EC Audit: H6: partly stored EC PUT is answered with "internal server error" and no object ID

未关闭
#4,331 2 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
3/5
预计耗时
1-2 天
新手友好度
68/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
go

调研方向

Start in pkg/services/object/put/distributed.go, especially handleECRule, Close, and the REP branch that uses newCompletionError. Compare how EC and REP failures are classified when some parts or replicas are stored, then add or update a focused test in pkg/services/object/put/ for the EC case. Done means a partial EC PUT returns INCOMPLETE with the object ID, consistent with the expected behavior in the issue.

由索引模型根据 Issue 内容生成。

描述

bug hal:proposed neofs-storage

Expected Behavior

When a PUT into an EC container stores some parts but not all, the client gets status INCOMPLETE and the object ID, as it does for a REP container. It can then tell "nothing was stored" from "stored but under-replicated", and can read or delete the object.

Current Behavior

The EC path returns errIncompletePut, which has neither Is nor Unwrap (put/distributed.go#L108-L121), and handleECRule passes it on as is (put/distributed.go#L468-L474). Close returns the object ID only for errors that match apistatus.ErrIncomplete (put/distributed.go#L305-L311), so this one comes back with a zero ID and is converted to status 1024. The REP path wraps its failure with newCompletionError (put/distributed.go#L557-L562) and gets INCOMPLETE.

Test below, three nodes, one of them refuses to store, PUT with node-side encoding:

Policy Stored Status Object ID in the reply
EC 2/1 2 of 3 parts 1024 internal server error none
REP 3 2 of 3 replicas 1 incomplete yes
status=1024 msg="incomplete object PUT by placement: only 2 EC parts were put successfully for 2/1 EC rule, latest error: save on SN #2: …"

In the EC case two parts of a 2/1 rule are stored, so the object is complete enough to be read. The client is told the PUT failed and has no ID. A retry stores a second object; the parts of the first stay.

Possible Solution

Wrap the EC failure with newCompletionError(err, <any part stored>) as the REP branch does, so that Close returns the ID with INCOMPLETE. The existing test for this error checks only the message.

Steps to Reproduce (for bugs)

  1. Save the test below as pkg/services/object/put/ecaudit_test.go.
  2. go test ./pkg/services/object/put/ -run TestECAuditH6 -v

It passes, that is, it asserts status 1024 and a zero ID for EC, and INCOMPLETE with an ID for REP.

ecaudit_test.go
package putsvc

import (
	"context"
	"errors"
	"fmt"
	"testing"

	iec "github.com/nspcc-dev/neofs-node/internal/ec"
	"github.com/nspcc-dev/neofs-node/internal/testutil"
	"github.com/nspcc-dev/neofs-node/pkg/services/object/common"
	objutil "github.com/nspcc-dev/neofs-node/pkg/services/object/util"
	svcutil "github.com/nspcc-dev/neofs-node/pkg/services/util"
	apistatus "github.com/nspcc-dev/neofs-sdk-go/client/status"
	"github.com/nspcc-dev/neofs-sdk-go/container"
	cidtest "github.com/nspcc-dev/neofs-sdk-go/container/id/test"
	"github.com/nspcc-dev/neofs-sdk-go/netmap"
	"github.com/nspcc-dev/neofs-sdk-go/object"
	oid "github.com/nspcc-dev/neofs-sdk-go/object/id"
	sessionv2 "github.com/nspcc-dev/neofs-sdk-go/session/v2"
	"github.com/nspcc-dev/neofs-sdk-go/user"
	usertest "github.com/nspcc-dev/neofs-sdk-go/user/test"
	"github.com/stretchr/testify/require"
)

func ecAuditECContainer(rules []iec.Rule, repRules []uint32, ip *netmap.InitialPlacementPolicy) container.Container {
	var policy netmap.PlacementPolicy
	ecRules := make([]netmap.ECRule, len(rules))
	for i := range rules {
		ecRules[i] = netmap.NewECRule(uint32(rules[i].DataPartNum), uint32(rules[i].ParityPartNum))
	}
	policy.SetECRules(ecRules)
	if len(repRules) > 0 {
		rds := make([]netmap.ReplicaDescriptor, len(repRules))
		for i := range repRules {
			rds[i].SetNumberOfObjects(repRules[i])
		}
		policy.SetReplicas(rds)
	}
	if ip != nil {
		policy.SetInitial(*ip)
	}
	var cnr container.Container
	cnr.SetPlacementPolicy(policy)
	return cnr
}

// EC-only cluster, all rules share one node list of given size.
func ecAuditECCluster(t *testing.T, rules []iec.Rule, nodes uint, ip *netmap.InitialPlacementPolicy) *testCluster {
	cluster := newTestClusterForRepPolicyWithContainer(t, nodes, 0, 0, ecAuditECContainer(rules, nil, ip))
	for i := range cluster.nodeNetworks {
		cn := &cluster.nodeNetworks[i].cnrNodes
		cn.repCounts = nil
		for range len(rules) - 1 {
			cn.unsorted = append(cn.unsorted, cn.unsorted[0])
			cn.sorted = append(cn.sorted, cn.sorted[0])
		}
		cn.ecRules = rules
	}
	return cluster
}

func ecAuditPut(svc *Service, obj object.Object, st2 *sessionv2.Token, ttl uint32) (oid.ID, error) {
	stream, err := svc.Put(context.Background())
	if err != nil {
		return oid.ID{}, err
	}
	commonPrm := objutil.CommonPrmFromRequest(ttl, nil, common.RequestTokens{Session: st2})
	if err = stream.Init(new(PutInitPrm).WithObject(obj.CutPayload()).WithCommonPrm(commonPrm)); err != nil {
		return oid.ID{}, fmt.Errorf("init: %w", err)
	}
	if err = stream.SendChunk(new(PutChunkPrm).WithChunk(obj.Payload())); err != nil {
		return oid.ID{}, fmt.Errorf("chunk: %w", err)
	}
	return stream.Close()
}

func ecAuditTokenForNode(t *testing.T, cluster *testCluster, nodeIdx int, owner usertest.UserSigner) *sessionv2.Token {
	tok := newSessionTokenV2(t, cidtest.ID(), owner, nil, []sessionv2.Verb{sessionv2.VerbObjectPut})
	pk := cluster.nodeSessions[nodeIdx].signer.ECDSAPrivateKey.PublicKey
	require.NoError(t, tok.SetSubjects([]sessionv2.Target{sessionv2.NewTargetUser(user.NewFromECDSAPublicKey(pk))}))
	require.NoError(t, tok.Sign(owner))
	return tok
}

func ecAuditSrcObject(owner usertest.UserSigner, ln int) object.Object {
	var obj object.Object
	obj.SetContainerID(cidtest.ID())
	obj.SetOwner(owner.UserID())
	obj.SetAttributes(object.NewAttribute("attr1", "val1"))
	obj.SetPayload(testutil.RandByteSlice(ln))
	return obj
}

func TestECAuditH6_PartialECPut(t *testing.T) {
	owner := usertest.User()

	t.Run("EC 2/1, one of three nodes fails", func(t *testing.T) {
		rules := []iec.Rule{{DataPartNum: 2, ParityPartNum: 1}}
		cluster := ecAuditECCluster(t, rules, 3, nil)
		cluster.nodeLocalStorages[2].err = errors.New("cannot store object")

		id, err := ecAuditPut(cluster.nodeServices[0].svc, ecAuditSrcObject(owner, 100), ecAuditTokenForNode(t, cluster, 0, owner), 2)
		require.Error(t, err)
		require.NotErrorIs(t, err, apistatus.ErrIncomplete)
		require.True(t, id.IsZero())

		// same wrapping and conversion as putStream.close + Server.sendPutResponse
		st := svcutil.ToStatus(fmt.Errorf("close put stream: %w", err))
		require.EqualValues(t, 1024, st.GetCode())

		var stored int
		for i := range cluster.nodeLocalStorages {
			stored += len(cluster.nodeLocalStorages[i].objs)
		}
		require.Equal(t, 2, stored) // == DataPartNum, i.e. object is fully recoverable
		t.Logf("H6/EC: id=%s status=%d msg=%q, parts stored cluster-wide=%d", id, st.GetCode(), st.GetMessage(), stored)
	})

	t.Run("REP 3, one of three nodes fails (contrast)", func(t *testing.T) {
		cluster := newTestClusterForRepPolicy(t, 3, 0, 0)
		cluster.nodeLocalStorages[2].err = errors.New("cannot store object")

		id, err := ecAuditPut(cluster.nodeServices[0].svc, ecAuditSrcObject(owner, 100), ecAuditTokenForNode(t, cluster, 0, owner), 2)
		require.ErrorIs(t, err, apistatus.ErrIncomplete)
		require.False(t, id.IsZero())
		st := svcutil.ToStatus(fmt.Errorf("close put stream: %w", err))
		t.Logf("H6/REP: id=%s status=%d", id, st.GetCode())
	})
}

Context

Found by an audit of the erasure coding code. The audit was static: code reading plus unit-level probe tests in the packages concerned. No cluster was run, so there is no end-to-end reproduction.

Regression

Not known.

Your Environment

  • Version used: 0.58.2 (1f0d51b5d)
  • Server setup and configuration: none, unit tests only
  • Operating System and version (uname -a): Darwin 25.6.0 arm64, Go 1.27.3
主要语言
Go
星标
37
派生
56
平均合并
1 天 18 小时
30 天内合并 PR
42

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

nspcc-dev/neofs-node 的其他 Issue

查看 nspcc-dev/neofs-node 的全部 Issue

相似的 Issue

更多 Go Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。