.combiner.Close()
}
return nil
}
// Delete deletes the pieces specified in the requests waiting until success threshold is reached.
func (service *Service) Delete(ctx context.Context, requests []Request, successThreshold float64) (err error) {
defer mon.Task()(&ctx, len(requests), requestsPieceCount(requests), successThreshold)(&err)
if len(requests) == 0 {
return nil
}
// wait for combiner and dialer to set themselves up.
if !service.running.Wait(ctx) {
return Error.Wrap(ctx.Err())
}
for i, req := range requests {
if !req.IsValid() {
return Error.New("request #%d is invalid", i)
}
}
// When number of pieces are more than the maximum limit, we let it overflow,
// so we don't have to split requests in to separate batches.
totalPieceCount := requestsPieceCount(requests)
if totalPieceCount > service.config.MaxConcurrentPieces {
totalPieceCount = service.config.MaxConcurrentPieces
}
if err := service.concurrentRequests.Acquire(ctx, int64(totalPieceCount)); err != nil {
return Error.Wrap(err)
}
defer service.concurrentRequests.Release(int64(totalPieceCount))
// Create a map for matching node information with the corresponding
// request.
nodesReqs := make(map[storj.NodeID]Request, len(requests))
nodeIDs := []storj.NodeID{}
for _, req := range requests {
if req.Node.Address == "" {
nodeIDs = append(nodeIDs, req.Node.ID)
}
nodesReqs[req.Node.ID] = req
}
if len(nodeIDs) > 0 {
nodes, err := service.nodesDB.KnownReliable(ctx, nodeIDs)
if err != nil {
// Pieces will be collected by garbage collector
return Error.Wrap(err)
}
for _, node := range nodes {
req := nodesReqs[node.Id]
nodesReqs[node.Id] = Request{
Node: storj.NodeURL{
ID: node.Id,
Address: node.Address.Address,
},
Pieces: req.Pieces,
}
}
}
threshold, err := sync2.NewSuccessThreshold(len(nodesReqs), successThreshold)
if err != nil {
return Error.Wrap(err)
}
for _, req := range nodesReqs {
service.combiner.Enqueue(req.Node, Job{
Pieces: req.Pieces,
Resolve: threshold,
})
}
threshold.Wait(ctx)
return nil
}
// Request defines a deletion requests for a node.
type Request struct {
Node storj.NodeURL
Pieces []storj.PieceID
}
// IsValid returns whether the request is valid.
func (req *Request) IsValid() bool {
return !req.Node.ID.IsZero() && len(req.Pieces) > 0
}
func requestsPieceCount(requests []Request) int {
total := 0
for _, r := range requests {
total += len(r.Pieces)
}
return total
}