storj/satellite/repair/repairer/segments.go

801 lines
30 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 repairer
import (
"context"
"errors"
"fmt"
"io"
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
"math"
"strings"
2019-03-19 13:14:59 +00:00
"time"
2019-01-29 20:42:27 +00:00
"github.com/zeebo/errs"
"go.uber.org/zap"
"golang.org/x/exp/maps"
2019-01-29 20:42:27 +00:00
"storj.io/common/pb"
"storj.io/common/storj"
"storj.io/common/storj/location"
"storj.io/common/sync2"
"storj.io/storj/satellite/audit"
"storj.io/storj/satellite/metabase"
2019-03-28 20:09:23 +00:00
"storj.io/storj/satellite/orders"
"storj.io/storj/satellite/overlay"
"storj.io/storj/satellite/repair"
"storj.io/storj/satellite/repair/checker"
"storj.io/storj/satellite/repair/queue"
"storj.io/uplink/private/eestream"
"storj.io/uplink/private/piecestore"
)
var (
metainfoGetError = errs.Class("metainfo db get")
metainfoPutError = errs.Class("metainfo db put")
invalidRepairError = errs.Class("invalid repair")
overlayQueryError = errs.Class("overlay query failure")
orderLimitFailureError = errs.Class("order limits failure")
repairReconstructError = errs.Class("repair reconstruction failure")
repairPutError = errs.Class("repair could not store repaired pieces")
// segmentVerificationError is the errs class when the repaired segment can not be verified during repair.
segmentVerificationError = errs.Class("segment verification failed")
// segmentDeletedError is the errs class when the repaired segment was deleted during the repair.
segmentDeletedError = errs.Class("segment deleted during repair")
// segmentModifiedError is the errs class used when a segment has been changed in any way.
segmentModifiedError = errs.Class("segment has been modified")
)
// irreparableError identifies situations where a segment could not be repaired due to reasons
// which are hopefully transient (e.g. too many pieces unavailable). The segment should be added
// to the irreparableDB.
type irreparableError struct {
piecesAvailable int32
piecesRequired int32
}
func (ie *irreparableError) Error() string {
return fmt.Sprintf("%d available pieces < %d required", ie.piecesAvailable, ie.piecesRequired)
}
// PieceFetchResult combines a piece pointer with the error we got when we tried
// to acquire that piece.
type PieceFetchResult struct {
Piece metabase.Piece
Err error
}
// FetchResultReport contains a categorization of a set of pieces based on the results of
// GET operations.
type FetchResultReport struct {
Successful []PieceFetchResult
Failed []PieceFetchResult
Offline []PieceFetchResult
Contained []PieceFetchResult
Unknown []PieceFetchResult
}
// SegmentRepairer for segments.
type SegmentRepairer struct {
log *zap.Logger
statsCollector *statsCollector
metabase *metabase.DB
orders *orders.Service
overlay *overlay.Service
ec *ECRepairer
timeout time.Duration
reporter audit.Reporter
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
reputationUpdateEnabled bool
doDeclumping bool
doPlacementCheck bool
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
// multiplierOptimalThreshold is the value that multiplied by the optimal
// threshold results in the maximum limit of number of nodes to upload
// repaired pieces
multiplierOptimalThreshold float64
// repairOverrides is the set of values configured by the checker to override the repair threshold for various RS schemes.
repairOverrides checker.RepairOverridesMap
excludedCountryCodes map[location.CountryCode]struct{}
nowFn func() time.Time
OnTestingCheckSegmentAlteredHook func()
OnTestingPiecesReportHook func(pieces FetchResultReport)
placementRules overlay.PlacementRules
}
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
// NewSegmentRepairer creates a new instance of SegmentRepairer.
//
// excessPercentageOptimalThreshold is the percentage to apply over the optimal
// threshould to determine the maximum limit of nodes to upload repaired pieces,
// when negative, 0 is applied.
func NewSegmentRepairer(
log *zap.Logger,
metabase *metabase.DB,
orders *orders.Service,
overlay *overlay.Service,
reporter audit.Reporter,
ecRepairer *ECRepairer,
placementRules overlay.PlacementRules,
repairOverrides checker.RepairOverrides,
config Config,
) *SegmentRepairer {
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
excessOptimalThreshold := config.MaxExcessRateOptimalThreshold
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
if excessOptimalThreshold < 0 {
excessOptimalThreshold = 0
}
excludedCountryCodes := make(map[location.CountryCode]struct{})
for _, countryCode := range config.RepairExcludedCountryCodes {
if cc := location.ToCountryCode(countryCode); cc != location.None {
excludedCountryCodes[cc] = struct{}{}
}
}
return &SegmentRepairer{
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
log: log,
statsCollector: newStatsCollector(),
metabase: metabase,
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
orders: orders,
overlay: overlay,
ec: ecRepairer,
timeout: config.Timeout,
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
multiplierOptimalThreshold: 1 + excessOptimalThreshold,
repairOverrides: repairOverrides.GetMap(),
excludedCountryCodes: excludedCountryCodes,
reporter: reporter,
reputationUpdateEnabled: config.ReputationUpdateEnabled,
doDeclumping: config.DoDeclumping,
doPlacementCheck: config.DoPlacementCheck,
placementRules: placementRules,
nowFn: time.Now,
}
}
// Repair retrieves an at-risk segment and repairs and stores lost pieces on new nodes
// note that shouldDelete is used even in the case where err is not null
// note that it will update audit status as failed for nodes that failed piece hash verification during repair downloading.
func (repairer *SegmentRepairer) Repair(ctx context.Context, queueSegment *queue.InjuredSegment) (shouldDelete bool, err error) {
defer mon.Task()(&ctx, queueSegment.StreamID.String(), queueSegment.Position.Encode())(&err)
log := repairer.log.With(zap.Stringer("Stream ID", queueSegment.StreamID), zap.Uint64("Position", queueSegment.Position.Encode()))
segment, err := repairer.metabase.GetSegmentByPosition(ctx, metabase.GetSegmentByPosition{
StreamID: queueSegment.StreamID,
Position: queueSegment.Position,
})
if err != nil {
if metabase.ErrSegmentNotFound.Has(err) {
mon.Meter("repair_unnecessary").Mark(1) //mon:locked
mon.Meter("segment_deleted_before_repair").Mark(1) //mon:locked
log.Info("segment was deleted")
return true, nil
}
return false, metainfoGetError.Wrap(err)
}
if segment.Inline() {
return true, invalidRepairError.New("cannot repair inline segment")
}
// ignore segment if expired
if segment.Expired(repairer.nowFn()) {
mon.Meter("repair_unnecessary").Mark(1)
mon.Meter("segment_expired_before_repair").Mark(1)
log.Info("segment has expired")
return true, nil
}
redundancy, err := eestream.NewRedundancyStrategyFromStorj(segment.Redundancy)
if err != nil {
return true, invalidRepairError.New("invalid redundancy strategy: %w", err)
}
stats := repairer.getStatsByRS(&pb.RedundancyScheme{
Type: pb.RedundancyScheme_SchemeType(segment.Redundancy.Algorithm),
ErasureShareSize: segment.Redundancy.ShareSize,
MinReq: int32(segment.Redundancy.RequiredShares),
RepairThreshold: int32(segment.Redundancy.RepairShares),
SuccessThreshold: int32(segment.Redundancy.OptimalShares),
Total: int32(segment.Redundancy.TotalShares),
})
mon.Meter("repair_attempts").Mark(1) //mon:locked
stats.repairAttempts.Mark(1)
mon.IntVal("repair_segment_size").Observe(int64(segment.EncryptedSize)) //mon:locked
stats.repairSegmentSize.Observe(int64(segment.EncryptedSize))
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
allNodeIDs := make([]storj.NodeID, len(segment.Pieces))
for i, p := range segment.Pieces {
allNodeIDs[i] = p.StorageNode
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
selectedNodes, err := repairer.overlay.GetNodes(ctx, allNodeIDs)
if err != nil {
return false, overlayQueryError.New("error identifying missing pieces: %w", err)
}
if len(selectedNodes) != len(segment.Pieces) {
log.Error("GetNodes returned an invalid result", zap.Any("pieces", segment.Pieces), zap.Any("selectedNodes", selectedNodes))
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
return false, overlayQueryError.New("GetNodes returned an invalid result")
}
pieces := segment.Pieces
piecesCheck := repair.ClassifySegmentPieces(pieces, selectedNodes, repairer.excludedCountryCodes, repairer.doPlacementCheck, repairer.doDeclumping, repairer.placementRules(segment.Placement), allNodeIDs)
// irreparable segment
if piecesCheck.Retrievable.Size() < int(segment.Redundancy.RequiredShares) {
mon.Counter("repairer_segments_below_min_req").Inc(1) //mon:locked
stats.repairerSegmentsBelowMinReq.Inc(1)
mon.Meter("repair_nodes_unavailable").Mark(1) //mon:locked
stats.repairerNodesUnavailable.Mark(1)
log.Warn("irreparable segment",
zap.Int("piecesAvailable", piecesCheck.Retrievable.Size()),
zap.Int16("piecesRequired", segment.Redundancy.RequiredShares),
)
return false, nil
}
// ensure we get values, even if only zero values, so that redash can have an alert based on this
mon.Counter("repairer_segments_below_min_req").Inc(0) //mon:locked
stats.repairerSegmentsBelowMinReq.Inc(0)
repairThreshold := int32(segment.Redundancy.RepairShares)
pbRedundancy := &pb.RedundancyScheme{
MinReq: int32(segment.Redundancy.RequiredShares),
RepairThreshold: int32(segment.Redundancy.RepairShares),
SuccessThreshold: int32(segment.Redundancy.OptimalShares),
Total: int32(segment.Redundancy.TotalShares),
}
overrideValue := repairer.repairOverrides.GetOverrideValuePB(pbRedundancy)
if overrideValue != 0 {
repairThreshold = overrideValue
}
if piecesCheck.Healthy.Size() > int(repairThreshold) {
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
// No repair is needed (note Healthy does not include pieces in ForcingRepair).
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
var dropPieces metabase.Pieces
if piecesCheck.ForcingRepair.Size() > 0 {
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
// No repair is needed, but remove forcing-repair pieces without a repair operation,
// as we will still be above the repair threshold.
for _, piece := range pieces {
if piecesCheck.ForcingRepair.Contains(int(piece.Number)) {
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
dropPieces = append(dropPieces, piece)
}
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
if len(dropPieces) > 0 {
newPieces, err := segment.Pieces.Update(nil, dropPieces)
if err != nil {
return false, metainfoPutError.Wrap(err)
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
err = repairer.metabase.UpdateSegmentPieces(ctx, metabase.UpdateSegmentPieces{
StreamID: segment.StreamID,
Position: segment.Position,
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
OldPieces: segment.Pieces,
NewRedundancy: segment.Redundancy,
NewPieces: newPieces,
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
NewRepairedAt: time.Now(),
})
if err != nil {
return false, metainfoPutError.Wrap(err)
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
mon.Meter("dropped_undesirable_pieces_without_repair").Mark(len(dropPieces))
}
}
mon.Meter("repair_unnecessary").Mark(1) //mon:locked
stats.repairUnnecessary.Mark(1)
log.Info("segment above repair threshold", zap.Int("numHealthy", piecesCheck.Healthy.Size()), zap.Int32("repairThreshold", repairThreshold),
zap.Int("numClumped", piecesCheck.Clumped.Size()), zap.Int("numExiting", piecesCheck.Exiting.Size()), zap.Int("numOffPieces", piecesCheck.OutOfPlacement.Size()),
zap.Int("numExcluded", piecesCheck.InExcludedCountry.Size()), zap.Int("droppedPieces", len(dropPieces)))
return true, nil
}
2019-05-28 15:54:31 +01:00
healthyRatioBeforeRepair := 0.0
if segment.Redundancy.TotalShares != 0 {
healthyRatioBeforeRepair = float64(piecesCheck.Healthy.Size()) / float64(segment.Redundancy.TotalShares)
2019-05-28 15:54:31 +01:00
}
mon.FloatVal("healthy_ratio_before_repair").Observe(healthyRatioBeforeRepair) //mon:locked
stats.healthyRatioBeforeRepair.Observe(healthyRatioBeforeRepair)
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
// Create the order limits for the GET_REPAIR action
retrievablePieces := make(metabase.Pieces, 0, piecesCheck.Retrievable.Size())
2019-05-20 14:22:03 +01:00
for _, piece := range pieces {
if piecesCheck.Retrievable.Contains(int(piece.Number)) {
retrievablePieces = append(retrievablePieces, piece)
}
}
getOrderLimits, getPrivateKey, cachedNodesInfo, err := repairer.orders.CreateGetRepairOrderLimits(ctx, segment, retrievablePieces)
2019-03-28 20:09:23 +00:00
if err != nil {
if orders.ErrDownloadFailedNotEnoughPieces.Has(err) {
mon.Counter("repairer_segments_below_min_req").Inc(1) //mon:locked
stats.repairerSegmentsBelowMinReq.Inc(1)
mon.Meter("repair_nodes_unavailable").Mark(1) //mon:locked
stats.repairerNodesUnavailable.Mark(1)
log.Warn("irreparable segment: too many nodes offline",
zap.Int("piecesAvailable", len(retrievablePieces)),
zap.Int16("piecesRequired", segment.Redundancy.RequiredShares),
zap.Error(err),
)
}
return false, orderLimitFailureError.New("could not create GET_REPAIR order limits: %w", err)
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
// Double check for retrievable pieces which were recognized as irretrievable during the
// call to CreateGetRepairOrderLimits. Add or remove them from the appropriate sets.
for _, piece := range retrievablePieces {
if getOrderLimits[piece.Number] == nil {
piecesCheck.Missing.Include(int(piece.Number))
piecesCheck.Unhealthy.Include(int(piece.Number))
piecesCheck.Healthy.Remove(int(piece.Number))
piecesCheck.Retrievable.Remove(int(piece.Number))
piecesCheck.UnhealthyRetrievable.Remove(int(piece.Number))
}
}
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
var requestCount int
{
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
totalNeeded := int(math.Ceil(float64(redundancy.OptimalThreshold()) * repairer.multiplierOptimalThreshold))
if totalNeeded > redundancy.TotalCount() {
totalNeeded = redundancy.TotalCount()
}
requestCount = totalNeeded - piecesCheck.Healthy.Size()
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
}
minSuccessfulNeeded := redundancy.OptimalThreshold() - piecesCheck.Healthy.Size()
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
// Request Overlay for n-h new storage nodes
2019-03-23 08:06:11 +00:00
request := overlay.FindStorageNodesRequest{
[V3-1927] Repairer uploads to max threshold instead of success… (#2423) * pkg/datarepair: Add test to check num upload pieces Add a new test for ensuring the number of pieces that the repair process upload when a segment is injured. * satellite/orders: Don't create "put order limits" over total Repair must not create "put order limits" more than the total count. * pkg/datarepair: Update upload repair pieces test Update the test which checks the number of pieces which are uploaded during a repair for using the same excess over the success threshold value than the implementation. * satellites/orders: Limit repair put order for not being total Limit the number of put orders to be used by repair for only uploading pieces to a % excess over the successful threshold. * pkg/datarepair: Change DataRepair test to pass again Make some changes in the DataRepair test to make pass again after the repair upload repaired pieces only until a % excess over success threshold. Also update the steps description of the DataRepair test after it has been changed, to match on what's now, besides to leave it more generic for avoiding having to update it on minimal future refactorings. * satellite: Make repair excess optimal threshold configurable Add a new configuration parameter to the satellite for being able to configure the percentage excess over the optimal threshold, used for determining how many pieces should be repaired/uploaded, rather than having the value hard coded. * repairer: Add configurable param to segments/repairer Add a new parameters to the segment/repairer to calculate the maximum number of excess nodes, based on the optimal threshold, that repaired pieces can be uploaded. This new parameter has been added for not returning more nodes than the number of upload orders for data repair satellite service calculate for repairing pieces. * pkg/storage/ec: Update log message in clien.Repair * satellite: Update configuration lock file
2019-07-11 23:44:47 +01:00
RequestedCount: requestCount,
ExcludedIDs: piecesCheck.ExcludeNodeIDs,
Placement: segment.Placement,
}
newNodes, err := repairer.overlay.FindStorageNodesForUpload(ctx, request)
if err != nil {
return false, overlayQueryError.Wrap(err)
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
// Create the order limits for the PUT_REPAIR action. We want to keep pieces in Healthy
// as well as pieces in InExcludedCountry (our policy is to let those nodes keep the
// pieces they have, as long as they are kept intact and retrievable).
maxToKeep := int(segment.Redundancy.TotalShares) - len(newNodes)
toKeep := map[uint16]struct{}{}
// TODO how to avoid this two loops
for _, piece := range pieces {
if piecesCheck.Healthy.Contains(int(piece.Number)) {
toKeep[piece.Number] = struct{}{}
}
}
for _, piece := range pieces {
if piecesCheck.InExcludedCountry.Contains(int(piece.Number)) {
if len(toKeep) >= maxToKeep {
break
}
toKeep[piece.Number] = struct{}{}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
}
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
putLimits, putPrivateKey, err := repairer.orders.CreatePutRepairOrderLimits(ctx, segment, getOrderLimits, toKeep, newNodes)
2019-03-28 20:09:23 +00:00
if err != nil {
return false, orderLimitFailureError.New("could not create PUT_REPAIR order limits: %w", err)
}
// Download the segment using just the retrievable pieces
segmentReader, piecesReport, err := repairer.ec.Get(ctx, getOrderLimits, cachedNodesInfo, getPrivateKey, redundancy, int64(segment.EncryptedSize))
// ensure we get values, even if only zero values, so that redash can have an alert based on this
mon.Meter("repair_too_many_nodes_failed").Mark(0) //mon:locked
mon.Meter("repair_suspected_network_problem").Mark(0) //mon:locked
stats.repairTooManyNodesFailed.Mark(0)
if repairer.OnTestingPiecesReportHook != nil {
repairer.OnTestingPiecesReportHook(piecesReport)
}
// Check if segment has been altered
checkSegmentError := repairer.checkIfSegmentAltered(ctx, segment)
if checkSegmentError != nil {
if segmentDeletedError.Has(checkSegmentError) {
// mon.Meter("segment_deleted_during_repair").Mark(1) //mon:locked
log.Info("segment deleted during Repair")
return true, nil
}
if segmentModifiedError.Has(checkSegmentError) {
// mon.Meter("segment_modified_during_repair").Mark(1) //mon:locked
log.Info("segment modified during Repair")
return true, nil
}
return false, segmentVerificationError.Wrap(checkSegmentError)
}
if len(piecesReport.Contained) > 0 {
log.Debug("unexpected contained pieces during repair", zap.Int("count", len(piecesReport.Contained)))
}
if err != nil {
// If the context was closed during the Get phase, it will appear here as though
// we just failed to download enough pieces to reconstruct the segment. Check for
// a closed context before doing any further error processing.
if ctxErr := ctx.Err(); ctxErr != nil {
return false, ctxErr
}
// If Get failed because of input validation, then it will keep failing. But if it
// gave us irreparableError, then we failed to download enough pieces and must try
// to wait for nodes to come back online.
var irreparableErr *irreparableError
if errors.As(err, &irreparableErr) {
// piecesReport.Offline:
// Nodes which were online recently, but which we couldn't contact for
// this operation.
//
// piecesReport.Failed:
// Nodes which we contacted successfully but which indicated they
// didn't have the piece we wanted.
//
// piecesReport.Contained:
// Nodes which we contacted successfully but timed out after we asked
// for the piece.
//
// piecesReport.Unknown:
// Something else went wrong, and we don't know what.
//
// In a network failure scenario, we expect more than half of the outcomes
// will be in Offline or Contained.
if len(piecesReport.Offline)+len(piecesReport.Contained) > len(piecesReport.Successful)+len(piecesReport.Failed)+len(piecesReport.Unknown) {
mon.Meter("repair_suspected_network_problem").Mark(1) //mon:locked
} else {
mon.Meter("repair_too_many_nodes_failed").Mark(1) //mon:locked
}
stats.repairTooManyNodesFailed.Mark(1)
failedNodeIDs := make([]string, 0, len(piecesReport.Failed))
offlineNodeIDs := make([]string, 0, len(piecesReport.Offline))
timedOutNodeIDs := make([]string, 0, len(piecesReport.Contained))
unknownErrs := make([]string, 0, len(piecesReport.Unknown))
for _, outcome := range piecesReport.Failed {
failedNodeIDs = append(failedNodeIDs, outcome.Piece.StorageNode.String())
}
for _, outcome := range piecesReport.Offline {
offlineNodeIDs = append(offlineNodeIDs, outcome.Piece.StorageNode.String())
}
for _, outcome := range piecesReport.Contained {
timedOutNodeIDs = append(timedOutNodeIDs, outcome.Piece.StorageNode.String())
}
for _, outcome := range piecesReport.Unknown {
// We are purposefully using the error's string here, as opposed
// to wrapping the error. It is not likely that we need the local-side
// traceback of where this error was initially wrapped, and this will
// keep the logs more readable.
unknownErrs = append(unknownErrs, fmt.Sprintf("node ID [%s] err: %v", outcome.Piece.StorageNode, outcome.Err))
}
log.Warn("irreparable segment: could not acquire enough shares",
zap.Int32("piecesAvailable", irreparableErr.piecesAvailable),
zap.Int32("piecesRequired", irreparableErr.piecesRequired),
zap.Int("numFailedNodes", len(failedNodeIDs)),
zap.Stringer("failedNodes", commaSeparatedArray(failedNodeIDs)),
zap.Int("numOfflineNodes", len(offlineNodeIDs)),
zap.Stringer("offlineNodes", commaSeparatedArray(offlineNodeIDs)),
zap.Int("numTimedOutNodes", len(timedOutNodeIDs)),
zap.Stringer("timedOutNodes", commaSeparatedArray(timedOutNodeIDs)),
zap.Stringer("unknownErrors", commaSeparatedArray(unknownErrs)),
)
// repair will be attempted again if the segment remains unhealthy.
return false, nil
}
// The segment's redundancy strategy is invalid, or else there was an internal error.
return true, repairReconstructError.New("segment could not be reconstructed: %w", err)
}
defer func() { err = errs.Combine(err, segmentReader.Close()) }()
// only report audit result when segment can be successfully downloaded
cachedNodesReputation := make(map[storj.NodeID]overlay.ReputationStatus, len(cachedNodesInfo))
for id, info := range cachedNodesInfo {
cachedNodesReputation[id] = info.Reputation
}
report := audit.Report{
Segment: &segment,
NodesReputation: cachedNodesReputation,
}
for _, outcome := range piecesReport.Successful {
report.Successes = append(report.Successes, outcome.Piece.StorageNode)
}
for _, outcome := range piecesReport.Failed {
report.Fails = append(report.Fails, metabase.Piece{
StorageNode: outcome.Piece.StorageNode,
Number: outcome.Piece.Number,
})
}
for _, outcome := range piecesReport.Offline {
report.Offlines = append(report.Offlines, outcome.Piece.StorageNode)
}
for _, outcome := range piecesReport.Unknown {
report.Unknown = append(report.Unknown, outcome.Piece.StorageNode)
}
if repairer.reputationUpdateEnabled {
repairer.reporter.RecordAudits(ctx, report)
}
// Upload the repaired pieces
successfulNodes, _, err := repairer.ec.Repair(ctx, putLimits, putPrivateKey, redundancy, segmentReader, repairer.timeout, minSuccessfulNeeded)
if err != nil {
return false, repairPutError.Wrap(err)
}
pieceSize := eestream.CalcPieceSize(int64(segment.EncryptedSize), redundancy)
var bytesRepaired int64
// Add the successfully uploaded pieces to repairedPieces
var repairedPieces metabase.Pieces
repairedMap := make(map[uint16]bool)
for i, node := range successfulNodes {
if node == nil {
continue
}
bytesRepaired += pieceSize
piece := metabase.Piece{
Number: uint16(i),
StorageNode: node.Id,
}
repairedPieces = append(repairedPieces, piece)
repairedMap[uint16(i)] = true
}
mon.Meter("repair_bytes_uploaded").Mark64(bytesRepaired) //mon:locked
healthyAfterRepair := piecesCheck.Healthy.Size() + len(repairedPieces)
switch {
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
case healthyAfterRepair >= int(segment.Redundancy.OptimalShares):
mon.Meter("repair_success").Mark(1) //mon:locked
stats.repairSuccess.Mark(1)
case healthyAfterRepair <= int(segment.Redundancy.RepairShares):
// Important: this indicates a failure to PUT enough pieces to the network to pass
// the repair threshold, and _not_ a failure to reconstruct the segment. But we
// put at least one piece, else ec.Repair() would have returned an error. So the
// repair "succeeded" in that the segment is now healthier than it was, but it is
// not as healthy as we want it to be.
mon.Meter("repair_failed").Mark(1) //mon:locked
stats.repairFailed.Mark(1)
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
default:
mon.Meter("repair_partial").Mark(1) //mon:locked
stats.repairPartial.Mark(1)
}
2019-05-28 15:54:31 +01:00
healthyRatioAfterRepair := 0.0
if segment.Redundancy.TotalShares != 0 {
healthyRatioAfterRepair = float64(healthyAfterRepair) / float64(segment.Redundancy.TotalShares)
2019-05-28 15:54:31 +01:00
}
mon.FloatVal("healthy_ratio_after_repair").Observe(healthyRatioAfterRepair) //mon:locked
stats.healthyRatioAfterRepair.Observe(healthyRatioAfterRepair)
toRemove := make(map[uint16]metabase.Piece, piecesCheck.Unhealthy.Size())
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
switch {
case healthyAfterRepair >= int(segment.Redundancy.OptimalShares):
// Repair was fully successful; remove all unhealthy pieces except those in
// (Retrievable AND InExcludedCountry). Those, we allow to remain on the nodes as
// long as the nodes are keeping the pieces intact and available.
for _, piece := range pieces {
if piecesCheck.Unhealthy.Contains(int(piece.Number)) {
retrievable := piecesCheck.Retrievable.Contains(int(piece.Number))
inExcludedCountry := piecesCheck.InExcludedCountry.Contains(int(piece.Number))
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
if retrievable && inExcludedCountry {
continue
}
toRemove[piece.Number] = piece
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
}
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
case healthyAfterRepair > int(segment.Redundancy.RepairShares):
// Repair was successful enough that we still want to drop all out-of-placement
// pieces. We want to do that wherever possible, except where doing so puts data in
// jeopardy.
for _, piece := range pieces {
if piecesCheck.OutOfPlacement.Contains(int(piece.Number)) {
toRemove[piece.Number] = piece
}
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
default:
// Repair improved the health of the piece, but it is still at or below the
// repair threshold (not counting unhealthy-but-retrievable pieces). To be safe,
// we will keep unhealthy-but-retrievable pieces in the segment for now.
}
// in any case, we want to remove pieces for which we have replacements now.
for _, piece := range pieces {
if repairedMap[piece.Number] {
toRemove[piece.Number] = piece
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
}
}
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
// add pieces that failed piece hash verification to the removal list
for _, outcome := range piecesReport.Failed {
toRemove[outcome.Piece.Number] = outcome.Piece
}
newPieces, err := segment.Pieces.Update(repairedPieces, maps.Values(toRemove))
if err != nil {
return false, repairPutError.Wrap(err)
}
err = repairer.metabase.UpdateSegmentPieces(ctx, metabase.UpdateSegmentPieces{
StreamID: segment.StreamID,
Position: segment.Position,
OldPieces: segment.Pieces,
NewRedundancy: segment.Redundancy,
NewPieces: newPieces,
NewRepairedAt: time.Now(),
})
if err != nil {
return false, metainfoPutError.Wrap(err)
}
repairedAt := time.Time{}
if segment.RepairedAt != nil {
repairedAt = *segment.RepairedAt
}
var segmentAge time.Duration
if segment.CreatedAt.Before(repairedAt) {
segmentAge = time.Since(repairedAt)
} else {
segmentAge = time.Since(segment.CreatedAt)
}
mon.IntVal("segment_time_until_repair").Observe(int64(segmentAge.Seconds())) //mon:locked
stats.segmentTimeUntilRepair.Observe(int64(segmentAge.Seconds()))
log.Info("repaired segment",
zap.Int("clumped pieces", piecesCheck.Clumped.Size()),
zap.Int("exiting-node pieces", piecesCheck.Exiting.Size()),
zap.Int("out of placement pieces", piecesCheck.OutOfPlacement.Size()),
zap.Int("in excluded countries", piecesCheck.InExcludedCountry.Size()),
zap.Int("missing pieces", piecesCheck.Missing.Size()),
zap.Int("removed pieces", len(toRemove)),
zap.Int("repaired pieces", len(repairedPieces)),
zap.Int("retrievable pieces", piecesCheck.Retrievable.Size()),
zap.Int("healthy before repair", piecesCheck.Healthy.Size()),
satellite/repair: unify repair logic The repair checker and repair worker both need to determine which pieces are healthy, which are retrievable, and which should be replaced, but they have been doing it in different ways in different code, which has been the cause of bugs. The same term could have very similar but subtly different meanings between the two, causing much confusion. With this change, the piece- and node-classification logic is consolidated into one place within the satellite/repair package, so that both subsystems can use it. This ought to make decision-making code more concise and more readable. The consolidated classification logic has been expanded to create more sets, so that the decision-making code does not need to do as much precalculation. It should now be clearer in comments and code that a piece can belong to multiple sets arbitrarily (except where the definition of the sets makes this logically impossible), and what the precise meaning of each set is. These sets include Missing, Suspended, Clumped, OutOfPlacement, InExcludedCountry, ForcingRepair, UnhealthyRetrievable, Unhealthy, Retrievable, and Healthy. Some other side effects of this change: * CreatePutRepairOrderLimits no longer needs to special-case excluded countries; it can just create as many order limits as requested (by way of len(newNodes)). * The repair checker will now queue a segment for repair when there are any pieces out of placement. The code calls this "forcing a repair". * The checker.ReliabilityCache is now accessed by way of a GetNodes() function similar to the one on the overlay. The classification methods like MissingPieces(), OutOfPlacementPieces(), and PiecesNodesLastNetsInOrder() are removed in favor of the classification logic in satellite/repair/classification.go. This means the reliability cache no longer needs access to the placement rules or excluded countries list. Change-Id: I105109fb94ee126952f07d747c6e11131164fadb
2023-09-11 05:07:39 +01:00
zap.Int("healthy after repair", healthyAfterRepair),
zap.Int("total before repair", len(piecesCheck.ExcludeNodeIDs)),
zap.Int("total after repair", len(newPieces)))
return true, nil
}
// checkIfSegmentAltered checks if oldSegment has been altered since it was selected for audit.
func (repairer *SegmentRepairer) checkIfSegmentAltered(ctx context.Context, oldSegment metabase.Segment) (err error) {
defer mon.Task()(&ctx)(&err)
if repairer.OnTestingCheckSegmentAlteredHook != nil {
repairer.OnTestingCheckSegmentAlteredHook()
}
newSegment, err := repairer.metabase.GetSegmentByPosition(ctx, metabase.GetSegmentByPosition{
StreamID: oldSegment.StreamID,
Position: oldSegment.Position,
})
if err != nil {
if metabase.ErrSegmentNotFound.Has(err) {
return segmentDeletedError.New("StreamID: %q Position: %d", oldSegment.StreamID.String(), oldSegment.Position.Encode())
}
return err
}
if !oldSegment.Pieces.Equal(newSegment.Pieces) {
return segmentModifiedError.New("StreamID: %q Position: %d", oldSegment.StreamID.String(), oldSegment.Position.Encode())
}
return nil
}
func (repairer *SegmentRepairer) getStatsByRS(redundancy *pb.RedundancyScheme) *stats {
rsString := getRSString(repairer.loadRedundancy(redundancy))
return repairer.statsCollector.getStatsByRS(rsString)
}
func (repairer *SegmentRepairer) loadRedundancy(redundancy *pb.RedundancyScheme) (int, int, int, int) {
repair := int(redundancy.RepairThreshold)
overrideValue := repairer.repairOverrides.GetOverrideValuePB(redundancy)
if overrideValue != 0 {
repair = int(overrideValue)
}
return int(redundancy.MinReq), repair, int(redundancy.SuccessThreshold), int(redundancy.Total)
}
// SetNow allows tests to have the server act as if the current time is whatever they want.
func (repairer *SegmentRepairer) SetNow(nowFn func() time.Time) {
repairer.nowFn = nowFn
}
// AdminFetchInfo groups together all the information about a piece that should be retrievable
// from storage nodes.
type AdminFetchInfo struct {
Reader io.ReadCloser
Hash *pb.PieceHash
GetLimit *pb.AddressedOrderLimit
OriginalLimit *pb.OrderLimit
FetchError error
}
// AdminFetchPieces retrieves raw pieces and the associated hashes and original order
// limits from the storage nodes on which they are stored, and returns them intact to
// the caller rather than decoding or decrypting or verifying anything. This is to be
// used for debugging purposes.
func (repairer *SegmentRepairer) AdminFetchPieces(ctx context.Context, seg *metabase.Segment, saveDir string) (pieceInfos []AdminFetchInfo, err error) {
if seg.Inline() {
return nil, errs.New("cannot download an inline segment")
}
if len(seg.Pieces) < int(seg.Redundancy.RequiredShares) {
return nil, errs.New("segment only has %d pieces; needs %d for reconstruction", seg.Pieces, seg.Redundancy.RequiredShares)
}
// we treat all pieces as "healthy" for our purposes here; we want to download as many
// of them as we reasonably can. Thus, we pass in seg.Pieces for 'healthy'
getOrderLimits, getPrivateKey, cachedNodesInfo, err := repairer.orders.CreateGetRepairOrderLimits(ctx, *seg, seg.Pieces)
if err != nil {
return nil, errs.New("could not create order limits: %w", err)
}
pieceSize := seg.PieceSize()
pieceInfos = make([]AdminFetchInfo, len(getOrderLimits))
limiter := sync2.NewLimiter(int(seg.Redundancy.RequiredShares))
for currentLimitIndex, limit := range getOrderLimits {
if limit == nil {
continue
}
pieceInfos[currentLimitIndex].GetLimit = limit
currentLimitIndex, limit := currentLimitIndex, limit
limiter.Go(ctx, func() {
info := cachedNodesInfo[limit.GetLimit().StorageNodeId]
address := limit.GetStorageNodeAddress().GetAddress()
var triedLastIPPort bool
if info.LastIPPort != "" && info.LastIPPort != address {
address = info.LastIPPort
triedLastIPPort = true
}
pieceReadCloser, hash, originalLimit, err := repairer.ec.downloadAndVerifyPiece(ctx, limit, address, getPrivateKey, saveDir, pieceSize)
// if piecestore dial with last ip:port failed try again with node address
if triedLastIPPort && piecestore.Error.Has(err) {
if pieceReadCloser != nil {
_ = pieceReadCloser.Close()
}
pieceReadCloser, hash, originalLimit, err = repairer.ec.downloadAndVerifyPiece(ctx, limit, limit.GetStorageNodeAddress().GetAddress(), getPrivateKey, saveDir, pieceSize)
}
pieceInfos[currentLimitIndex].Reader = pieceReadCloser
pieceInfos[currentLimitIndex].Hash = hash
pieceInfos[currentLimitIndex].OriginalLimit = originalLimit
pieceInfos[currentLimitIndex].FetchError = err
})
}
limiter.Wait()
return pieceInfos, nil
}
// commaSeparatedArray concatenates an array into a comma-separated string,
// lazily.
type commaSeparatedArray []string
func (c commaSeparatedArray) String() string {
return strings.Join(c, ", ")
}