frostfs-node/pkg/services/control/server/evacuate.go
Dmitrii Stepanov 3a797e4682
All checks were successful
DCO action / DCO (pull_request) Successful in 13m4s
Vulncheck / Vulncheck (pull_request) Successful in 15m32s
Build / Build Components (1.22) (pull_request) Successful in 16m30s
Build / Build Components (1.21) (pull_request) Successful in 16m37s
Tests and linters / gopls check (pull_request) Successful in 20m33s
Tests and linters / Staticcheck (pull_request) Successful in 21m17s
Tests and linters / Lint (pull_request) Successful in 21m37s
Tests and linters / Tests with -race (pull_request) Successful in 26m40s
Tests and linters / Tests (1.21) (pull_request) Successful in 26m54s
Tests and linters / Tests (1.22) (pull_request) Successful in 27m59s
Pre-commit hooks / Pre-commit (pull_request) Successful in 39m39s
[#1222] engine: Fix tree evacuation
Do not fail if it is unable to evacuate tree to other node.

Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
2024-07-04 10:38:10 +03:00

188 lines
4.8 KiB
Go

package control
import (
"bytes"
"context"
"crypto/sha256"
"encoding/hex"
"errors"
"fmt"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/engine"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/local_object_storage/pilorama"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/services/control"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/services/control/server/ctrlmessage"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/services/object_manager/placement"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/services/replicator"
"git.frostfs.info/TrueCloudLab/frostfs-node/pkg/services/tree"
cid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/container/id"
"git.frostfs.info/TrueCloudLab/frostfs-sdk-go/netmap"
objectSDK "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object"
oid "git.frostfs.info/TrueCloudLab/frostfs-sdk-go/object/id"
"github.com/nspcc-dev/neo-go/pkg/crypto/keys"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/status"
)
var errFailedToBuildListOfContainerNodes = errors.New("can't build a list of container nodes")
func (s *Server) EvacuateShard(ctx context.Context, req *control.EvacuateShardRequest) (*control.EvacuateShardResponse, error) {
err := s.isValidRequest(req)
if err != nil {
return nil, status.Error(codes.PermissionDenied, err.Error())
}
prm := engine.EvacuateShardPrm{
ShardID: s.getShardIDList(req.GetBody().GetShard_ID()),
IgnoreErrors: req.GetBody().GetIgnoreErrors(),
ObjectsHandler: s.replicateObject,
Scope: engine.EvacuateScopeObjects,
}
res, err := s.s.Evacuate(ctx, prm)
if err != nil {
return nil, status.Error(codes.Internal, err.Error())
}
resp := &control.EvacuateShardResponse{
Body: &control.EvacuateShardResponse_Body{
Count: uint32(res.ObjectsEvacuated()),
},
}
err = ctrlmessage.Sign(s.key, resp)
if err != nil {
return nil, status.Error(codes.Internal, err.Error())
}
return resp, nil
}
func (s *Server) replicateObject(ctx context.Context, addr oid.Address, obj *objectSDK.Object) (bool, error) {
cid, ok := obj.ContainerID()
if !ok {
// Return nil to prevent situations where a shard can't be evacuated
// because of a single bad/corrupted object.
return false, nil
}
nodes, err := s.getContainerNodes(cid)
if err != nil {
return false, err
}
if len(nodes) == 0 {
return false, nil
}
var res replicatorResult
task := replicator.Task{
NumCopies: 1,
Addr: addr,
Obj: obj,
Nodes: nodes,
}
s.replicator.HandleReplicationTask(ctx, task, &res)
if res.count == 0 {
return false, errors.New("object was not replicated")
}
return true, nil
}
func (s *Server) replicateTree(ctx context.Context, contID cid.ID, treeID string, forest pilorama.Forest) (bool, string, error) {
nodes, err := s.getContainerNodes(contID)
if err != nil {
return false, "", err
}
if len(nodes) == 0 {
return false, "", nil
}
for _, node := range nodes {
err = s.replicateTreeToNode(ctx, forest, contID, treeID, node)
if err == nil {
return true, hex.EncodeToString(node.PublicKey()), nil
}
}
return false, "", err
}
func (s *Server) replicateTreeToNode(ctx context.Context, forest pilorama.Forest, contID cid.ID, treeID string, node netmap.NodeInfo) error {
rawCID := make([]byte, sha256.Size)
contID.Encode(rawCID)
var height uint64
for {
op, err := forest.TreeGetOpLog(ctx, contID, treeID, height)
if err != nil {
return err
}
if op.Time == 0 {
return nil
}
req := &tree.ApplyRequest{
Body: &tree.ApplyRequest_Body{
ContainerId: rawCID,
TreeId: treeID,
Operation: &tree.LogMove{
ParentId: op.Parent,
Meta: op.Meta.Bytes(),
ChildId: op.Child,
},
},
}
err = tree.SignMessage(req, s.key)
if err != nil {
return fmt.Errorf("can't message apply request: %w", err)
}
err = s.treeService.ReplicateTreeOp(ctx, node, req)
if err != nil {
return err
}
height = op.Time + 1
}
}
func (s *Server) getContainerNodes(contID cid.ID) ([]netmap.NodeInfo, error) {
nm, err := s.netMapSrc.GetNetMap(0)
if err != nil {
return nil, err
}
c, err := s.cnrSrc.Get(contID)
if err != nil {
return nil, err
}
binCnr := make([]byte, sha256.Size)
contID.Encode(binCnr)
ns, err := nm.ContainerNodes(c.Value.PlacementPolicy(), binCnr)
if err != nil {
return nil, errFailedToBuildListOfContainerNodes
}
nodes := placement.FlattenNodes(ns)
bs := (*keys.PublicKey)(&s.key.PublicKey).Bytes()
for i := 0; i < len(nodes); i++ {
if bytes.Equal(nodes[i].PublicKey(), bs) {
copy(nodes[i:], nodes[i+1:])
nodes = nodes[:len(nodes)-1]
}
}
return nodes, nil
}
type replicatorResult struct {
count int
}
// SubmitSuccessfulReplication implements the replicator.TaskResult interface.
func (r *replicatorResult) SubmitSuccessfulReplication(_ netmap.NodeInfo) {
r.count++
}