forked from TrueCloudLab/frostfs-node
Dmitrii Stepanov
8a57c78f5f
1. Add forgotten metrics for client requests 2. Include execIfNotBlocked into metrics Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
461 lines
12 KiB
Go
461 lines
12 KiB
Go
package engine
|
|
|
|
import (
|
|
"context"
|
|
"errors"
|
|
|
|
"git.frostfs.info/TrueCloudLab/frostfs-node/internal/logs"
|
|
meta "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/metabase"
|
|
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/shard"
|
|
tracingPkg "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/tracing"
|
|
"git.frostfs.info/TrueCloudLab/frostfs-observability/tracing"
|
|
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/client"
|
|
apistatus "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/client/status"
|
|
cid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/container/id"
|
|
objectSDK "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object"
|
|
oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id"
|
|
"go.opentelemetry.io/otel/attribute"
|
|
"go.opentelemetry.io/otel/trace"
|
|
"go.uber.org/zap"
|
|
)
|
|
|
|
// InhumePrm encapsulates parameters for inhume operation.
|
|
type InhumePrm struct {
|
|
tombstone *oid.Address
|
|
addrs []oid.Address
|
|
|
|
forceRemoval bool
|
|
}
|
|
|
|
// InhumeRes encapsulates results of inhume operation.
|
|
type InhumeRes struct{}
|
|
|
|
// WithTarget sets a list of objects that should be inhumed and tombstone address
|
|
// as the reason for inhume operation.
|
|
//
|
|
// tombstone should not be nil, addr should not be empty.
|
|
// Should not be called along with MarkAsGarbage.
|
|
func (p *InhumePrm) WithTarget(tombstone oid.Address, addrs ...oid.Address) {
|
|
p.addrs = addrs
|
|
p.tombstone = &tombstone
|
|
}
|
|
|
|
// MarkAsGarbage marks an object to be physically removed from local storage.
|
|
//
|
|
// Should not be called along with WithTarget.
|
|
func (p *InhumePrm) MarkAsGarbage(addrs ...oid.Address) {
|
|
p.addrs = addrs
|
|
p.tombstone = nil
|
|
}
|
|
|
|
// WithForceRemoval inhumes objects specified via MarkAsGarbage with GC mark
|
|
// without any object restrictions checks.
|
|
func (p *InhumePrm) WithForceRemoval() {
|
|
p.forceRemoval = true
|
|
p.tombstone = nil
|
|
}
|
|
|
|
var errInhumeFailure = errors.New("inhume operation failed")
|
|
|
|
// Inhume calls metabase. Inhume method to mark an object as removed. It won't be
|
|
// removed physically from the shard until `Delete` operation.
|
|
//
|
|
// Allows inhuming non-locked objects only. Returns apistatus.ObjectLocked
|
|
// if at least one object is locked.
|
|
//
|
|
// NOTE: Marks any object as removed (despite any prohibitions on operations
|
|
// with that object) if WithForceRemoval option has been provided.
|
|
//
|
|
// Returns an error if executions are blocked (see BlockExecution).
|
|
func (e *StorageEngine) Inhume(ctx context.Context, prm InhumePrm) (res InhumeRes, err error) {
|
|
ctx, span := tracing.StartSpanFromContext(ctx, "StorageEngine.Inhume")
|
|
defer span.End()
|
|
defer elapsed("Inhume", e.metrics.AddMethodDuration)()
|
|
|
|
err = e.execIfNotBlocked(func() error {
|
|
res, err = e.inhume(ctx, prm)
|
|
return err
|
|
})
|
|
|
|
return
|
|
}
|
|
|
|
func (e *StorageEngine) inhume(ctx context.Context, prm InhumePrm) (InhumeRes, error) {
|
|
var shPrm shard.InhumePrm
|
|
if prm.forceRemoval {
|
|
shPrm.ForceRemoval()
|
|
}
|
|
|
|
for i := range prm.addrs {
|
|
if !prm.forceRemoval {
|
|
locked, err := e.IsLocked(ctx, prm.addrs[i])
|
|
if err != nil {
|
|
e.log.Warn(logs.EngineRemovingAnObjectWithoutFullLockingCheck,
|
|
zap.Error(err),
|
|
zap.Stringer("addr", prm.addrs[i]),
|
|
zap.String("trace_id", tracingPkg.GetTraceID(ctx)))
|
|
} else if locked {
|
|
return InhumeRes{}, new(apistatus.ObjectLocked)
|
|
}
|
|
}
|
|
|
|
if prm.tombstone != nil {
|
|
shPrm.SetTarget(*prm.tombstone, prm.addrs[i])
|
|
} else {
|
|
shPrm.MarkAsGarbage(prm.addrs[i])
|
|
}
|
|
|
|
ok, err := e.inhumeAddr(ctx, prm.addrs[i], shPrm, true)
|
|
if err != nil {
|
|
return InhumeRes{}, err
|
|
}
|
|
if !ok {
|
|
ok, err := e.inhumeAddr(ctx, prm.addrs[i], shPrm, false)
|
|
if err != nil {
|
|
return InhumeRes{}, err
|
|
} else if !ok {
|
|
return InhumeRes{}, errInhumeFailure
|
|
}
|
|
}
|
|
}
|
|
|
|
return InhumeRes{}, nil
|
|
}
|
|
|
|
// Returns ok if object was inhumed during this invocation or before.
|
|
func (e *StorageEngine) inhumeAddr(ctx context.Context, addr oid.Address, prm shard.InhumePrm, checkExists bool) (bool, error) {
|
|
root := false
|
|
var existPrm shard.ExistsPrm
|
|
var retErr error
|
|
var ok bool
|
|
|
|
e.iterateOverSortedShards(addr, func(_ int, sh hashedShard) (stop bool) {
|
|
defer func() {
|
|
// if object is root we continue since information about it
|
|
// can be presented in other shards
|
|
if checkExists && root {
|
|
stop = false
|
|
}
|
|
}()
|
|
|
|
if checkExists {
|
|
existPrm.Address = addr
|
|
exRes, err := sh.Exists(ctx, existPrm)
|
|
if err != nil {
|
|
if client.IsErrObjectAlreadyRemoved(err) || shard.IsErrObjectExpired(err) {
|
|
// inhumed once - no need to be inhumed again
|
|
ok = true
|
|
return true
|
|
}
|
|
|
|
var siErr *objectSDK.SplitInfoError
|
|
var ecErr *objectSDK.ECInfoError
|
|
if !(errors.As(err, &siErr) || errors.As(err, &ecErr)) {
|
|
e.reportShardError(sh, "could not check for presents in shard", err, zap.Stringer("address", addr))
|
|
return
|
|
}
|
|
|
|
root = true
|
|
} else if !exRes.Exists() {
|
|
return
|
|
}
|
|
}
|
|
|
|
_, err := sh.Inhume(ctx, prm)
|
|
if err != nil {
|
|
var errLocked *apistatus.ObjectLocked
|
|
switch {
|
|
case errors.As(err, &errLocked):
|
|
retErr = new(apistatus.ObjectLocked)
|
|
return true
|
|
case errors.Is(err, shard.ErrLockObjectRemoval):
|
|
retErr = meta.ErrLockObjectRemoval
|
|
return true
|
|
case errors.Is(err, shard.ErrReadOnlyMode) || errors.Is(err, shard.ErrDegradedMode):
|
|
retErr = err
|
|
return true
|
|
}
|
|
|
|
e.reportShardError(sh, "could not inhume object in shard", err, zap.Stringer("address", addr))
|
|
return false
|
|
}
|
|
|
|
ok = true
|
|
return true
|
|
})
|
|
|
|
return ok, retErr
|
|
}
|
|
|
|
// IsLocked checks whether an object is locked according to StorageEngine's state.
|
|
func (e *StorageEngine) IsLocked(ctx context.Context, addr oid.Address) (bool, error) {
|
|
ctx, span := tracing.StartSpanFromContext(ctx, "StorageEngine.IsLocked",
|
|
trace.WithAttributes(
|
|
attribute.String("address", addr.EncodeToString()),
|
|
))
|
|
defer span.End()
|
|
|
|
var locked bool
|
|
var err error
|
|
var outErr error
|
|
|
|
e.iterateOverUnsortedShards(func(h hashedShard) (stop bool) {
|
|
locked, err = h.Shard.IsLocked(ctx, addr)
|
|
if err != nil {
|
|
e.reportShardError(h, "can't check object's lockers", err, zap.Stringer("address", addr),
|
|
zap.String("trace_id", tracingPkg.GetTraceID(ctx)))
|
|
outErr = err
|
|
return false
|
|
}
|
|
|
|
return locked
|
|
})
|
|
|
|
if locked {
|
|
return locked, nil
|
|
}
|
|
|
|
return locked, outErr
|
|
}
|
|
|
|
// GetLocked return lock id's if object is locked according to StorageEngine's state.
|
|
func (e *StorageEngine) GetLocked(ctx context.Context, addr oid.Address) ([]oid.ID, error) {
|
|
ctx, span := tracing.StartSpanFromContext(ctx, "StorageEngine.GetLocked",
|
|
trace.WithAttributes(
|
|
attribute.String("address", addr.EncodeToString()),
|
|
))
|
|
defer span.End()
|
|
|
|
var locked []oid.ID
|
|
var outErr error
|
|
|
|
e.iterateOverUnsortedShards(func(h hashedShard) (stop bool) {
|
|
ld, err := h.Shard.GetLocked(ctx, addr)
|
|
if err != nil {
|
|
e.reportShardError(h, logs.EngineInterruptGettingLockers, err, zap.Stringer("address", addr),
|
|
zap.String("trace_id", tracingPkg.GetTraceID(ctx)))
|
|
outErr = err
|
|
}
|
|
locked = append(locked, ld...)
|
|
return false
|
|
})
|
|
if len(locked) > 0 {
|
|
return locked, nil
|
|
}
|
|
return locked, outErr
|
|
}
|
|
|
|
func (e *StorageEngine) processExpiredTombstones(ctx context.Context, addrs []meta.TombstonedObject) {
|
|
e.iterateOverUnsortedShards(func(sh hashedShard) (stop bool) {
|
|
sh.HandleExpiredTombstones(ctx, addrs)
|
|
|
|
select {
|
|
case <-ctx.Done():
|
|
return true
|
|
default:
|
|
return false
|
|
}
|
|
})
|
|
}
|
|
|
|
func (e *StorageEngine) processExpiredLocks(ctx context.Context, epoch uint64, lockers []oid.Address) {
|
|
e.iterateOverUnsortedShards(func(sh hashedShard) (stop bool) {
|
|
sh.HandleExpiredLocks(ctx, epoch, lockers)
|
|
|
|
select {
|
|
case <-ctx.Done():
|
|
e.log.Info(logs.EngineInterruptProcessingTheExpiredLocks, zap.Error(ctx.Err()))
|
|
return true
|
|
default:
|
|
return false
|
|
}
|
|
})
|
|
}
|
|
|
|
func (e *StorageEngine) processDeletedLocks(ctx context.Context, lockers []oid.Address) {
|
|
e.iterateOverUnsortedShards(func(sh hashedShard) (stop bool) {
|
|
sh.HandleDeletedLocks(lockers)
|
|
|
|
select {
|
|
case <-ctx.Done():
|
|
e.log.Info(logs.EngineInterruptProcessingTheDeletedLocks, zap.Error(ctx.Err()))
|
|
return true
|
|
default:
|
|
return false
|
|
}
|
|
})
|
|
}
|
|
|
|
func (e *StorageEngine) processZeroSizeContainers(ctx context.Context, ids []cid.ID) {
|
|
if len(ids) == 0 {
|
|
return
|
|
}
|
|
|
|
idMap, err := e.selectNonExistentIDs(ctx, ids)
|
|
if err != nil {
|
|
return
|
|
}
|
|
|
|
if len(idMap) == 0 {
|
|
return
|
|
}
|
|
|
|
var failed bool
|
|
var prm shard.ContainerSizePrm
|
|
e.iterateOverUnsortedShards(func(sh hashedShard) bool {
|
|
select {
|
|
case <-ctx.Done():
|
|
e.log.Info(logs.EngineInterruptProcessingZeroSizeContainers, zap.Error(ctx.Err()))
|
|
failed = true
|
|
return true
|
|
default:
|
|
}
|
|
|
|
var drop []cid.ID
|
|
for id := range idMap {
|
|
prm.SetContainerID(id)
|
|
s, err := sh.ContainerSize(prm)
|
|
if err != nil {
|
|
e.log.Warn(logs.EngineFailedToGetContainerSize, zap.Stringer("container_id", id), zap.Error(err))
|
|
failed = true
|
|
return true
|
|
}
|
|
if s.Size() > 0 {
|
|
drop = append(drop, id)
|
|
}
|
|
}
|
|
for _, id := range drop {
|
|
delete(idMap, id)
|
|
}
|
|
|
|
return len(idMap) == 0
|
|
})
|
|
|
|
if failed || len(idMap) == 0 {
|
|
return
|
|
}
|
|
|
|
e.iterateOverUnsortedShards(func(sh hashedShard) bool {
|
|
select {
|
|
case <-ctx.Done():
|
|
e.log.Info(logs.EngineInterruptProcessingZeroSizeContainers, zap.Error(ctx.Err()))
|
|
failed = true
|
|
return true
|
|
default:
|
|
}
|
|
|
|
for id := range idMap {
|
|
if err := sh.DeleteContainerSize(ctx, id); err != nil {
|
|
e.log.Warn(logs.EngineFailedToDeleteContainerSize, zap.Stringer("container_id", id), zap.Error(err))
|
|
failed = true
|
|
return true
|
|
}
|
|
}
|
|
|
|
return false
|
|
})
|
|
|
|
if failed {
|
|
return
|
|
}
|
|
|
|
for id := range idMap {
|
|
e.metrics.DeleteContainerSize(id.EncodeToString())
|
|
}
|
|
}
|
|
|
|
func (e *StorageEngine) processZeroCountContainers(ctx context.Context, ids []cid.ID) {
|
|
if len(ids) == 0 {
|
|
return
|
|
}
|
|
|
|
idMap, err := e.selectNonExistentIDs(ctx, ids)
|
|
if err != nil {
|
|
return
|
|
}
|
|
|
|
if len(idMap) == 0 {
|
|
return
|
|
}
|
|
|
|
var failed bool
|
|
var prm shard.ContainerCountPrm
|
|
e.iterateOverUnsortedShards(func(sh hashedShard) bool {
|
|
select {
|
|
case <-ctx.Done():
|
|
e.log.Info(logs.EngineInterruptProcessingZeroCountContainers, zap.Error(ctx.Err()))
|
|
failed = true
|
|
return true
|
|
default:
|
|
}
|
|
|
|
var drop []cid.ID
|
|
for id := range idMap {
|
|
prm.ContainerID = id
|
|
s, err := sh.ContainerCount(ctx, prm)
|
|
if err != nil {
|
|
e.log.Warn(logs.EngineFailedToGetContainerCounters, zap.Stringer("container_id", id), zap.Error(err))
|
|
failed = true
|
|
return true
|
|
}
|
|
if s.User > 0 || s.Logic > 0 || s.Phy > 0 {
|
|
drop = append(drop, id)
|
|
}
|
|
}
|
|
for _, id := range drop {
|
|
delete(idMap, id)
|
|
}
|
|
|
|
return len(idMap) == 0
|
|
})
|
|
|
|
if failed || len(idMap) == 0 {
|
|
return
|
|
}
|
|
|
|
e.iterateOverUnsortedShards(func(sh hashedShard) bool {
|
|
select {
|
|
case <-ctx.Done():
|
|
e.log.Info(logs.EngineInterruptProcessingZeroCountContainers, zap.Error(ctx.Err()))
|
|
failed = true
|
|
return true
|
|
default:
|
|
}
|
|
|
|
for id := range idMap {
|
|
if err := sh.DeleteContainerCount(ctx, id); err != nil {
|
|
e.log.Warn(logs.EngineFailedToDeleteContainerSize, zap.Stringer("container_id", id), zap.Error(err))
|
|
failed = true
|
|
return true
|
|
}
|
|
}
|
|
|
|
return false
|
|
})
|
|
|
|
if failed {
|
|
return
|
|
}
|
|
|
|
for id := range idMap {
|
|
e.metrics.DeleteContainerCount(id.EncodeToString())
|
|
}
|
|
}
|
|
|
|
func (e *StorageEngine) selectNonExistentIDs(ctx context.Context, ids []cid.ID) (map[cid.ID]struct{}, error) {
|
|
cs := e.containerSource.Load()
|
|
|
|
idMap := make(map[cid.ID]struct{})
|
|
for _, id := range ids {
|
|
isAvailable, err := cs.IsContainerAvailable(ctx, id)
|
|
if err != nil {
|
|
e.log.Warn(logs.EngineFailedToCheckContainerAvailability, zap.Stringer("container_id", id), zap.Error(err))
|
|
return nil, err
|
|
}
|
|
if isAvailable {
|
|
continue
|
|
}
|
|
idMap[id] = struct{}{}
|
|
}
|
|
return idMap, nil
|
|
}
|