storj/satellite/accounting/tally/tally.go

332 lines
11 KiB
Go
Raw Normal View History

2019-01-24 20:15:10 +00:00
// Copyright (C) 2019 Storj Labs, Inc.
// See LICENSE for copying information.
package tally
import (
"context"
"time"
"github.com/spacemonkeygo/monkit/v3"
"github.com/zeebo/errs"
"go.uber.org/zap"
"storj.io/common/sync2"
"storj.io/common/uuid"
"storj.io/storj/satellite/accounting"
"storj.io/storj/satellite/metabase"
)
// Error is a standard error class for this package.
var (
Error = errs.Class("tally")
mon = monkit.Package()
)
// Config contains configurable values for the tally service.
2019-01-23 19:58:44 +00:00
type Config struct {
Interval time.Duration `help:"how frequently the tally service should run" releaseDefault:"1h" devDefault:"30s" testDefault:"$TESTINTERVAL"`
SaveRollupBatchSize int `help:"how large of batches SaveRollup should process at a time" default:"1000"`
ReadRollupBatchSize int `help:"how large of batches GetBandwidthSince should process at a time" default:"10000"`
ListLimit int `help:"how many objects to query in a batch" default:"2500"`
AsOfSystemInterval time.Duration `help:"as of system interval" releaseDefault:"-5m" devDefault:"-1us" testDefault:"-1us"`
}
// Service is the tally service for data stored on each storage node.
2019-09-10 14:24:16 +01:00
//
// architecture: Chore
type Service struct {
log *zap.Logger
config Config
Loop *sync2.Cycle
metabase *metabase.DB
liveAccounting accounting.Cache
storagenodeAccountingDB accounting.StoragenodeAccounting
projectAccountingDB accounting.ProjectAccounting
nowFn func() time.Time
}
// New creates a new tally Service.
func New(log *zap.Logger, sdb accounting.StoragenodeAccounting, pdb accounting.ProjectAccounting, liveAccounting accounting.Cache, metabase *metabase.DB, config Config) *Service {
return &Service{
log: log,
config: config,
Loop: sync2.NewCycle(config.Interval),
metabase: metabase,
liveAccounting: liveAccounting,
storagenodeAccountingDB: sdb,
projectAccountingDB: pdb,
nowFn: time.Now,
}
}
// Run the tally service loop.
func (service *Service) Run(ctx context.Context) (err error) {
defer mon.Task()(&ctx)(&err)
return service.Loop.Run(ctx, func(ctx context.Context) error {
err := service.Tally(ctx)
if err != nil {
service.log.Error("tally failed", zap.Error(err))
}
return nil
})
}
// Close stops the service and releases any resources.
func (service *Service) Close() error {
service.Loop.Close()
return nil
}
// SetNow allows tests to have the Service act as if the current time is whatever
// they want. This avoids races and sleeping, making tests more reliable and efficient.
func (service *Service) SetNow(now func() time.Time) {
service.nowFn = now
}
// Tally calculates data-at-rest usage once.
//
// How live accounting is calculated:
//
// At the beginning of the tally iteration, we get a map containing the current
// project totals from the cache- initialLiveTotals (our current estimation of
// the project totals). At the end of the tally iteration, we have the totals
// from what we saw during the metainfo loop.
//
// However, data which was uploaded during the loop may or may not have been
// seen in the metainfo loop. For this reason, we also read the live accounting
// totals again at the end of the tally iteration- latestLiveTotals.
//
// The difference between latest and initial indicates how much data was
// uploaded during the metainfo loop and is assigned to delta. However, again,
// we aren't certain how much of the delta is accounted for in the metainfo
// totals. For the reason we make an assumption that 50% of the data is
// accounted for. So to calculate the new live accounting totals, we sum the
// metainfo totals and 50% of the deltas.
func (service *Service) Tally(ctx context.Context) (err error) {
defer mon.Task()(&ctx)(&err)
// No-op unless that there isn't an error getting the
// liveAccounting.GetAllProjectTotals
updateLiveAccountingTotals := func(_ map[uuid.UUID]accounting.Usage) {}
satellite/accounting: refactor live accounting to hold current estimated totals live accounting used to be a cache to store writes before they are picked up during the tally iteration, after which the cache is cleared. This created a window in which users could potentially exceed the storage limit. This PR refactors live accounting to hold current estimations of space used per project. This should also reduce DB load since we no longer need to query the satellite DB when checking space used for limiting. The mechanism by which the new live accounting system works is as follows: During the upload of any segment, the size of that segment is added to its respective project total in live accounting. At the beginning of the tally iteration we record the current values in live accounting as `initialLiveTotals`. At the end of the tally iteration we again record the current totals in live accounting as `latestLiveTotals`. The metainfo loop observer in tally allows us to get the project totals from what it observed in metainfo DB which are stored in `tallyProjectTotals`. However, for any particular segment uploaded during the metainfo loop, the observer may or may not have seen it. Thus, we take half of the difference between `latestLiveTotals` and `initialLiveTotals`, and add that to the total that was found during tally and set that as the new live accounting total. Initially, live accounting was storing the total stored amount across all nodes rather than the segment size, which is inconsistent with how we record amounts stored in the project accounting DB, so we have refactored live accounting to record segment size Change-Id: Ie48bfdef453428fcdc180b2d781a69d58fd927fb
2019-10-31 17:27:38 +00:00
initialLiveTotals, err := service.liveAccounting.GetAllProjectTotals(ctx)
if err != nil {
service.log.Error(
"tally won't update the live accounting storage usages of the projects in this cycle",
zap.Error(err),
)
} else {
updateLiveAccountingTotals = func(tallyProjectTotals map[uuid.UUID]accounting.Usage) {
latestLiveTotals, err := service.liveAccounting.GetAllProjectTotals(ctx)
if err != nil {
service.log.Error(
"tally isn't updating the live accounting storage usages of the projects in this cycle",
zap.Error(err),
)
return
}
// empty projects are not returned by the metainfo observer. If a project exists
// in live accounting, but not in tally projects, we would not update it in live accounting.
// Thus, we add them and set the total to 0.
for projectID := range latestLiveTotals {
if _, ok := tallyProjectTotals[projectID]; !ok {
tallyProjectTotals[projectID] = accounting.Usage{}
}
}
for projectID, tallyTotal := range tallyProjectTotals {
delta := latestLiveTotals[projectID].Storage - initialLiveTotals[projectID].Storage
if delta < 0 {
delta = 0
}
// read the method documentation why the increase passed to this method
// is calculated in this way
err = service.liveAccounting.AddProjectStorageUsage(ctx, projectID, -latestLiveTotals[projectID].Storage+tallyTotal.Storage+(delta/2))
if err != nil {
if accounting.ErrSystemOrNetError.Has(err) {
service.log.Error(
"tally isn't updating the live accounting storage usages of the projects in this cycle",
zap.Error(err),
)
return
}
service.log.Error(
"tally isn't updating the live accounting storage usage of the project in this cycle",
zap.String("projectID", projectID.String()),
zap.Error(err),
)
}
// difference between cached project totals and latest tally collector
increment := tallyTotal.Segments - latestLiveTotals[projectID].Segments
err = service.liveAccounting.UpdateProjectSegmentUsage(ctx, projectID, increment)
if err != nil {
if accounting.ErrSystemOrNetError.Has(err) {
service.log.Error(
"tally isn't updating the live accounting segment usages of the projects in this cycle",
zap.Error(err),
)
return
}
service.log.Error(
"tally isn't updating the live accounting segment usage of the project in this cycle",
zap.String("projectID", projectID.String()),
zap.Error(err),
)
}
}
}
}
// add up all buckets
collector := NewBucketTallyCollector(service.log.Named("observer"), service.nowFn(), service.metabase, service.config)
err = collector.Run(ctx)
if err != nil {
return Error.Wrap(err)
}
finishTime := service.nowFn()
// save the new results
var errAtRest error
if len(collector.Bucket) > 0 {
satellite/accounting: refactor live accounting to hold current estimated totals live accounting used to be a cache to store writes before they are picked up during the tally iteration, after which the cache is cleared. This created a window in which users could potentially exceed the storage limit. This PR refactors live accounting to hold current estimations of space used per project. This should also reduce DB load since we no longer need to query the satellite DB when checking space used for limiting. The mechanism by which the new live accounting system works is as follows: During the upload of any segment, the size of that segment is added to its respective project total in live accounting. At the beginning of the tally iteration we record the current values in live accounting as `initialLiveTotals`. At the end of the tally iteration we again record the current totals in live accounting as `latestLiveTotals`. The metainfo loop observer in tally allows us to get the project totals from what it observed in metainfo DB which are stored in `tallyProjectTotals`. However, for any particular segment uploaded during the metainfo loop, the observer may or may not have seen it. Thus, we take half of the difference between `latestLiveTotals` and `initialLiveTotals`, and add that to the total that was found during tally and set that as the new live accounting total. Initially, live accounting was storing the total stored amount across all nodes rather than the segment size, which is inconsistent with how we record amounts stored in the project accounting DB, so we have refactored live accounting to record segment size Change-Id: Ie48bfdef453428fcdc180b2d781a69d58fd927fb
2019-10-31 17:27:38 +00:00
// record bucket tallies to DB
err = service.projectAccountingDB.SaveTallies(ctx, finishTime, collector.Bucket)
if err != nil {
errAtRest = Error.New("ProjectAccounting.SaveTallies failed: %v", err)
}
satellite/accounting: refactor live accounting to hold current estimated totals live accounting used to be a cache to store writes before they are picked up during the tally iteration, after which the cache is cleared. This created a window in which users could potentially exceed the storage limit. This PR refactors live accounting to hold current estimations of space used per project. This should also reduce DB load since we no longer need to query the satellite DB when checking space used for limiting. The mechanism by which the new live accounting system works is as follows: During the upload of any segment, the size of that segment is added to its respective project total in live accounting. At the beginning of the tally iteration we record the current values in live accounting as `initialLiveTotals`. At the end of the tally iteration we again record the current totals in live accounting as `latestLiveTotals`. The metainfo loop observer in tally allows us to get the project totals from what it observed in metainfo DB which are stored in `tallyProjectTotals`. However, for any particular segment uploaded during the metainfo loop, the observer may or may not have seen it. Thus, we take half of the difference between `latestLiveTotals` and `initialLiveTotals`, and add that to the total that was found during tally and set that as the new live accounting total. Initially, live accounting was storing the total stored amount across all nodes rather than the segment size, which is inconsistent with how we record amounts stored in the project accounting DB, so we have refactored live accounting to record segment size Change-Id: Ie48bfdef453428fcdc180b2d781a69d58fd927fb
2019-10-31 17:27:38 +00:00
updateLiveAccountingTotals(projectTotalsFromBuckets(collector.Bucket))
}
2019-06-13 17:58:40 +01:00
if len(collector.Bucket) > 0 {
var total accounting.BucketTally
// TODO for now we don't have access to inline/remote stats per bucket
// but that may change in the future. To get back those stats we would
// most probably need to add inline/remote information to object in
// metabase. We didn't decide yet if that is really needed right now.
for _, bucket := range collector.Bucket {
monAccounting.IntVal("bucket_objects").Observe(bucket.ObjectCount) //mon:locked
monAccounting.IntVal("bucket_segments").Observe(bucket.Segments()) //mon:locked
// monAccounting.IntVal("bucket_inline_segments").Observe(bucket.InlineSegments) //mon:locked
// monAccounting.IntVal("bucket_remote_segments").Observe(bucket.RemoteSegments) //mon:locked
monAccounting.IntVal("bucket_bytes").Observe(bucket.Bytes()) //mon:locked
// monAccounting.IntVal("bucket_inline_bytes").Observe(bucket.InlineBytes) //mon:locked
// monAccounting.IntVal("bucket_remote_bytes").Observe(bucket.RemoteBytes) //mon:locked
total.Combine(bucket)
}
monAccounting.IntVal("total_objects").Observe(total.ObjectCount) //mon:locked
monAccounting.IntVal("total_segments").Observe(total.Segments()) //mon:locked
monAccounting.IntVal("total_bytes").Observe(total.Bytes()) //mon:locked
monAccounting.IntVal("total_pending_objects").Observe(total.PendingObjectCount)
}
// return errors if something went wrong.
return errAtRest
}
// BucketTallyCollector collects and adds up tallies for buckets.
type BucketTallyCollector struct {
Now time.Time
Log *zap.Logger
Bucket map[metabase.BucketLocation]*accounting.BucketTally
metabase *metabase.DB
config Config
}
// NewBucketTallyCollector returns an collector that adds up totals for buckets.
// The now argument controls when the collector considers objects to be expired.
func NewBucketTallyCollector(log *zap.Logger, now time.Time, db *metabase.DB, config Config) *BucketTallyCollector {
return &BucketTallyCollector{
Now: now,
Log: log,
Bucket: make(map[metabase.BucketLocation]*accounting.BucketTally),
metabase: db,
config: config,
}
}
// Run runs collecting bucket tallies.
func (observer *BucketTallyCollector) Run(ctx context.Context) (err error) {
defer mon.Task()(&ctx)(&err)
startingTime, err := observer.metabase.Now(ctx)
if err != nil {
return err
}
return observer.metabase.IterateLoopObjects(ctx, metabase.IterateLoopObjects{
BatchSize: observer.config.ListLimit,
AsOfSystemTime: startingTime,
AsOfSystemInterval: observer.config.AsOfSystemInterval,
}, func(ctx context.Context, it metabase.LoopObjectsIterator) (err error) {
var entry metabase.LoopObjectEntry
for it.Next(ctx, &entry) {
err = observer.object(ctx, entry)
if err != nil {
return err
}
}
return nil
})
}
// ensureBucket returns bucket corresponding to the passed in path.
func (observer *BucketTallyCollector) ensureBucket(location metabase.ObjectLocation) *accounting.BucketTally {
bucketLocation := location.Bucket()
bucket, exists := observer.Bucket[bucketLocation]
if !exists {
bucket = &accounting.BucketTally{}
bucket.BucketLocation = bucketLocation
observer.Bucket[bucketLocation] = bucket
}
return bucket
}
// Object is called for each object once.
func (observer *BucketTallyCollector) object(ctx context.Context, object metabase.LoopObjectEntry) (err error) {
defer mon.Task()(&ctx)(&err)
if object.Expired(observer.Now) {
return nil
}
bucket := observer.ensureBucket(object.ObjectStream.Location())
bucket.TotalSegments += int64(object.SegmentCount)
bucket.TotalBytes += object.TotalEncryptedSize
bucket.MetadataSize += int64(object.EncryptedMetadataSize)
bucket.ObjectCount++
if object.Status == metabase.Pending {
bucket.PendingObjectCount++
}
return nil
}
func projectTotalsFromBuckets(buckets map[metabase.BucketLocation]*accounting.BucketTally) map[uuid.UUID]accounting.Usage {
projectTallyTotals := make(map[uuid.UUID]accounting.Usage)
satellite/accounting: refactor live accounting to hold current estimated totals live accounting used to be a cache to store writes before they are picked up during the tally iteration, after which the cache is cleared. This created a window in which users could potentially exceed the storage limit. This PR refactors live accounting to hold current estimations of space used per project. This should also reduce DB load since we no longer need to query the satellite DB when checking space used for limiting. The mechanism by which the new live accounting system works is as follows: During the upload of any segment, the size of that segment is added to its respective project total in live accounting. At the beginning of the tally iteration we record the current values in live accounting as `initialLiveTotals`. At the end of the tally iteration we again record the current totals in live accounting as `latestLiveTotals`. The metainfo loop observer in tally allows us to get the project totals from what it observed in metainfo DB which are stored in `tallyProjectTotals`. However, for any particular segment uploaded during the metainfo loop, the observer may or may not have seen it. Thus, we take half of the difference between `latestLiveTotals` and `initialLiveTotals`, and add that to the total that was found during tally and set that as the new live accounting total. Initially, live accounting was storing the total stored amount across all nodes rather than the segment size, which is inconsistent with how we record amounts stored in the project accounting DB, so we have refactored live accounting to record segment size Change-Id: Ie48bfdef453428fcdc180b2d781a69d58fd927fb
2019-10-31 17:27:38 +00:00
for _, bucket := range buckets {
projectUsage := projectTallyTotals[bucket.ProjectID]
projectUsage.Storage += bucket.TotalBytes
projectUsage.Segments += bucket.TotalSegments
projectTallyTotals[bucket.ProjectID] = projectUsage
satellite/accounting: refactor live accounting to hold current estimated totals live accounting used to be a cache to store writes before they are picked up during the tally iteration, after which the cache is cleared. This created a window in which users could potentially exceed the storage limit. This PR refactors live accounting to hold current estimations of space used per project. This should also reduce DB load since we no longer need to query the satellite DB when checking space used for limiting. The mechanism by which the new live accounting system works is as follows: During the upload of any segment, the size of that segment is added to its respective project total in live accounting. At the beginning of the tally iteration we record the current values in live accounting as `initialLiveTotals`. At the end of the tally iteration we again record the current totals in live accounting as `latestLiveTotals`. The metainfo loop observer in tally allows us to get the project totals from what it observed in metainfo DB which are stored in `tallyProjectTotals`. However, for any particular segment uploaded during the metainfo loop, the observer may or may not have seen it. Thus, we take half of the difference between `latestLiveTotals` and `initialLiveTotals`, and add that to the total that was found during tally and set that as the new live accounting total. Initially, live accounting was storing the total stored amount across all nodes rather than the segment size, which is inconsistent with how we record amounts stored in the project accounting DB, so we have refactored live accounting to record segment size Change-Id: Ie48bfdef453428fcdc180b2d781a69d58fd927fb
2019-10-31 17:27:38 +00:00
}
return projectTallyTotals
}
// using custom name to avoid breaking monitoring.
var monAccounting = monkit.ScopeNamed("storj.io/storj/satellite/accounting")