Files
prysm/beacon-chain/sync/data_column_sidecars.go
Manu NALEPA 2773bdef89 Remove NUMBER_OF_COLUMNS and MAX_CELLS_IN_EXTENDED_MATRIX configuration. (#16073)
**What type of PR is this?**
Other

**What does this PR do? Why is it needed?**
This pull request removes `NUMBER_OF_COLUMNS` and
`MAX_CELLS_IN_EXTENDED_MATRIX` configuration.

**Other notes for review**
Please read commit by commit, with commit messages.

**Acknowledgements**
- [x] I have read
[CONTRIBUTING.md](https://github.com/prysmaticlabs/prysm/blob/develop/CONTRIBUTING.md).
- [x] I have included a uniquely named [changelog fragment
file](https://github.com/prysmaticlabs/prysm/blob/develop/CONTRIBUTING.md#maintaining-changelogmd).
- [x] I have added a description to this PR with sufficient context for
reviewers to understand this PR.
2025-11-29 09:30:54 +00:00

1186 lines
43 KiB
Go

package sync
import (
"bytes"
"context"
"maps"
"slices"
"sync"
"time"
"github.com/OffchainLabs/prysm/v7/beacon-chain/blockchain"
"github.com/OffchainLabs/prysm/v7/beacon-chain/core/helpers"
"github.com/OffchainLabs/prysm/v7/beacon-chain/core/peerdas"
"github.com/OffchainLabs/prysm/v7/beacon-chain/db/filesystem"
prysmP2P "github.com/OffchainLabs/prysm/v7/beacon-chain/p2p"
p2ptypes "github.com/OffchainLabs/prysm/v7/beacon-chain/p2p/types"
"github.com/OffchainLabs/prysm/v7/beacon-chain/verification"
fieldparams "github.com/OffchainLabs/prysm/v7/config/fieldparams"
"github.com/OffchainLabs/prysm/v7/config/params"
"github.com/OffchainLabs/prysm/v7/consensus-types/blocks"
"github.com/OffchainLabs/prysm/v7/consensus-types/primitives"
leakybucket "github.com/OffchainLabs/prysm/v7/container/leaky-bucket"
"github.com/OffchainLabs/prysm/v7/crypto/rand"
ethpb "github.com/OffchainLabs/prysm/v7/proto/prysm/v1alpha1"
goPeer "github.com/libp2p/go-libp2p/core/peer"
"github.com/pkg/errors"
"github.com/sirupsen/logrus"
)
// DataColumnSidecarsParams stores the common parameters needed to
// fetch data column sidecars from peers.
type DataColumnSidecarsParams struct {
Ctx context.Context // Context
Tor blockchain.TemporalOracle // Temporal oracle, useful to get the current slot
P2P prysmP2P.P2P // P2P network interface
RateLimiter *leakybucket.Collector // Rate limiter for outgoing requests
CtxMap ContextByteVersions // Context map, useful to know if a message is mapped to the correct fork
Storage filesystem.DataColumnStorageReader // Data columns storage
NewVerifier verification.NewDataColumnsVerifier // Data columns verifier to check to conformity of incoming data column sidecars
DownscorePeerOnRPCFault bool // Downscore a peer if it commits an RPC fault. Not responding sidecars at all is considered as a fault.
}
// FetchDataColumnSidecars retrieves data column sidecars for the given blocks and indices
// using a series of fallback strategies.
//
// For each block in `roBlocks` that has commitments, the function attempts to obtain
// all sidecars corresponding to the indices listed in `requestedIndices`.
//
// The function returns:
// - A map from block root to the sidecars successfully retrieved.
// - A set of block roots for which not all requested sidecars could be retrieved.
//
// Retrieval strategy (proceeds to the next step only if not all requested sidecars
// were successfully obtained at the current step):
// 1. Attempt to load the requested sidecars from storage, reconstructing them from
// other available sidecars in storage if necessary.
// 2. Request any missing sidecars from peers. If some are still missing, attempt to
// reconstruct them using both stored sidecars and those retrieved from peers.
// 3. Request all remaining possible sidecars from peers that are not already in storage
// or retrieved in step 2. Stop once either all requested sidecars are retrieved,
// or enough sidecars are available (from storage, step 2, and step 3) to reconstruct
// the requested ones.
func FetchDataColumnSidecars(
params DataColumnSidecarsParams,
roBlocks []blocks.ROBlock,
requestedIndices map[uint64]bool,
) (map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, map[[fieldparams.RootLength]byte]map[uint64]bool, error) {
if len(roBlocks) == 0 || len(requestedIndices) == 0 {
return nil, nil, nil
}
blockCount := len(roBlocks)
// We first consider all requested roots as incomplete, and remove roots from this
// set as we retrieve them.
incompleteRoots := make(map[[fieldparams.RootLength]byte]bool, blockCount)
slotsWithCommitments := make(map[primitives.Slot]bool, blockCount)
slotByRoot := make(map[[fieldparams.RootLength]byte]primitives.Slot, blockCount)
storedIndicesByRoot := make(map[[fieldparams.RootLength]byte]map[uint64]bool, blockCount)
for _, roBlock := range roBlocks {
block := roBlock.Block()
commitments, err := block.Body().BlobKzgCommitments()
if err != nil {
return nil, nil, errors.Wrapf(err, "get blob kzg commitments for block root %#x", roBlock.Root())
}
if len(commitments) == 0 {
continue
}
root := roBlock.Root()
slot := block.Slot()
incompleteRoots[root] = true
slotByRoot[root] = slot
slotsWithCommitments[slot] = true
storedIndices := params.Storage.Summary(root).Stored()
if len(storedIndices) > 0 {
storedIndicesByRoot[root] = storedIndices
}
}
initialMissingRootCount := len(incompleteRoots)
// Request sidecars from storage (by reconstructing them from other available sidecars if needed).
result, err := requestSidecarsFromStorage(params.Storage, storedIndicesByRoot, requestedIndices, incompleteRoots)
if err != nil {
return nil, nil, errors.Wrap(err, "request sidecars from storage")
}
log := log.WithField("initialMissingRootCount", initialMissingRootCount)
if len(incompleteRoots) == 0 {
log.WithField("finalMissingRootCount", 0).Debug("Fetched data column sidecars from storage")
return result, nil, nil
}
// Request direct sidecars from peers.
directSidecarsByRoot, err := requestDirectSidecarsFromPeers(params, slotByRoot, requestedIndices, slotsWithCommitments, storedIndicesByRoot, incompleteRoots)
if err != nil {
return nil, nil, errors.Wrap(err, "request direct sidecars from peers")
}
// Merge sidecars in storage and those received from peers. Reconstruct if needed.
mergedSidecarsByRoot, err := mergeAvailableSidecars(params.Storage, requestedIndices, storedIndicesByRoot, incompleteRoots, directSidecarsByRoot)
if err != nil {
return nil, nil, errors.Wrap(err, "try merge storage and mandatory inputs")
}
maps.Copy(result, mergedSidecarsByRoot)
if len(incompleteRoots) == 0 {
log.WithField("finalMissingRootCount", 0).Debug("Fetched data column sidecars from storage and peers")
return result, nil, nil
}
// Request all possible indirect sidecars from peers which are neither stored nor in `directSidecarsByRoot`
indirectSidecarsByRoot, err := requestIndirectSidecarsFromPeers(params, slotByRoot, slotsWithCommitments, storedIndicesByRoot, directSidecarsByRoot, requestedIndices, incompleteRoots)
if err != nil {
return nil, nil, errors.Wrap(err, "request all sidecars from peers")
}
// Merge sidecars in storage and those received from peers. Reconstruct if needed.
mergedSidecarsByRoot, err = mergeAvailableSidecars(params.Storage, requestedIndices, storedIndicesByRoot, incompleteRoots, indirectSidecarsByRoot)
if err != nil {
return nil, nil, errors.Wrap(err, "try merge storage and all inputs")
}
maps.Copy(result, mergedSidecarsByRoot)
if len(incompleteRoots) == 0 {
log.WithField("finalMissingRootCount", 0).Debug("Fetched data column sidecars from storage and peers using rescue mode")
return result, nil, nil
}
// For remaining incomplete roots, assemble what is available.
incompleteSidecarsByRoot, missingByRoot, err := assembleAvailableSidecars(params.Storage, requestedIndices, incompleteRoots, directSidecarsByRoot)
if err != nil {
return nil, nil, errors.Wrap(err, "assemble available sidecars for incomplete roots")
}
maps.Copy(result, incompleteSidecarsByRoot)
log.WithField("finalMissingRootCount", len(incompleteRoots)).Warning("Failed to fetch data column sidecars")
return result, missingByRoot, nil
}
// requestSidecarsFromStorage attempts to retrieve data column sidecars for each block root in `roots`
// and for all indices specified in `requestedIndices`.
//
// If not all requested sidecars can be obtained for a given root, that root is excluded from the result.
// It returns a map from each root to its successfully retrieved sidecars.
//
// WARNING: This function mutates `roots` by removing entries for which all requested sidecars
// were successfully retrieved.
func requestSidecarsFromStorage(
storage filesystem.DataColumnStorageReader,
storedIndicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool,
requestedIndicesMap map[uint64]bool,
roots map[[fieldparams.RootLength]byte]bool,
) (map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, error) {
requestedIndices := helpers.SortedSliceFromMap(requestedIndicesMap)
result := make(map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, len(roots))
for root := range roots {
storedIndices := storedIndicesByRoot[root]
// Check if all requested indices are stored.
allAvailable := true
for index := range requestedIndicesMap {
if !storedIndices[index] {
allAvailable = false
break
}
}
// Skip if not all requested indices are stored.
if !allAvailable {
continue
}
// All requested indices are stored, retrieve them.
verifiedRoSidecars, err := storage.Get(root, requestedIndices)
if err != nil {
return nil, errors.Wrapf(err, "storage get for block root %#x", root)
}
result[root] = verifiedRoSidecars
delete(roots, root)
}
return result, nil
}
// requestDirectSidecarsFromPeers tries to fetch missing data column sidecars from connected peers.
// It searches through the available peers to identify those responsible for the requested columns,
// and returns only after all columns have either been successfully retrieved or all candidate peers
// have been exhausted.
//
// It returns a map from each root to its successfully retrieved sidecars.
func requestDirectSidecarsFromPeers(
params DataColumnSidecarsParams,
slotByRoot map[[fieldparams.RootLength]byte]primitives.Slot,
requestedIndices map[uint64]bool,
slotsWithCommitments map[primitives.Slot]bool,
storedIndicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool,
incompleteRoots map[[fieldparams.RootLength]byte]bool,
) (map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, error) {
start := time.Now()
// Create a new random source for peer selection.
randomSource := rand.NewGenerator()
// Determine all sidecars each peers are expected to custody.
connectedPeersSlice := params.P2P.Peers().Connected()
connectedPeers := make(map[goPeer.ID]bool, len(connectedPeersSlice))
for _, peer := range connectedPeersSlice {
connectedPeers[peer] = true
}
// Compute missing indices by root, excluding those already in storage.
missingIndicesByRoot := make(map[[fieldparams.RootLength]byte]map[uint64]bool, len(incompleteRoots))
for root := range incompleteRoots {
storedIndices := storedIndicesByRoot[root]
missingIndices := make(map[uint64]bool, len(requestedIndices))
for index := range requestedIndices {
if !storedIndices[index] {
missingIndices[index] = true
}
}
if len(missingIndices) > 0 {
missingIndicesByRoot[root] = missingIndices
}
}
initialMissingCount := computeTotalCount(missingIndicesByRoot)
indicesByRootByPeer, err := computeIndicesByRootByPeer(params.P2P, slotByRoot, missingIndicesByRoot, connectedPeers)
if err != nil {
return nil, errors.Wrap(err, "explore peers")
}
verifiedColumnsByRoot := make(map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn)
for len(missingIndicesByRoot) > 0 && len(indicesByRootByPeer) > 0 {
// Select peers to query the missing sidecars from.
indicesByRootByPeerToQuery, err := selectPeers(params, randomSource, len(missingIndicesByRoot), indicesByRootByPeer)
if err != nil {
return nil, errors.Wrap(err, "select peers")
}
// Remove selected peers from the maps.
for peer := range indicesByRootByPeerToQuery {
delete(connectedPeers, peer)
}
// Fetch the sidecars from the chosen peers.
roDataColumnsByPeer := fetchDataColumnSidecarsFromPeers(params, slotByRoot, slotsWithCommitments, indicesByRootByPeerToQuery)
// Verify the received data column sidecars.
verifiedRoDataColumnSidecars, err := verifyDataColumnSidecarsByPeer(params.P2P, params.NewVerifier, roDataColumnsByPeer)
if err != nil {
return nil, errors.Wrap(err, "verify data columns sidecars by peer")
}
// Remove the verified sidecars from the missing indices map and compute the new verified columns by root.
localVerifiedColumnsByRoot := updateResults(verifiedRoDataColumnSidecars, missingIndicesByRoot)
for root, verifiedRoDataColumns := range localVerifiedColumnsByRoot {
verifiedColumnsByRoot[root] = append(verifiedColumnsByRoot[root], verifiedRoDataColumns...)
}
// Compute indices by root by peers with the updated missing indices and connected peers.
indicesByRootByPeer, err = computeIndicesByRootByPeer(params.P2P, slotByRoot, missingIndicesByRoot, connectedPeers)
if err != nil {
return nil, errors.Wrap(err, "explore peers")
}
}
log.WithFields(logrus.Fields{
"duration": time.Since(start),
"initialMissingCount": initialMissingCount,
"finalMissingCount": computeTotalCount(missingIndicesByRoot),
}).Debug("Requested direct data column sidecars from peers")
return verifiedColumnsByRoot, nil
}
// requestIndirectSidecarsFromPeers requests, for all roots in `missingIndicesbyRootOrig`,
// for all possible peers, taking into account sidecars available in `inputs` and in the storage,
// all possible sidecars until either, for each root:
// - all indices in `indices` are available, or
// - enough sidecars are available to trigger a reconstruction, or
// - all peers are exhausted.
func requestIndirectSidecarsFromPeers(
p DataColumnSidecarsParams,
slotByRoot map[[fieldparams.RootLength]byte]primitives.Slot,
slotsWithCommitments map[primitives.Slot]bool,
storedIndicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool,
alreadyAvailableByRoot map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn,
requestedIndices map[uint64]bool,
roots map[[fieldparams.RootLength]byte]bool,
) (map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, error) {
start := time.Now()
const numberOfColumns = uint64(fieldparams.NumberOfColumns)
minimumColumnCountToReconstruct := peerdas.MinimumColumnCountToReconstruct()
// Create a new random source for peer selection.
randomSource := rand.NewGenerator()
// For each root compute all possible data column sidecar indices excluding
// those already stored or already available.
indicesToRetrieveByRoot := make(map[[fieldparams.RootLength]byte]map[uint64]bool)
for root := range roots {
alreadyAvailableIndices := make(map[uint64]bool, len(alreadyAvailableByRoot[root]))
for _, sidecar := range alreadyAvailableByRoot[root] {
alreadyAvailableIndices[sidecar.Index] = true
}
storedIndices := storedIndicesByRoot[root]
indicesToRetrieve := make(map[uint64]bool, numberOfColumns)
for index := range numberOfColumns {
if !(storedIndices[index] || alreadyAvailableIndices[index]) {
indicesToRetrieve[index] = true
}
}
if len(indicesToRetrieve) > 0 {
indicesToRetrieveByRoot[root] = indicesToRetrieve
}
}
initialToRetrieveRootCount := len(indicesToRetrieveByRoot)
// Determine all sidecars each peers are expected to custody.
connectedPeersSlice := p.P2P.Peers().Connected()
connectedPeers := make(map[goPeer.ID]bool, len(connectedPeersSlice))
for _, peer := range connectedPeersSlice {
connectedPeers[peer] = true
}
// Compute which peers have which of the missing indices.
indicesByRootByPeer, err := computeIndicesByRootByPeer(p.P2P, slotByRoot, indicesToRetrieveByRoot, connectedPeers)
if err != nil {
return nil, errors.Wrap(err, "explore peers")
}
// Already add into results all sidecars present in `alreadyAvailableByRoot`.
result := make(map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn)
for root := range roots {
alreadyAvailable := alreadyAvailableByRoot[root]
result[root] = append(result[root], alreadyAvailable...)
}
for len(indicesToRetrieveByRoot) > 0 && len(indicesByRootByPeer) > 0 {
// Select peers to query the missing sidecars from.
indicesByRootByPeerToQuery, err := selectPeers(p, randomSource, len(indicesToRetrieveByRoot), indicesByRootByPeer)
if err != nil {
return nil, errors.Wrap(err, "select peers")
}
// Remove selected peers from the maps.
for peer := range indicesByRootByPeerToQuery {
delete(connectedPeers, peer)
}
// Fetch the sidecars from the chosen peers.
roDataColumnsByPeer := fetchDataColumnSidecarsFromPeers(p, slotByRoot, slotsWithCommitments, indicesByRootByPeerToQuery)
// Verify the received data column sidecars.
verifiedRoDataColumnSidecars, err := verifyDataColumnSidecarsByPeer(p.P2P, p.NewVerifier, roDataColumnsByPeer)
if err != nil {
return nil, errors.Wrap(err, "verify data columns sidecars by peer")
}
// Add to results all verified sidecars.
localVerifiedColumnsByRoot := updateResults(verifiedRoDataColumnSidecars, indicesToRetrieveByRoot)
for root, verifiedRoDataColumns := range localVerifiedColumnsByRoot {
result[root] = append(result[root], verifiedRoDataColumns...)
}
// Unlabel a root as to retrieve if enough sidecars are retrieved to enable a reconstruction,
// or if all requested sidecars are now available for this root.
for root, indicesToRetrieve := range indicesToRetrieveByRoot {
storedIndices := storedIndicesByRoot[root]
storedCount := uint64(len(storedIndices))
resultCount := uint64(len(result[root]))
if storedCount+resultCount >= minimumColumnCountToReconstruct {
delete(indicesToRetrieveByRoot, root)
continue
}
allRequestedIndicesAvailable := true
for index := range requestedIndices {
if indicesToRetrieve[index] {
// Still need this index.
allRequestedIndicesAvailable = false
break
}
}
if allRequestedIndicesAvailable {
delete(indicesToRetrieveByRoot, root)
}
}
// Compute indices by root by peers with the updated missing indices and connected peers.
indicesByRootByPeer, err = computeIndicesByRootByPeer(p.P2P, slotByRoot, indicesToRetrieveByRoot, connectedPeers)
if err != nil {
return nil, errors.Wrap(err, "explore peers")
}
}
log.WithFields(logrus.Fields{
"duration": time.Since(start),
"initialToRetrieveRootCount": initialToRetrieveRootCount,
"finalToRetrieveRootCount": len(indicesToRetrieveByRoot),
}).Debug("Requested all data column sidecars from peers")
return result, nil
}
// mergeAvailableSidecars retrieves missing data column sidecars by combining
// what is available in storage with the sidecars provided in `alreadyAvailableByRoot`,
// reconstructing them when necessary.
//
// The function works in two modes depending on sidecar availability:
// - If all requested sidecars are already available (no reconstruction needed),
// it simply returns them directly from storage and inputs.
// - If storage + inputs together provide enough sidecars to reconstruct all requested ones,
// it reconstructs and returns the requested sidecars.
//
// If a root cannot yield all requested sidecars, that root is omitted from the result.
//
// Note: It is assumed that no sidecar in `alreadyAvailableByRoot` is already present in storage.
//
// WARNING: This function mutates `roots`, removing any block roots
// for which all requested sidecars were successfully retrieved.
func mergeAvailableSidecars(
storage filesystem.DataColumnStorageReader,
requestedIndices map[uint64]bool,
storedIndicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool,
roots map[[fieldparams.RootLength]byte]bool,
alreadyAvailableByRoot map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn,
) (map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, error) {
minimumColumnsCountToReconstruct := peerdas.MinimumColumnCountToReconstruct()
result := make(map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, len(roots))
for root := range roots {
storedIndices := storedIndicesByRoot[root]
alreadyAvailable := alreadyAvailableByRoot[root]
// Compute already available indices.
alreadyAvailableIndices := make(map[uint64]bool, len(alreadyAvailable))
for _, sidecar := range alreadyAvailable {
alreadyAvailableIndices[sidecar.Index] = true
}
// Check if reconstruction is needed.
isReconstructionNeeded := false
for index := range requestedIndices {
if !(storedIndices[index] || alreadyAvailableIndices[index]) {
isReconstructionNeeded = true
break
}
}
// Check if reconstruction is possible.
storedCount := uint64(len(storedIndices))
alreadyAvailableCount := uint64(len(alreadyAvailableIndices))
isReconstructionPossible := storedCount+alreadyAvailableCount >= minimumColumnsCountToReconstruct
// Skip if the reconstruction is needed and not possible.
if isReconstructionNeeded && !isReconstructionPossible {
continue
}
// Reconstruct if reconstruction is needed and possible.
if isReconstructionNeeded && isReconstructionPossible {
// Load all we have in the store.
stored, err := storage.Get(root, nil)
if err != nil {
return nil, errors.Wrapf(err, "storage get for root %#x", root)
}
allAvailable := make([]blocks.VerifiedRODataColumn, 0, storedCount+alreadyAvailableCount)
allAvailable = append(allAvailable, stored...)
allAvailable = append(allAvailable, alreadyAvailable...)
// Attempt reconstruction.
reconstructedSidecars, err := peerdas.ReconstructDataColumnSidecars(allAvailable)
if err != nil {
return nil, errors.Wrapf(err, "reconstruct data column sidecars for root %#x", root)
}
// Select only sidecars we need.
for _, sidecar := range reconstructedSidecars {
if requestedIndices[sidecar.Index] {
result[root] = append(result[root], sidecar)
}
}
delete(roots, root)
continue
}
// Reconstruction is not needed, simply assemble what is available in storage and already available.
allAvailable, err := assembleAvailableSidecarsForRoot(storage, alreadyAvailableByRoot, root, requestedIndices)
if err != nil {
return nil, errors.Wrap(err, "assemble available sidecars")
}
result[root] = allAvailable
delete(roots, root)
}
return result, nil
}
// assembleAvailableSidecars assembles all sidecars available in storage
// and in `alreadyAvailableByRoot` corresponding to `roots`.
// It also returns all missing indices by root.
func assembleAvailableSidecars(
storage filesystem.DataColumnStorageReader,
requestedIndices map[uint64]bool,
roots map[[fieldparams.RootLength]byte]bool,
alreadyAvailableByRoot map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn,
) (map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, map[[fieldparams.RootLength]byte]map[uint64]bool, error) {
// Assemble results.
result := make(map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn, len(roots))
for root := range roots {
allAvailable, err := assembleAvailableSidecarsForRoot(storage, alreadyAvailableByRoot, root, requestedIndices)
if err != nil {
return nil, nil, errors.Wrap(err, "assemble sidecars for root")
}
if len(allAvailable) > 0 {
result[root] = allAvailable
}
}
// Compute still missing sidecars.
missingByRoot := make(map[[fieldparams.RootLength]byte]map[uint64]bool, len(roots))
for root := range roots {
missing := make(map[uint64]bool, len(requestedIndices))
for index := range requestedIndices {
missing[index] = true
}
allAvailable := result[root]
for _, sidecar := range allAvailable {
delete(missing, sidecar.Index)
}
if len(missing) > 0 {
missingByRoot[root] = missing
}
}
return result, missingByRoot, nil
}
// assembleAvailableSidecarsForRoot assembles all sidecars available in storage
// and in `alreadyAvailableByRoot` corresponding to `root` and `indices`.
func assembleAvailableSidecarsForRoot(
storage filesystem.DataColumnStorageReader,
alreadyAvailableByRoot map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn,
root [fieldparams.RootLength]byte,
indices map[uint64]bool,
) ([]blocks.VerifiedRODataColumn, error) {
stored, err := storage.Get(root, helpers.SortedSliceFromMap(indices))
if err != nil {
return nil, errors.Wrapf(err, "storage get for root %#x", root)
}
alreadyAvailable := alreadyAvailableByRoot[root]
allAvailable := make([]blocks.VerifiedRODataColumn, 0, len(stored)+len(alreadyAvailable))
allAvailable = append(allAvailable, stored...)
allAvailable = append(allAvailable, alreadyAvailable...)
return allAvailable, nil
}
// selectPeers selects peers to query the sidecars.
// It begins by randomly selecting a peer in `origIndicesByRootByPeer` that has enough bandwidth,
// and assigns to it all its available sidecars. Then, it randomly select an other peer, until
// all sidecars in `missingIndicesByRoot` are covered.
func selectPeers(
p DataColumnSidecarsParams,
randomSource *rand.Rand,
count int,
origIndicesByRootByPeer map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool,
) (map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool, error) {
const randomPeerTimeout = 2 * time.Minute
// Select peers to query the missing sidecars from.
indicesByRootByPeer := copyIndicesByRootByPeer(origIndicesByRootByPeer)
internalIndicesByRootByPeer := copyIndicesByRootByPeer(indicesByRootByPeer)
indicesByRootByPeerToQuery := make(map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool)
for len(internalIndicesByRootByPeer) > 0 {
// Randomly select a peer with enough bandwidth.
peer, err := func() (goPeer.ID, error) {
ctx, cancel := context.WithTimeout(p.Ctx, randomPeerTimeout)
defer cancel()
peer, err := randomPeer(ctx, randomSource, p.RateLimiter, count, internalIndicesByRootByPeer)
if err != nil {
return "", errors.Wrap(err, "select random peer")
}
return peer, err
}()
if err != nil {
return nil, err
}
// Query all the sidecars that peer can offer us.
newIndicesByRoot, ok := internalIndicesByRootByPeer[peer]
if !ok {
return nil, errors.Errorf("peer %s not found in internal indices by root by peer map", peer)
}
indicesByRootByPeerToQuery[peer] = newIndicesByRoot
// Remove this peer from the maps to avoid re-selection.
delete(indicesByRootByPeer, peer)
delete(internalIndicesByRootByPeer, peer)
// Delete the corresponding sidecars from other peers in the internal map
// to avoid re-selection during this iteration.
for peer, indicesByRoot := range internalIndicesByRootByPeer {
for root, indices := range indicesByRoot {
newIndices := newIndicesByRoot[root]
for index := range newIndices {
delete(indices, index)
}
if len(indices) == 0 {
delete(indicesByRoot, root)
}
}
if len(indicesByRoot) == 0 {
delete(internalIndicesByRootByPeer, peer)
}
}
}
return indicesByRootByPeerToQuery, nil
}
// updateResults updates the missing indices and verified sidecars maps based on the newly verified sidecars.
// WARNING: This function alters `missingIndicesByRoot` by removing verified sidecars.
// After running this function, the user can check the content of the (modified) `missingIndicesByRoot` map
// to check if some sidecars are still missing.
func updateResults(
verifiedSidecars []blocks.VerifiedRODataColumn,
missingIndicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool,
) map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn {
verifiedSidecarsByRoot := make(map[[fieldparams.RootLength]byte][]blocks.VerifiedRODataColumn)
for _, verifiedSidecar := range verifiedSidecars {
blockRoot := verifiedSidecar.BlockRoot()
index := verifiedSidecar.Index
// Add to the result map grouped by block root
verifiedSidecarsByRoot[blockRoot] = append(verifiedSidecarsByRoot[blockRoot], verifiedSidecar)
if indices, ok := missingIndicesByRoot[blockRoot]; ok {
delete(indices, index)
if len(indices) == 0 {
delete(missingIndicesByRoot, blockRoot)
}
}
}
return verifiedSidecarsByRoot
}
// fetchDataColumnSidecarsFromPeers retrieves data column sidecars from peers.
func fetchDataColumnSidecarsFromPeers(
params DataColumnSidecarsParams,
slotByRoot map[[fieldparams.RootLength]byte]primitives.Slot,
slotsWithCommitments map[primitives.Slot]bool,
indicesByRootByPeer map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool,
) map[goPeer.ID][]blocks.RODataColumn {
var (
wg sync.WaitGroup
mut sync.Mutex
)
roDataColumnsByPeer := make(map[goPeer.ID][]blocks.RODataColumn)
wg.Add(len(indicesByRootByPeer))
for peerID, indicesByRoot := range indicesByRootByPeer {
go func(peerID goPeer.ID, indicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool) {
defer wg.Done()
requestedCount := 0
for _, indices := range indicesByRoot {
requestedCount += len(indices)
}
log := log.WithFields(logrus.Fields{
"peerID": peerID,
"agent": agentString(peerID, params.P2P.Host()),
"blockCount": len(indicesByRoot),
"totalRequestedCount": requestedCount,
})
roDataColumns, err := sendDataColumnSidecarsRequest(params, slotByRoot, slotsWithCommitments, peerID, indicesByRoot)
if err != nil {
log.WithError(err).Debug("Failed to send data column sidecars request")
return
}
mut.Lock()
defer mut.Unlock()
roDataColumnsByPeer[peerID] = roDataColumns
}(peerID, indicesByRoot)
}
wg.Wait()
return roDataColumnsByPeer
}
func sendDataColumnSidecarsRequest(
params DataColumnSidecarsParams,
slotByRoot map[[fieldparams.RootLength]byte]primitives.Slot,
slotsWithCommitments map[primitives.Slot]bool,
peerID goPeer.ID,
indicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool,
) ([]blocks.RODataColumn, error) {
const batchSize = 32
rootCount := int64(len(indicesByRoot))
requestedSidecarsCount := 0
for _, indices := range indicesByRoot {
requestedSidecarsCount += len(indices)
}
log := log.WithFields(logrus.Fields{
"peerID": peerID,
"agent": agentString(peerID, params.P2P.Host()),
"requestedSidecars": requestedSidecarsCount,
})
// Try to build a by range byRangeRequest first.
byRangeRequests, err := buildByRangeRequests(slotByRoot, slotsWithCommitments, indicesByRoot, batchSize)
if err != nil {
return nil, errors.Wrap(err, "craft by range request")
}
// If we have a valid by range request, send it.
if len(byRangeRequests) > 0 {
count := 0
for _, indices := range indicesByRoot {
count += len(indices)
}
start := time.Now()
roDataColumns := make([]blocks.RODataColumn, 0, count)
for _, request := range byRangeRequests {
if params.RateLimiter != nil {
params.RateLimiter.Add(peerID.String(), rootCount)
}
localRoDataColumns, err := SendDataColumnSidecarsByRangeRequest(params, peerID, request)
if err != nil {
return nil, errors.Wrapf(err, "send data column sidecars by range request to peer %s", peerID)
}
roDataColumns = append(roDataColumns, localRoDataColumns...)
}
if logrus.GetLevel() >= logrus.DebugLevel {
prettyByRangeRequests := make([]map[string]any, 0, len(byRangeRequests))
for _, request := range byRangeRequests {
prettyRequest := map[string]any{
"startSlot": request.StartSlot,
"count": request.Count,
"columns": helpers.PrettySlice(request.Columns),
}
prettyByRangeRequests = append(prettyByRangeRequests, prettyRequest)
}
log.WithFields(logrus.Fields{
"respondedSidecars": len(roDataColumns),
"requestCount": len(byRangeRequests),
"type": "byRange",
"duration": time.Since(start),
"requests": prettyByRangeRequests,
}).Debug("Received data column sidecars")
}
return roDataColumns, nil
}
// Build identifiers for the by root request.
byRootRequest := buildByRootRequest(indicesByRoot)
// Send the by root request.
start := time.Now()
if params.RateLimiter != nil {
params.RateLimiter.Add(peerID.String(), rootCount)
}
roDataColumns, err := SendDataColumnSidecarsByRootRequest(params, peerID, byRootRequest)
if err != nil {
return nil, errors.Wrapf(err, "send data column sidecars by root request to peer %s", peerID)
}
log.WithFields(logrus.Fields{
"respondedSidecars": len(roDataColumns),
"requests": 1,
"type": "byRoot",
"duration": time.Since(start),
}).Debug("Received data column sidecars")
return roDataColumns, nil
}
// buildByRangeRequests constructs a by range request from the given indices,
// only if the indices are the same all blocks and if the blocks are contiguous.
// (Missing blocks or blocks without commitments do count as contiguous)
// If one of this condition is not met, returns nil.
func buildByRangeRequests(
slotByRoot map[[fieldparams.RootLength]byte]primitives.Slot,
slotsWithCommitments map[primitives.Slot]bool,
indicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool,
batchSize uint64,
) ([]*ethpb.DataColumnSidecarsByRangeRequest, error) {
if len(indicesByRoot) == 0 {
return nil, nil
}
var reference map[uint64]bool
slots := make([]primitives.Slot, 0, len(slotByRoot))
for root, indices := range indicesByRoot {
if reference == nil {
reference = indices
}
if !compareIndices(reference, indices) {
return nil, nil
}
slot, ok := slotByRoot[root]
if !ok {
return nil, errors.Errorf("slot not found for block root %#x", root)
}
slots = append(slots, slot)
}
slices.Sort(slots)
for i := 1; i < len(slots); i++ {
previous, current := slots[i-1], slots[i]
if current == previous+1 {
continue
}
for j := previous + 1; j < current; j++ {
if slotsWithCommitments[j] {
return nil, nil
}
}
}
columns := helpers.SortedSliceFromMap(reference)
startSlot, endSlot := slots[0], slots[len(slots)-1]
totalCount := uint64(endSlot - startSlot + 1)
requests := make([]*ethpb.DataColumnSidecarsByRangeRequest, 0, totalCount/batchSize)
for start := startSlot; start <= endSlot; start += primitives.Slot(batchSize) {
end := min(start+primitives.Slot(batchSize)-1, endSlot)
request := &ethpb.DataColumnSidecarsByRangeRequest{
StartSlot: start,
Count: uint64(end - start + 1),
Columns: columns,
}
requests = append(requests, request)
}
return requests, nil
}
// buildByRootRequest constructs a by root request from the given indices.
func buildByRootRequest(indicesByRoot map[[fieldparams.RootLength]byte]map[uint64]bool) p2ptypes.DataColumnsByRootIdentifiers {
identifiers := make(p2ptypes.DataColumnsByRootIdentifiers, 0, len(indicesByRoot))
for root, indices := range indicesByRoot {
identifier := &ethpb.DataColumnsByRootIdentifier{
BlockRoot: root[:],
Columns: helpers.SortedSliceFromMap(indices),
}
identifiers = append(identifiers, identifier)
}
// Sort identifiers to have a deterministic output.
slices.SortFunc(identifiers, func(left, right *ethpb.DataColumnsByRootIdentifier) int {
if cmp := bytes.Compare(left.BlockRoot, right.BlockRoot); cmp != 0 {
return cmp
}
return slices.Compare(left.Columns, right.Columns)
})
return identifiers
}
// verifyDataColumnSidecarsByPeer verifies the received data column sidecars.
// If at least one sidecar from a peer is invalid, the peer is downscored and
// all its sidecars are rejected. (Sidecars from other peers are still accepted.)
func verifyDataColumnSidecarsByPeer(
p2p prysmP2P.P2P,
newVerifier verification.NewDataColumnsVerifier,
roDataColumnsByPeer map[goPeer.ID][]blocks.RODataColumn,
) ([]blocks.VerifiedRODataColumn, error) {
// First optimistically verify all received data columns in a single batch.
count := 0
for _, columns := range roDataColumnsByPeer {
count += len(columns)
}
roDataColumnSidecars := make([]blocks.RODataColumn, 0, count)
for _, columns := range roDataColumnsByPeer {
roDataColumnSidecars = append(roDataColumnSidecars, columns...)
}
verifiedRoDataColumnSidecars, err := verifyByRootDataColumnSidecars(newVerifier, roDataColumnSidecars)
if err == nil {
// This is the happy path where all sidecars are verified.
return verifiedRoDataColumnSidecars, nil
}
// An error occurred during verification, which means that at least one sidecar is invalid.
// Reverify peer by peer to identify faulty peer(s), reject all its sidecars, and downscore it.
verifiedRoDataColumnSidecars = make([]blocks.VerifiedRODataColumn, 0, count)
for peer, columns := range roDataColumnsByPeer {
peerVerifiedRoDataColumnSidecars, err := verifyByRootDataColumnSidecars(newVerifier, columns)
if err != nil {
// This peer has invalid sidecars.
log := log.WithError(err).WithField("peerID", peer)
newScore := p2p.Peers().Scorers().BadResponsesScorer().Increment(peer)
log.Warning("Peer returned invalid data column sidecars")
log.WithFields(logrus.Fields{"reason": "invalidDataColumnSidecars", "newScore": newScore}).Debug("Downscore peer")
}
verifiedRoDataColumnSidecars = append(verifiedRoDataColumnSidecars, peerVerifiedRoDataColumnSidecars...)
}
return verifiedRoDataColumnSidecars, nil
}
// verifyByRootDataColumnSidecars verifies the provided read-only data columns against the
// requirements for data column sidecars received via the by root request.
func verifyByRootDataColumnSidecars(newVerifier verification.NewDataColumnsVerifier, roDataColumns []blocks.RODataColumn) ([]blocks.VerifiedRODataColumn, error) {
verifier := newVerifier(roDataColumns, verification.ByRootRequestDataColumnSidecarRequirements)
if err := verifier.ValidFields(); err != nil {
return nil, errors.Wrap(err, "valid fields")
}
if err := verifier.SidecarInclusionProven(); err != nil {
return nil, errors.Wrap(err, "sidecar inclusion proven")
}
if err := verifier.SidecarKzgProofVerified(); err != nil {
return nil, errors.Wrap(err, "sidecar KZG proof verified")
}
verifiedRoDataColumns, err := verifier.VerifiedRODataColumns()
if err != nil {
return nil, errors.Wrap(err, "verified RO data columns - should never happen")
}
return verifiedRoDataColumns, nil
}
// computeIndicesByRootByPeer returns a peers->root->indices map only for
// root and indices given in `indicesByBlockRoot`. It also only selects peers
// for a given root only if its head state is higher than the block slot.
func computeIndicesByRootByPeer(
p2p prysmP2P.P2P,
slotByBlockRoot map[[fieldparams.RootLength]byte]primitives.Slot,
indicesByBlockRoot map[[fieldparams.RootLength]byte]map[uint64]bool,
peers map[goPeer.ID]bool,
) (map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool, error) {
slotsPerEpoch := params.BeaconConfig().SlotsPerEpoch
// First, compute custody columns for all peers
peersByIndex := make(map[uint64]map[goPeer.ID]bool)
headSlotByPeer := make(map[goPeer.ID]primitives.Slot)
for peer := range peers {
log := log.WithField("peerID", peer)
// Computes the custody columns for each peer
nodeID, err := prysmP2P.ConvertPeerIDToNodeID(peer)
if err != nil {
log.WithError(err).Debug("Failed to convert peer ID to node ID")
continue
}
custodyGroupCount := p2p.CustodyGroupCountFromPeer(peer)
dasInfo, _, err := peerdas.Info(nodeID, custodyGroupCount)
if err != nil {
log.WithError(err).Debug("Failed to get peer DAS info")
continue
}
for column := range dasInfo.CustodyColumns {
if _, exists := peersByIndex[column]; !exists {
peersByIndex[column] = make(map[goPeer.ID]bool)
}
peersByIndex[column][peer] = true
}
// Compute the head slot for each peer
peerChainState, err := p2p.Peers().ChainState(peer)
if err != nil {
log.WithError(err).Debug("Failed to get peer chain state")
continue
}
if peerChainState == nil {
log.Debug("Peer chain state is nil")
continue
}
// Our view of the head slot of a peer is not updated in real time.
// We add an epoch to take into account the fact the real head slot of the peer
// is higher than our view of it.
headSlotByPeer[peer] = peerChainState.HeadSlot + slotsPerEpoch
}
// For each block root and its indices, find suitable peers
indicesByRootByPeer := make(map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool)
for blockRoot, indices := range indicesByBlockRoot {
blockSlot, ok := slotByBlockRoot[blockRoot]
if !ok {
return nil, errors.Errorf("slot not found for block root %#x", blockRoot)
}
for index := range indices {
peers := peersByIndex[index]
for peer := range peers {
peerHeadSlot, ok := headSlotByPeer[peer]
if !ok {
return nil, errors.Errorf("head slot not found for peer %s", peer)
}
if peerHeadSlot < blockSlot {
continue
}
// Build peers->root->indices map
if _, exists := indicesByRootByPeer[peer]; !exists {
indicesByRootByPeer[peer] = make(map[[fieldparams.RootLength]byte]map[uint64]bool)
}
if _, exists := indicesByRootByPeer[peer][blockRoot]; !exists {
indicesByRootByPeer[peer][blockRoot] = make(map[uint64]bool)
}
indicesByRootByPeer[peer][blockRoot][index] = true
}
}
}
return indicesByRootByPeer, nil
}
// randomPeer selects a random peer. If no peers has enough bandwidth, it will wait and retry.
// Returns the selected peer ID and any error.
func randomPeer(
ctx context.Context,
randomSource *rand.Rand,
rateLimiter *leakybucket.Collector,
count int,
indicesByRootByPeer map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool,
) (goPeer.ID, error) {
const waitPeriod = 5 * time.Second
peerCount := len(indicesByRootByPeer)
if peerCount == 0 {
return "", errors.New("no peers available")
}
for ctx.Err() == nil {
nonRateLimitedPeers := make([]goPeer.ID, 0, len(indicesByRootByPeer))
for peer := range indicesByRootByPeer {
if rateLimiter == nil || rateLimiter.Remaining(peer.String()) >= int64(count) {
nonRateLimitedPeers = append(nonRateLimitedPeers, peer)
}
}
if len(nonRateLimitedPeers) > 0 {
slices.Sort(nonRateLimitedPeers)
randomIndex := randomSource.Intn(len(nonRateLimitedPeers))
return nonRateLimitedPeers[randomIndex], nil
}
log.WithFields(logrus.Fields{
"peerCount": peerCount,
"delay": waitPeriod,
}).Debug("Waiting for a peer with enough bandwidth for data column sidecars")
select {
case <-time.After(waitPeriod):
case <-ctx.Done():
}
}
return "", ctx.Err()
}
// copyIndicesByRootByPeer creates a deep copy of the given nested map.
// Returns a new map with the same structure and contents.
func copyIndicesByRootByPeer(original map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool) map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool {
copied := make(map[goPeer.ID]map[[fieldparams.RootLength]byte]map[uint64]bool, len(original))
for peer, indicesByRoot := range original {
copied[peer] = copyIndicesByRoot(indicesByRoot)
}
return copied
}
// copyIndicesByRoot creates a deep copy of the given nested map.
// Returns a new map with the same structure and contents.
func copyIndicesByRoot(original map[[fieldparams.RootLength]byte]map[uint64]bool) map[[fieldparams.RootLength]byte]map[uint64]bool {
copied := make(map[[fieldparams.RootLength]byte]map[uint64]bool, len(original))
for root, indexMap := range original {
copied[root] = make(map[uint64]bool, len(indexMap))
maps.Copy(copied[root], indexMap)
}
return copied
}
// compareIndices compares two map[uint64]bool and returns true if they are equal.
func compareIndices(left, right map[uint64]bool) bool {
if len(left) != len(right) {
return false
}
for key, leftValue := range left {
rightValue, exists := right[key]
if !exists || leftValue != rightValue {
return false
}
}
return true
}
// computeTotalCount calculates the total count of indices across all roots.
func computeTotalCount(input map[[fieldparams.RootLength]byte]map[uint64]bool) int {
totalCount := 0
for _, indices := range input {
totalCount += len(indices)
}
return totalCount
}