frostfs-node/pkg/services/policer/ec.go
Aleksey Savchuk f0c43c8d80
[#1502] Use zap.Error for logging errors
Use `zap.Error` instead of `zap.String` for logging errors: change all expressions like
`zap.String("error", err.Error())` or `zap.String("err", err.Error())` to `zap.Error(err)`.
Leave similar expressions with other messages unchanged, for example,
`zap.String("last_error", lastErr.Error())` or `zap.String("reason", ctx.Err().Error())`.

This change was made by applying the following patch:
```diff
@@
var err expression
@@
-zap.String("error", err.Error())
+zap.Error(err)

@@
var err expression
@@
-zap.String("err", err.Error())
+zap.Error(err)
```

Signed-off-by: Aleksey Savchuk <a.savchuk@yadro.com>
2024-12-16 11:13:42 +03:00

395 lines
14 KiB
Go

package policer
import (
"context"
"encoding/hex"
"errors"
"fmt"
"git.frostfs.info/TrueCloudLab/frostfs-node/internal/logs"
objectcore "git.frostfs.info/TrueCloudLab/frostfs-node/pkg/core/object"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/services/replicator"
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/client"
containerSDK "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/container"
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/netmap"
objectSDK "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object"
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/erasurecode"
oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id"
"go.uber.org/zap"
"golang.org/x/sync/errgroup"
)
var errNoECinfoReturnded = errors.New("no EC info returned")
type ecChunkProcessResult struct {
validPlacement bool
removeLocal bool
}
var errInvalidECPlacement = errors.New("invalid EC placement: EC placement must have one placement vector with at least one node")
func (p *Policer) processECContainerObject(ctx context.Context, objInfo objectcore.Info, cnr containerSDK.Container) error {
if objInfo.ECInfo == nil {
return p.processECContainerRepObject(ctx, objInfo, cnr.PlacementPolicy())
}
return p.processECContainerECObject(ctx, objInfo, cnr)
}
// processECContainerRepObject processes non erasure coded objects in EC container: tombstones, locks and linking objects.
// All of them must be stored on all of the container nodes.
func (p *Policer) processECContainerRepObject(ctx context.Context, objInfo objectcore.Info, policy netmap.PlacementPolicy) error {
objID := objInfo.Address.Object()
nn, err := p.placementBuilder.BuildPlacement(objInfo.Address.Container(), &objID, policy)
if err != nil {
return fmt.Errorf("%s: %w", logs.PolicerCouldNotBuildPlacementVectorForObject, err)
}
if len(nn) != 1 || len(nn[0]) == 0 {
return errInvalidECPlacement
}
c := &placementRequirements{}
checkedNodes := newNodeCache()
select {
case <-ctx.Done():
return ctx.Err()
default:
}
p.processRepNodes(ctx, c, objInfo, nn[0], uint32(len(nn[0])), checkedNodes)
if !c.needLocalCopy && c.removeLocalCopy {
p.log.Info(ctx, logs.PolicerRedundantLocalObjectCopyDetected,
zap.Stringer("object", objInfo.Address),
)
p.cbRedundantCopy(ctx, objInfo.Address)
}
return nil
}
func (p *Policer) processECContainerECObject(ctx context.Context, objInfo objectcore.Info, cnr containerSDK.Container) error {
nn, err := p.placementBuilder.BuildPlacement(objInfo.Address.Container(), &objInfo.ECInfo.ParentID, cnr.PlacementPolicy())
if err != nil {
return fmt.Errorf("%s: %w", logs.PolicerCouldNotBuildPlacementVectorForObject, err)
}
if len(nn) != 1 || len(nn[0]) == 0 {
return errInvalidECPlacement
}
select {
case <-ctx.Done():
return ctx.Err()
default:
}
res := p.processECChunk(ctx, objInfo, nn[0])
if !res.validPlacement {
// drop local chunk only if all required chunks are in place
res.removeLocal = res.removeLocal && p.pullRequiredECChunks(ctx, objInfo, nn[0], cnr)
}
p.adjustECPlacement(ctx, objInfo, nn[0], cnr)
if res.removeLocal {
p.log.Info(ctx, logs.PolicerRedundantLocalObjectCopyDetected, zap.Stringer("object", objInfo.Address))
p.cbRedundantCopy(ctx, objInfo.Address)
}
return nil
}
// processECChunk replicates EC chunk if needed.
func (p *Policer) processECChunk(ctx context.Context, objInfo objectcore.Info, nodes []netmap.NodeInfo) ecChunkProcessResult {
var removeLocalChunk bool
requiredNode := nodes[int(objInfo.ECInfo.Index)%(len(nodes))]
if p.cfg.netmapKeys.IsLocalKey(requiredNode.PublicKey()) {
// current node is required node, we are happy
return ecChunkProcessResult{
validPlacement: true,
}
}
if requiredNode.Status().IsMaintenance() {
// consider maintenance mode has object, but do not drop local copy
p.log.Debug(ctx, logs.PolicerConsiderNodeUnderMaintenanceAsOK, zap.String("node", netmap.StringifyPublicKey(requiredNode)))
return ecChunkProcessResult{}
}
callCtx, cancel := context.WithTimeout(ctx, p.headTimeout)
_, err := p.remoteHeader(callCtx, requiredNode, objInfo.Address, false)
cancel()
if err == nil {
removeLocalChunk = true
} else if client.IsErrObjectNotFound(err) {
p.log.Debug(ctx, logs.PolicerShortageOfObjectCopiesDetected, zap.Stringer("object", objInfo.Address), zap.Uint32("shortage", 1))
task := replicator.Task{
NumCopies: 1,
Addr: objInfo.Address,
Nodes: []netmap.NodeInfo{requiredNode},
}
p.replicator.HandleReplicationTask(ctx, task, newNodeCache())
} else if client.IsErrNodeUnderMaintenance(err) {
// consider maintenance mode has object, but do not drop local copy
p.log.Debug(ctx, logs.PolicerConsiderNodeUnderMaintenanceAsOK, zap.String("node", netmap.StringifyPublicKey(requiredNode)))
} else {
p.log.Error(ctx, logs.PolicerReceiveObjectHeaderToCheckPolicyCompliance, zap.Stringer("object", objInfo.Address), zap.Error(err))
}
return ecChunkProcessResult{
removeLocal: removeLocalChunk,
}
}
func (p *Policer) pullRequiredECChunks(ctx context.Context, objInfo objectcore.Info, nodes []netmap.NodeInfo, cnr containerSDK.Container) bool {
var parentAddress oid.Address
parentAddress.SetContainer(objInfo.Address.Container())
parentAddress.SetObject(objInfo.ECInfo.ParentID)
requiredChunkIndexes := p.collectRequiredECChunks(nodes, objInfo)
if len(requiredChunkIndexes) == 0 {
p.log.Info(ctx, logs.PolicerNodeIsNotECObjectNode, zap.Stringer("object", objInfo.ECInfo.ParentID))
return true
}
err := p.resolveLocalECChunks(ctx, parentAddress, requiredChunkIndexes)
if err != nil {
p.log.Error(ctx, logs.PolicerFailedToGetLocalECChunks, zap.Error(err), zap.Stringer("object", parentAddress))
return false
}
if len(requiredChunkIndexes) == 0 {
return true
}
indexToObjectID := make(map[uint32]oid.ID)
success := p.resolveRemoteECChunks(ctx, parentAddress, nodes, requiredChunkIndexes, indexToObjectID)
if !success {
return false
}
for index, candidates := range requiredChunkIndexes {
var addr oid.Address
addr.SetContainer(objInfo.Address.Container())
addr.SetObject(indexToObjectID[index])
p.replicator.HandlePullTask(ctx, replicator.Task{
Addr: addr,
Nodes: candidates,
Container: cnr,
})
}
// there was some missing chunks, it's not ok
return false
}
func (p *Policer) collectRequiredECChunks(nodes []netmap.NodeInfo, objInfo objectcore.Info) map[uint32][]netmap.NodeInfo {
requiredChunkIndexes := make(map[uint32][]netmap.NodeInfo)
for i, n := range nodes {
if uint32(i) == objInfo.ECInfo.Total {
break
}
if p.cfg.netmapKeys.IsLocalKey(n.PublicKey()) {
requiredChunkIndexes[uint32(i)] = []netmap.NodeInfo{}
}
}
return requiredChunkIndexes
}
func (p *Policer) resolveLocalECChunks(ctx context.Context, parentAddress oid.Address, required map[uint32][]netmap.NodeInfo) error {
_, err := p.localHeader(ctx, parentAddress)
var eiErr *objectSDK.ECInfoError
if err == nil { // should not be happen
return errNoECinfoReturnded
}
if !errors.As(err, &eiErr) {
return err
}
for _, ch := range eiErr.ECInfo().Chunks {
delete(required, ch.Index)
}
return nil
}
func (p *Policer) resolveRemoteECChunks(ctx context.Context, parentAddress oid.Address, nodes []netmap.NodeInfo, required map[uint32][]netmap.NodeInfo, indexToObjectID map[uint32]oid.ID) bool {
var eiErr *objectSDK.ECInfoError
for _, n := range nodes {
if p.cfg.netmapKeys.IsLocalKey(n.PublicKey()) {
continue
}
_, err := p.remoteHeader(ctx, n, parentAddress, true)
if !errors.As(err, &eiErr) {
continue
}
for _, ch := range eiErr.ECInfo().Chunks {
if candidates, ok := required[ch.Index]; ok {
candidates = append(candidates, n)
required[ch.Index] = candidates
var chunkID oid.ID
if err := chunkID.ReadFromV2(ch.ID); err != nil {
p.log.Error(ctx, logs.PolicerFailedToDecodeECChunkID, zap.Error(err), zap.Stringer("object", parentAddress))
return false
}
if existed, ok := indexToObjectID[ch.Index]; ok && existed != chunkID {
p.log.Error(ctx, logs.PolicerDifferentObjectIDForTheSameECChunk, zap.Stringer("first", existed),
zap.Stringer("second", chunkID), zap.Stringer("object", parentAddress), zap.Uint32("index", ch.Index))
return false
}
indexToObjectID[ch.Index] = chunkID
}
}
}
for index, candidates := range required {
if len(candidates) == 0 {
p.log.Error(ctx, logs.PolicerMissingECChunk, zap.Stringer("object", parentAddress), zap.Uint32("index", index))
return false
}
}
return true
}
func (p *Policer) adjustECPlacement(ctx context.Context, objInfo objectcore.Info, nodes []netmap.NodeInfo, cnr containerSDK.Container) {
var parentAddress oid.Address
parentAddress.SetContainer(objInfo.Address.Container())
parentAddress.SetObject(objInfo.ECInfo.ParentID)
var eiErr *objectSDK.ECInfoError
resolved := make(map[uint32][]netmap.NodeInfo)
chunkIDs := make(map[uint32]oid.ID)
restore := true // do not restore EC chunks if some node returned error
for idx, n := range nodes {
if uint32(idx) >= objInfo.ECInfo.Total && uint32(len(resolved)) == objInfo.ECInfo.Total {
return
}
var err error
if p.cfg.netmapKeys.IsLocalKey(n.PublicKey()) {
_, err = p.localHeader(ctx, parentAddress)
} else {
_, err = p.remoteHeader(ctx, n, parentAddress, true)
}
if errors.As(err, &eiErr) {
for _, ch := range eiErr.ECInfo().Chunks {
resolved[ch.Index] = append(resolved[ch.Index], n)
var ecInfoChunkID oid.ID
if err := ecInfoChunkID.ReadFromV2(ch.ID); err != nil {
p.log.Error(ctx, logs.PolicerFailedToDecodeECChunkID, zap.Error(err), zap.Stringer("object", parentAddress))
return
}
if chunkID, exist := chunkIDs[ch.Index]; exist && chunkID != ecInfoChunkID {
p.log.Error(ctx, logs.PolicerDifferentObjectIDForTheSameECChunk, zap.Stringer("first", chunkID),
zap.Stringer("second", ecInfoChunkID), zap.Stringer("object", parentAddress), zap.Uint32("index", ch.Index))
return
}
chunkIDs[ch.Index] = ecInfoChunkID
}
} else if client.IsErrObjectAlreadyRemoved(err) {
restore = false
} else if !p.cfg.netmapKeys.IsLocalKey(n.PublicKey()) && uint32(idx) < objInfo.ECInfo.Total {
p.log.Warn(ctx, logs.PolicerCouldNotGetObjectFromNodeMoving, zap.String("node", hex.EncodeToString(n.PublicKey())), zap.Stringer("object", parentAddress), zap.Error(err))
p.replicator.HandleReplicationTask(ctx, replicator.Task{
NumCopies: 1,
Addr: objInfo.Address,
Nodes: []netmap.NodeInfo{n},
}, newNodeCache())
restore = false
}
}
if !restore || uint32(len(resolved)) == objInfo.ECInfo.Total {
return
}
if objInfo.ECInfo.Total-uint32(len(resolved)) > cnr.PlacementPolicy().ReplicaDescriptor(0).GetECParityCount() {
var found []uint32
for i := range resolved {
found = append(found, i)
}
p.log.Error(ctx, logs.PolicerCouldNotRestoreObjectNotEnoughChunks, zap.Stringer("object", parentAddress), zap.Uint32s("found_chunks", found))
return
}
p.restoreECObject(ctx, objInfo, parentAddress, nodes, resolved, chunkIDs, cnr)
}
func (p *Policer) restoreECObject(ctx context.Context, objInfo objectcore.Info, parentAddress oid.Address, nodes []netmap.NodeInfo, existedChunks map[uint32][]netmap.NodeInfo, chunkIDs map[uint32]oid.ID,
cnr containerSDK.Container,
) {
c, err := erasurecode.NewConstructor(int(cnr.PlacementPolicy().ReplicaDescriptor(0).GetECDataCount()), int(cnr.PlacementPolicy().ReplicaDescriptor(0).GetECParityCount()))
if err != nil {
p.log.Error(ctx, logs.PolicerFailedToRestoreObject, zap.Stringer("object", parentAddress), zap.Error(err))
return
}
parts := p.collectExistedChunks(ctx, objInfo, existedChunks, parentAddress, chunkIDs)
if parts == nil {
return
}
key, err := p.keyStorage.GetKey(nil)
if err != nil {
p.log.Error(ctx, logs.PolicerFailedToRestoreObject, zap.Stringer("object", parentAddress), zap.Error(err))
return
}
required := make([]bool, len(parts))
for i, p := range parts {
if p == nil {
required[i] = true
}
}
if err := c.ReconstructParts(parts, required, key); err != nil {
p.log.Error(ctx, logs.PolicerFailedToRestoreObject, zap.Stringer("object", parentAddress), zap.Error(err))
return
}
for idx, part := range parts {
if _, exists := existedChunks[uint32(idx)]; exists {
continue
}
var addr oid.Address
addr.SetContainer(parentAddress.Container())
pID, _ := part.ID()
addr.SetObject(pID)
targetNode := nodes[idx%len(nodes)]
if p.cfg.netmapKeys.IsLocalKey(targetNode.PublicKey()) {
p.replicator.HandleLocalPutTask(ctx, replicator.Task{
Addr: addr,
Obj: part,
Container: cnr,
})
} else {
p.replicator.HandleReplicationTask(ctx, replicator.Task{
NumCopies: 1,
Addr: addr,
Nodes: []netmap.NodeInfo{targetNode},
Obj: part,
}, newNodeCache())
}
}
}
func (p *Policer) collectExistedChunks(ctx context.Context, objInfo objectcore.Info, existedChunks map[uint32][]netmap.NodeInfo, parentAddress oid.Address, chunkIDs map[uint32]oid.ID) []*objectSDK.Object {
parts := make([]*objectSDK.Object, objInfo.ECInfo.Total)
errGroup, egCtx := errgroup.WithContext(ctx)
for idx, nodes := range existedChunks {
errGroup.Go(func() error {
var objID oid.Address
objID.SetContainer(parentAddress.Container())
objID.SetObject(chunkIDs[idx])
var obj *objectSDK.Object
var err error
for _, node := range nodes {
if p.cfg.netmapKeys.IsLocalKey(node.PublicKey()) {
obj, err = p.localObject(egCtx, objID)
} else {
obj, err = p.remoteObject(egCtx, node, objID)
}
if err == nil {
break
}
p.log.Warn(ctx, logs.PolicerCouldNotGetChunk, zap.Stringer("object", parentAddress), zap.Stringer("chunkID", objID), zap.Error(err), zap.String("node", hex.EncodeToString(node.PublicKey())))
}
if obj != nil {
parts[idx] = obj
}
return nil
})
}
if err := errGroup.Wait(); err != nil {
p.log.Error(ctx, logs.PolicerCouldNotGetChunks, zap.Stringer("object", parentAddress), zap.Error(err))
return nil
}
return parts
}