Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Implement PeerDAS #14129

Draft
wants to merge 81 commits into
base: develop
Choose a base branch
from
Draft

Implement PeerDAS #14129

wants to merge 81 commits into from

Conversation

nalepae
Copy link
Contributor

@nalepae nalepae commented Jun 21, 2024

Remaining tasks:

  • Initial sync: Allow syncing for non super-nodes.
  • Initial sync: Backfill data columns when starting from a checkpoint.
  • Initial sync: Do not request more than 64 data columns: Request 64 data columns then reconstruct the rest.
  • Implement validator custody.
  • ⚠️ Integrate the output of sampling in the forkchoice ⚠️
  • Check why we cannot run initial sync when starting a node during epoch 0, in slot [1, 31]. (Starting a node since genesis works, starting a node from epoch 1 and later works.) Note: This issue may be already here with blobs (without data columns.)
  • Indroduce a nasty peer flag that holds AND/OR (both are important) alterate data columns on RPC responses.
  • Add unit tests where they are not yet added.
  • Replace the cKZG library by the goKZGone (when available) - In progress.
  • Implement Add MetadataV3 with custody_subnet_count ethereum/consensus-specs#3821.
  • ⚠️ Implement number of blobs as a CL parameter. ⚠️
  • Decouple network subnets from Core methods as referenced.
  • Data columns verification: Implement it once, and use it everywhere.
  • ⚠️ Implement /eth/v1/beacon/blob_sidecars/{block_id}
  • Update to latest config values in the spec. ⚠️ In progress here
  • Rewrite sampling tests to allow for a subnet count larger than 32, this is currently hardcoded for it.
  • Sampling: Find more peers if a column has no corresponding peers. (Can do like in internalBroadcastDataColumn)
  • Add PeerDAS specific metrics
  • (Not especially peerDAS related): Make consistent DEBUG p2p: Peer connected activePeers=2 log with INFO p2p: Connected peers log.
  • Clarify usage of --minimum-peers-per-subnet=<n>
  • Should we downscore peer on sampling? (Maybe the peer does not see the block)
  • Should we stop using a peer when failed on sampling? (To avoid DDoS)
  • A node should always serve data columns it has, even if it does not custody it. (Particularly, when proposing a block.)
  • Check metadataV3 interop with other clients if before EIP-7594
  • Refactor code where GetValidCustodyPeers is used. ==> We need to fetch data from peers that are not super nodes.
  • After a fork, for some reason, data columns by root requests are done more than once. (It may be normal if done to multiple peers, but we have to check.)
  • prysmctl: Implement byRoot and byRange data columns requests
  • Fix E2E tests
  • Using subnet sampling only: Should, when using data columns by root requests (and also by range, during initial sync), the node request only custody columns, or custody columns + extra subnet sampling columns? If requesting only custody columns, then a strange situation arises: When receiving a block with a extra subnet sampling column missing, then the node will ignore the block. A few sec after, when requesting the blocks and the associated columns, the node won't need the subnet sampling extra columns to import the block.
  • Find why sometimes ENR records are not correctly updated via discovery.
  • Find why, with Grandine in the mix, discv5 iterator.Next is soooo slow.
  • Test with only one bootnode (and not all nodes)
  • Initial sync: Save data columns? (Are blobs saved during initial sync?)
  • peersWithDataColumn: Select only peers capable to serve up to the target slot. Previously, we used f.pep.Peers().ChainState(peer).HeadSlot, but this info is only updated twice per epoch via status P2P messages. ==> Find a good way to do it. (Maybe send status message just before checking?)
  • Initial sync: How to handle orphaned blocks (Not to be stuck)?
  • By range (and by root) requests: Run verifications for finalized blocks?
  • ByRoot requests: choose all peers like on ByRange requests, not only super nodes.

nalepae and others added 11 commits November 27, 2024 10:11
* Add Support For Discovery Of Column Subnets

* Lint for SubnetsPerNode

* Manu's Review

* Change to a better name
* Add Data Column Subscriber

* Add Data Column Vaidator

* Wire all Handlers In

* Fix Build

* Fix Test

* Fix IP in Test

* Fix IP in Test
* Add RPC Handler

* Add Column Requests

* Update beacon-chain/db/filesystem/blob.go

Co-authored-by: Manu NALEPA <[email protected]>

* Update beacon-chain/p2p/rpc_topic_mappings.go

Co-authored-by: Manu NALEPA <[email protected]>

* Manu's Review

* Manu's Review

* Interface Fixes

* mock manager

---------

Co-authored-by: Manu NALEPA <[email protected]>
* Bump `c-kzg-4844` lib to the `das` branch.

* Implement `MerkleProofKZGCommitments`.

* Implement `das-core.md`.

* Use `peerdas.CustodyColumnSubnets` and `peerdas.CustodyColumns`.

* `CustodyColumnSubnets`: Include `i` in the for loop.

* Remove `computeSubscribedColumnSubnet`.

* Remove `peerdas.CustodyColumns` out of the for loop.
* Remove capital letter from error messages.

* `[4]byte` => `[fieldparams.VersionLength]byte`.

* Prometheus: Remove extra `committee`.

They are probably due to a bad copy/paste.

Note: The name of the probe itself is remaining,
to ensure backward compatibility.

* Implement Proposer RPC for data columns.

* Fix TestProposer_ProposeBlock_OK test.

* Remove default peerDAS activation.

* `validateDataColumn`: Workaround to return a `VerifiedRODataColumn`
* Add new DA check

* Exit early in the event no commitments exist.

* Gazelle

* Fix Mock Broadcaster

* Fix Test Setup

* Update beacon-chain/blockchain/process_block.go

Co-authored-by: Manu NALEPA <[email protected]>

* Manu's Review

* Fix Build

---------

Co-authored-by: Manu NALEPA <[email protected]>
* Update `consensus_spec_version` to `v1.5.0-alpha.1`.

* `CustodyColumns`: Fix and implement spec tests.

* Make deepsource happy.

* `^uint64(0)` => `math.MaxUint64`.

* Fix `TestLoadConfigFile` test.
nisdas and others added 29 commits November 27, 2024 10:24
* Persist All Changes

* Fix All Tests

* Fix Build

* Fix Build

* Fix Build

* Fix Test Again

* Add missing verification

* Add Test Cases for Data Column Validation

* Fix comments for methods

* Fix comments for methods

* Fix Test

* Manu's Review
)

* `parseIndices`: `O(n**2)` ==> `O(n)`.

* PeerDAS: Implement `/eth/v1/beacon/blob_sidecars/{block_id}`.

* Update beacon-chain/core/peerdas/helpers.go

Co-authored-by: Sammy Rosso <[email protected]>

* Rename some functions.

* `Blobs`: Fix empty slice.

* `recoverCellsAndProofs` --> Move function in `beacon-chain/core/peerdas`.

* peerDAS helpers: Add missing tests.

* Implement `CustodyColumnCount`.

* `RecoverCellsAndProofs`: Remove useless argument `columnsCount`.

* Tests: Add cleanups.

* `blobsFromStoredDataColumns`: Reconstruct if needed.

* Make deepsource happy.

* Beacon API: Use provided indices.

* Make deepsource happier.

---------

Co-authored-by: Sammy Rosso <[email protected]>
* Update go.yml

* Disable mnd

* Update .golangci.yml

* Update go.yml

* Update go.yml

* Update .golangci.yml

* Update go.yml

* Fix Lint Issues

* Remove comment

* Update .golangci.yml
* Update values

* Update Spec To v1.5.0-alpha.5

* Fix Discovery Tests

* Hardcode Subnet Count For Tests

* Fix All Initial Sync Tests

* Gazelle

* Less Chaotic Service Initialization

* Gazelle
* Use Data Column Validation Everywhere

* Fix Build

* Fix Lint

* Fix Clock Synchronizer

* Fix Panic
* Add Changes for Uint8 Csc

* Fix Build

* Fix Build for Sync

* Fix Discovery Test
* Fix Various Bugs in PeerDAS

* Remove Log

* Remove useless copy var.

---------

Co-authored-by: Manu NALEPA <[email protected]>
…or reconstruction (#14397)

* `broadcastAndReceiveDataColumns`: Use real `sidecar.ColumnIndex` instead of position in the slice.

And improve logging as well.

* `isDataColumnsAvailable`: Improve logging.

* `validateDataColumn`: Print `Accepted data column sidecar gossip` really at the end.

* Subscriber: Improve logging.

* `sendAndSaveDataColumnSidecars`: Use common used function for logging.

* `dataColumnSidecarByRootRPCHandler`: Logging - Pring `all` instead of all the columns for a super node.

* Verification: Improve logging.

* `DataColumnsWithholdCount`: Set as `uint64` instead `int`.

* `DataColumnFields`: Improve logging.

* Logging: Remove now useless private `columnFields`function.

* Avoid useless goroutines blocking for reconstruction.

* Update beacon-chain/sync/subscriber.go

Co-authored-by: Nishant Das <[email protected]>

* Address Nishant's comment.

* Improve logging.

---------

Co-authored-by: Nishant Das <[email protected]>
* Add Data Column Metrics

* Shift it All To Peerdas Package
* `pingPeers`: Add log with new ENR when modified.

* `p2p Start`: Use idiomatic go error syntax.

* P2P `start`: Fix error message.

* Use not bootnodes at all if the `--chain-config-file` flag is used and no `--bootstrap-node` flag is used.

Before this commit, if the  `--chain-config-file` flag is used and no `--bootstrap-node` flag is used, then bootnodes are (incorrectly) defaulted on `mainnet` ones.

* `validPeersExist`: Centralize logs.

* `AddConnectionHandler`: Improve logging.

"Peer connected" does not really reflect the fact that a new peer is actually connected. --> "New peer connection" is more clear.

Also, instead of writing `0`, `1`or `2` for direction, now it's writted "Unknown", "Inbound", "Outbound".

* Logging: Add 2 decimals for timestamt in text and JSON logs.

* Improve "no valid peers" logging.

* Improve "Some columns have no peers responsible for custody" logging.

* `pubsubSubscriptionRequestLimit`: Increase to be consistent with data columns.

* `sendPingRequest`: Improve logging.

* `FindPeersWithSubnet`: Regularly recheck in our current set of peers if we have enough peers for this topic.

Before this commit, new peers HAD to be found, even if current peers are eventually acceptable.
For very small network, it used to lead to infinite search.

* `subscribeDynamicWithSyncSubnets`: Use exactly the same subscription function initially and every slot.

* Make deepsource happier.

* Nishant's commend: Change peer disconnected log.

* NIshant's comment: Change `Too many incoming subscription` log from error to debug.

* `FindPeersWithSubnet`: Address Nishant's comment.

* `batchSize`: Address Nishant's comment.

* `pingPeers` ==> `pingPeersAndLogEnr`.

* Update beacon-chain/sync/subscriber.go

Co-authored-by: Nishant Das <[email protected]>

---------

Co-authored-by: Nishant Das <[email protected]>
* `CustodyCountFromRemotePeer`: Set happy path in the outer scope.

* `FindPeersWithSubnet`: Improve logging.

* `listenForNewNodes`: Avoid infinite loop in a small subnet.

* Address Nishant's comment.

* FIx Nishant's comment.
* `sendBatchRootRequest`: Refactor and add comments.

* `sendBatchRootRequest`: Do send requests to peers that custodies a superset of our columns.

Before this commit, we sent "data columns by root requests" for data columns peers do not custody.

* Data columns: Use subnet sampling only.

(Instead of peer sampling.)

aaa

* `areDataColumnsAvailable`: Improve logs.

* `GetBeaconBlock`: Improve logs.

Rationale: A `begin` log should always be followed by a `success` log or a `failure` log.
* `validateDataColumn`: Refactor logging.

* `dataColumnSidecarByRootRPCHandler`: Improve logging.

* `isDataAvailable`: Improve logging.

* Add hidden debug flag: `--data-columns-reject-slot-multiple`.

* Add more logs about peer disconnection.

* `validPeersExist` --> `enoughPeersAreConnected`

* `beaconBlocksByRangeRPCHandler`: Add remote Peer ID in logs.

* Stop calling twice `writeErrorResponseToStream` in case of rate limit.
* `scheduleReconstructedDataColumnsBroadcast`: Really minor refactor.

* `receivedDataColumnsFromRootLock` -> `dataColumnsFromRootLock`

* `reconstructDataColumns`: Stop looking into the DB to know if we have some columns.

Before this commit:
Each time we receive a column, we look into the filesystem for all columns we store.
==> For 128 columns, it looks for 1 + 2 + 3 + ... + 128 = 128(128+1)/2 = 8256 files look.

Also, as soon as a column is saved into the file system, then if, right after, we
look at the filesystem again, we assume the column will be available (strict consistency).
It happens not to be always true.

==> Sometimes, we can reconstruct and reseed columns more than once, because of this lack of filesystem strict consistency.

After this commit:
We use a (strictly consistent) cache to determine if we received a column or not.
==> No more consistency issue, and less stress for the filesystem.

* `dataColumnSidecarByRootRPCHandler`: Improve logging.

Before this commit, logged values assumed that all requested columns correspond to
the same block root, which is not always the case.

After this commit, we know which columns are requested for which root.

* Add a log when broadcasting a data column.

This is useful to debug "lost data columns" in devnet.

* Address Nishant's comment
* `columnErrBuilder`: Uses `Wrap` instead of `Join`.

Reason: `Join` makes a carriage return. The log is quite unreadable.

* `validateDataColumn`: Improve log.

* `areDataColumnsAvailable`: Improve log.

* `SendDataColumnSidecarByRoot` ==> `SendDataColumnSidecarsByRootRequest`.

* `handleDA`: Refactor error message.

* `sendRecentBeaconBlocksRequest` ==> `sendBeaconBlocksRequest`.

Reason: There is no notion at all of "recent" in the function.

If the caller decides to call this function only with "recent" blocks, that's fine.
However, the function itself will know nothing about the "recentness" of these blocks.

* `sendBatchRootRequest`: Improve comments.

* `sendBeaconBlocksRequest`: Avoid `else` usage and use map of bool instead of `struct{}`.

* `wrapAndReportValidation`: Remove `agent` from log.

Reason: This prevent the log to hold on one line, and it is not really useful to debug.

* `validateAggregateAndProof`: Add comments.

* `GetValidCustodyPeers`: Fix typo.

* `GetValidCustodyPeers` ==> `DataColumnsAdmissibleCustodyPeers`.

* `CustodyHandler` ==> `DataColumnsHandler`.

* `CustodyCountFromRemotePeer` ==> `DataColumnsCustodyCountFromRemotePeer`.

* Implement `DataColumnsAdmissibleSubnetSamplingPeers`.

* Use `SubnetSamplingSize` instead of `CustodySubnetCount` where needed.

* Revert "`wrapAndReportValidation`: Remove `agent` from log."

This reverts commit 55db351.
* `retrieveMissingDataColumnsFromPeers`: Improve logging.

* `dataColumnSidecarByRootRPCHandler`: Stop decreasing peer's score if asking for a column we do not custody.

* `dataColumnSidecarByRootRPCHandler`: If a data column is unavailable, stop waiting for it.

This behaviour was useful for peer sampling.
Now, just return the data column if we store it.
If we don't, skip.

* Dirty code comment.

* `retrieveMissingDataColumnsFromPeers`: Improve logs.

* `SendDataColumnsByRangeRequest`: Improve logs.

* `dataColumnSidecarsByRangeRPCHandler`: Improve logs.
* `BestFinalized`: Refactor (no functional change).

* `BestNonFinalized`: Refactor (no functional change).

* `beaconBlocksByRangeRPCHandler`: Remove useless log.

The same is already printed at the start of the function.

* `calculateHeadAndTargetEpochs`: Avoid `else`.

* `ConvertPeerIDToNodeID`: Improve error.

* Stop printing noisy "peer should be banned" logs.

* Initial sync: Request data columns from peers which:
- custody a superset of columns we need, and
- have a head slot >= our target slot.

* `requestDataColumnsFromPeers`: Shuffle peers before requesting.

Before this commit, we always requests peers in the same order,
until one responds something.
Without shuffling, we always requests data columns from the same
peer.

* `requestDataColumnsFromPeers`: If error from a peer, just log the error and skip the peer.

* Improve logging.

* Fix tests.
* Improve logging.

* `retrieveMissingDataColumnsFromPeers`: Limit to `512` items per request.

* `retrieveMissingDataColumnsFromPeers`: Allow `nil` peers.

Before this commit:
If, when this funcion is called, we are not yet connected to enough peers, then `peers` will be possibly not be satisfaying,
and, if new peers are connected, we will never see them.

After this commit:
If `peers` is `nil`, then we regularly check for all connected peers.
If `peers` is not `nil`, then we use them.
* `retrieveMissingDataColumnsFromPeers`: Search only for needed peers.

* Improve logging.
* Fix Commitments Check

* `highestFinalizedEpoch`: Refactor (no functional change).

* `retrieveMissingDataColumnsFromPeers`: Fix logs.

* `VerifyDataColumnSidecarKZGProofs`: Optimise with capacity.

* Save data columns when initial syncing.

* `dataColumnSidecarsByRangeRPCHandler`: Add logs when a request enters.

* Improve logging.

* Improve logging.

* `peersWithDataColumns: Do not filter any more on peer head slot.

* Fix Nishant's comment.

---------

Co-authored-by: Manu NALEPA <[email protected]>
…llow syncing from full nodes. (#14532)

* `validateDataColumnsByRange`: `current` ==> `currentSlot`.

* `validateRequest`: Extract `remotePeer` variable.

* `dataColumnSidecarsByRangeRPCHandler`: Small non functional refactor.

* `streamDataColumnBatch`: Fix major bug.

Before this commit, the node was unable to respond with a data column index higher than the count of stored data columns.
For example, if there is 8 data columns stored for a given block, the node was
able to respond for data columns indices 1, 3, and 5, but not for 10, 16 or 127.

The issue was visible only for full nodes, since super nodes always store 128 data columns.

* Initial sync: Fetch data columns from all peers.
(Not only from supernodes.)

* Nishant's comment: Fix `lastSlot` and `endSlot` duplication.

* Address Nishant's comment.
* `ColumnAlignsWithBlock`: Split lines.

* Data columns verifications: Batch

* Remove completely `DataColumnBatchVerifier`.

Only `DataColumnsVerifier` (with `s`) on columns remains.
It is the responsability of the function which receive the data column
(either by gossip, by range request or by root request) to verify the
data column wrt. corresponding checks.

* Fix Nishant's comment.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Blocked Blocked by research or external factors peerDAS
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants