frostfs-sdk-go/audit/result.go

378 lines
9.5 KiB
Go

package audit
import (
"errors"
"fmt"
"github.com/nspcc-dev/neofs-api-go/v2/audit"
"github.com/nspcc-dev/neofs-api-go/v2/refs"
cid "github.com/nspcc-dev/neofs-sdk-go/container/id"
oid "github.com/nspcc-dev/neofs-sdk-go/object/id"
"github.com/nspcc-dev/neofs-sdk-go/version"
)
// Result represents report on the results of the data audit in NeoFS system.
//
// Result is mutually binary-compatible with github.com/nspcc-dev/neofs-api-go/v2/audit.DataAuditResult
// message. See Marshal / Unmarshal methods.
//
// Instances can be created using built-in var declaration.
type Result struct {
versionEncoded bool
v2 audit.DataAuditResult
}
// Marshal encodes Result into a canonical NeoFS binary format (Protocol Buffers
// with direct field order).
//
// Writes version.Current() protocol version into the resulting message if Result
// hasn't been already decoded from such a message using Unmarshal.
//
// See also Unmarshal.
func (r *Result) Marshal() []byte {
if !r.versionEncoded {
var verV2 refs.Version
version.Current().WriteToV2(&verV2)
r.v2.SetVersion(&verV2)
r.versionEncoded = true
}
return r.v2.StableMarshal(nil)
}
var errCIDNotSet = errors.New("container ID is not set")
// Unmarshal decodes Result from its canonical NeoFS binary format (Protocol Buffers
// with direct field order). Returns an error describing a format violation.
//
// See also Marshal.
func (r *Result) Unmarshal(data []byte) error {
err := r.v2.Unmarshal(data)
if err != nil {
return err
}
r.versionEncoded = true
// format checks
var cID cid.ID
cidV2 := r.v2.GetContainerID()
if cidV2 == nil {
return errCIDNotSet
}
err = cID.ReadFromV2(*cidV2)
if err != nil {
return fmt.Errorf("could not convert V2 container ID: %w", err)
}
var (
oID oid.ID
oidV2 refs.ObjectID
)
for _, oidV2 = range r.v2.GetPassSG() {
err = oID.ReadFromV2(oidV2)
if err != nil {
return fmt.Errorf("invalid passed storage group ID: %w", err)
}
}
for _, oidV2 = range r.v2.GetFailSG() {
err = oID.ReadFromV2(oidV2)
if err != nil {
return fmt.Errorf("invalid failed storage group ID: %w", err)
}
}
return nil
}
// Epoch returns NeoFS epoch when the data associated with the Result was audited.
//
// Zero Result has zero epoch.
//
// See also ForEpoch.
func (r Result) Epoch() uint64 {
return r.v2.GetAuditEpoch()
}
// ForEpoch specifies NeoFS epoch when the data associated with the Result was audited.
//
// See also Epoch.
func (r *Result) ForEpoch(epoch uint64) {
r.v2.SetAuditEpoch(epoch)
}
// Container returns identifier of the container with which the data audit Result
// is associated and a bool that indicates container ID field presence in the Result.
//
// Zero Result does not have container ID.
//
// See also ForContainer.
func (r Result) Container() (cid.ID, bool) {
var cID cid.ID
cidV2 := r.v2.GetContainerID()
if cidV2 != nil {
_ = cID.ReadFromV2(*cidV2)
return cID, true
}
return cID, false
}
// ForContainer sets identifier of the container with which the data audit Result
// is associated.
//
// See also Container.
func (r *Result) ForContainer(cnr cid.ID) {
var cidV2 refs.ContainerID
cnr.WriteToV2(&cidV2)
r.v2.SetContainerID(&cidV2)
}
// AuditorKey returns public key of the auditing NeoFS Inner Ring node in
// a NeoFS binary key format.
//
// Zero Result has nil key. Return value MUST NOT be mutated: to do this,
// first make a copy.
//
// See also SetAuditorPublicKey.
func (r Result) AuditorKey() []byte {
return r.v2.GetPublicKey()
}
// SetAuditorKey specifies public key of the auditing NeoFS Inner Ring node in
// a NeoFS binary key format.
//
// Argument MUST NOT be mutated at least until the end of using the Result.
//
// See also AuditorKey.
func (r *Result) SetAuditorKey(key []byte) {
r.v2.SetPublicKey(key)
}
// Completed returns completion state of the data audit associated with the Result.
//
// Zero Result corresponds to incomplete data audit.
//
// See also Complete.
func (r Result) Completed() bool {
return r.v2.GetComplete()
}
// Complete marks the data audit associated with the Result as completed.
//
// See also Completed.
func (r *Result) Complete() {
r.v2.SetComplete(true)
}
// RequestsPoR returns number of requests made by Proof-of-Retrievability
// audit check to get all headers of the objects inside storage groups.
//
// Zero Result has zero requests.
//
// See also SetRequestsPoR.
func (r Result) RequestsPoR() uint32 {
return r.v2.GetRequests()
}
// SetRequestsPoR sets number of requests made by Proof-of-Retrievability
// audit check to get all headers of the objects inside storage groups.
//
// See also RequestsPoR.
func (r *Result) SetRequestsPoR(v uint32) {
r.v2.SetRequests(v)
}
// RetriesPoR returns number of retries made by Proof-of-Retrievability
// audit check to get all headers of the objects inside storage groups.
//
// Zero Result has zero retries.
//
// See also SetRetriesPoR.
func (r Result) RetriesPoR() uint32 {
return r.v2.GetRetries()
}
// SetRetriesPoR sets number of retries made by Proof-of-Retrievability
// audit check to get all headers of the objects inside storage groups.
//
// See also RetriesPoR.
func (r *Result) SetRetriesPoR(v uint32) {
r.v2.SetRetries(v)
}
// IteratePassedStorageGroups iterates over all storage groups that passed
// Proof-of-Retrievability audit check and passes them into f. Breaks on f's
// false return, f MUST NOT be nil.
//
// Zero Result has no passed storage groups and doesn't call f.
//
// See also SubmitPassedStorageGroup.
func (r Result) IteratePassedStorageGroups(f func(oid.ID) bool) {
r2 := r.v2.GetPassSG()
var id oid.ID
for i := range r2 {
_ = id.ReadFromV2(r2[i])
if !f(id) {
return
}
}
}
// SubmitPassedStorageGroup marks storage group as passed Proof-of-Retrievability
// audit check.
//
// See also IteratePassedStorageGroups.
func (r *Result) SubmitPassedStorageGroup(sg oid.ID) {
var idV2 refs.ObjectID
sg.WriteToV2(&idV2)
r.v2.SetPassSG(append(r.v2.GetPassSG(), idV2))
}
// IterateFailedStorageGroups is similar to IteratePassedStorageGroups but for failed groups.
//
// See also SubmitFailedStorageGroup.
func (r Result) IterateFailedStorageGroups(f func(oid.ID) bool) {
v := r.v2.GetFailSG()
var id oid.ID
for i := range v {
_ = id.ReadFromV2(v[i])
if !f(id) {
return
}
}
}
// SubmitFailedStorageGroup is similar to SubmitPassedStorageGroup but for failed groups.
//
// See also IterateFailedStorageGroups.
func (r *Result) SubmitFailedStorageGroup(sg oid.ID) {
var idV2 refs.ObjectID
sg.WriteToV2(&idV2)
r.v2.SetFailSG(append(r.v2.GetFailSG(), idV2))
}
// Hits returns number of sampled objects under audit placed
// in an optimal way according to the container's placement policy
// when checking Proof-of-Placement.
//
// Zero result has zero hits.
//
// See also SetHits.
func (r Result) Hits() uint32 {
return r.v2.GetHit()
}
// SetHits sets number of sampled objects under audit placed
// in an optimal way according to the containers placement policy
// when checking Proof-of-Placement.
//
// See also Hits.
func (r *Result) SetHits(hit uint32) {
r.v2.SetHit(hit)
}
// Misses returns number of sampled objects under audit placed
// in suboptimal way according to the container's placement policy,
// but still at a satisfactory level when checking Proof-of-Placement.
//
// Zero Result has zero misses.
//
// See also SetMisses.
func (r Result) Misses() uint32 {
return r.v2.GetMiss()
}
// SetMisses sets number of sampled objects under audit placed
// in suboptimal way according to the container's placement policy,
// but still at a satisfactory level when checking Proof-of-Placement.
//
// See also Misses.
func (r *Result) SetMisses(miss uint32) {
r.v2.SetMiss(miss)
}
// Failures returns number of sampled objects under audit stored
// in a way not confirming placement policy or not found at all
// when checking Proof-of-Placement.
//
// Zero result has zero failures.
//
// See also SetFailures.
func (r Result) Failures() uint32 {
return r.v2.GetFail()
}
// SetFailures sets number of sampled objects under audit stored
// in a way not confirming placement policy or not found at all
// when checking Proof-of-Placement.
//
// See also Failures.
func (r *Result) SetFailures(fail uint32) {
r.v2.SetFail(fail)
}
// IteratePassedStorageNodes iterates over all storage nodes that passed at least one
// Proof-of-Data-Possession audit check and passes their public keys into f. Breaks on
// f's false return.
//
// f MUST NOT be nil and MUST NOT mutate parameter passed into it at least until
// the end of using the Result.
//
// Zero Result has no passed storage nodes and doesn't call f.
//
// See also SubmitPassedStorageNode.
func (r Result) IteratePassedStorageNodes(f func([]byte) bool) {
v := r.v2.GetPassNodes()
for i := range v {
if !f(v[i]) {
return
}
}
}
// SubmitPassedStorageNodes marks storage node list as passed Proof-of-Data-Possession
// audit check. The list contains public keys.
//
// Argument and its elements MUST NOT be mutated at least until the end of using the Result.
//
// See also IteratePassedStorageNodes.
func (r *Result) SubmitPassedStorageNodes(list [][]byte) {
r.v2.SetPassNodes(list)
}
// IterateFailedStorageNodes is similar to IteratePassedStorageNodes but for failed nodes.
//
// See also SubmitPassedStorageNodes.
func (r Result) IterateFailedStorageNodes(f func([]byte) bool) {
v := r.v2.GetFailNodes()
for i := range v {
if !f(v[i]) {
return
}
}
}
// SubmitFailedStorageNodes is similar to SubmitPassedStorageNodes but for failed nodes.
//
// See also IterateFailedStorageNodes.
func (r *Result) SubmitFailedStorageNodes(list [][]byte) {
r.v2.SetFailNodes(list)
}