hotshot_query_service/data_source/

update.rs

// Copyright (c) 2022 Espresso Systems (espressosys.com)
// This file is part of the HotShot Query Service library.
//
// This program is free software: you can redistribute it and/or modify it under the terms of the GNU
// General Public License as published by the Free Software Foundation, either version 3 of the
// License, or (at your option) any later version.
// This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without
// even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
// General Public License for more details.
// You should have received a copy of the GNU General Public License along with this program. If not,
// see <https://www.gnu.org/licenses/>.

//! A generic algorithm for updating a HotShot Query Service data source with new data.
use std::iter::once;

use anyhow::{Context, ensure};
use async_trait::async_trait;
use committable::Committable;
use futures::future::Future;
use hotshot::types::EventType;
use hotshot_types::{
    data::{
        Leaf2, VidCommitment, VidCommon, VidDisperseShare, VidShare, ViewNumber,
        ns_table::parse_ns_table,
    },
    event::LeafInfo,
    new_protocol::CoordinatorEvent,
    traits::{
        block_contents::{BlockHeader, BlockPayload, EncodeBytes, GENESIS_VID_NUM_STORAGE_NODES},
        node_implementation::NodeType,
    },
    vid::{
        advz::advz_scheme,
        avidm::{AvidMScheme, init_avidm_param},
        avidm_gf2::{AvidmGf2Scheme, init_avidm_gf2_param},
    },
    vote::HasViewNumber,
};
use jf_advz::VidScheme;

use crate::{
    Header, Payload,
    availability::{
        BlockInfo, BlockQueryData, LeafQueryData, QueryableHeader, QueryablePayload,
        UpdateAvailabilityData, VidCommonQueryData,
    },
    types::HeightIndexed,
};

/// An extension trait for types which implement the update trait for each API module.
///
/// If a type implements [UpdateAvailabilityData] and
/// [UpdateStatusData](crate::status::UpdateStatusData), then it can be fully kept up to date
/// through two interfaces:
/// * [populate_metrics](crate::status::UpdateStatusData::populate_metrics), to get a handle for
///   populating the status metrics, which should be used when initializing a
///   [SystemContextHandle](hotshot::types::SystemContextHandle)
/// * [update](Self::update), provided by this extension trait, to update the query state when a new
///   HotShot event is emitted
#[async_trait]
pub trait UpdateDataSource<Types: NodeType>: UpdateAvailabilityData<Types> {
    /// Update query state based on consensus event.
    ///
    /// The caller is responsible for authenticating `event`. This function does not perform any
    /// authentication, and if given an invalid `event` (one which does not follow from the latest
    /// known state of the ledger) it may panic or silently accept the invalid `event`. This allows
    /// the best possible performance in the case where the query service and the HotShot instance
    /// are running in the same process (and thus the event stream, directly from HotShot) is
    /// trusted.
    ///
    /// If you want to update the data source with an untrusted event, for example one received from
    /// a peer over the network, you must authenticate it first.
    ///
    ///
    /// For each decided leaf the query service stores a `BlockInfo` containing the leaf paired
    /// with a QC that certifies it (`LeafQueryData`), the block payload and VID data when
    /// available, and only on the **newest** leaf in the batch, it stores the proof that finalizes it:
    /// a `qc_chain` for legacy protocol set via `BlockInfo::with_qc_chain`, or a `cert2` for new protocol set
    /// via `BlockInfo::with_cert2`. The two protocols differ only in where those pieces come
    /// from:
    ///
    /// In both events leaves arrive in **newest → oldest** order (`leaves[0]` is the leaf being
    /// finalized; each subsequent leaf is its ancestor reached via `justify_qc`). For the new
    /// protocol, `vid_shares` is parallel to `leaves` (same ordering, one share per leaf). The
    /// handler iterates in reverse so heights are appended ascending.
    ///
    /// - **Legacy (`CoordinatorEvent::LegacyEvent` → `EventType::Decide`).** The newest leaf is
    ///   certified by `committing_qc`; each older leaf is certified by the *next-newer* leaf's
    ///   `justify_qc`. The newest leaf's `qc_chain` is set to `[committing_qc, deciding_qc]` —
    ///   the two consecutive QCs that decide it under the legacy 3-chain rule.
    ///
    /// - **New protocol (`CoordinatorEvent::NewDecide`).** The newest leaf is certified by
    ///   `cert1`; older leaves are again certified by the next leaf's `justify_qc`. When a
    ///   `cert2` is present, it is attached to the newest leaf. Under the new protocol a single
    ///   `cert2` finalizes that leaf directly, replacing the legacy QC chain.
    ///
    /// # Returns
    ///
    /// If all provided data is successfully inserted into the database, returns `Ok(())`. If any
    /// error occurred, the error is logged, and the return value is the height of the first leaf
    /// which failed to be inserted.
    async fn update(&self, event: &CoordinatorEvent<Types>) -> Result<(), u64>;
}

#[async_trait]
impl<Types: NodeType, T> UpdateDataSource<Types> for T
where
    T: UpdateAvailabilityData<Types> + Send + Sync,
    Header<Types>: QueryableHeader<Types>,
    Payload<Types>: QueryablePayload<Types>,
{
    async fn update(&self, event: &CoordinatorEvent<Types>) -> Result<(), u64> {
        match event {
            CoordinatorEvent::LegacyEvent(event) => {
                let EventType::Decide {
                    leaf_chain,
                    committing_qc,
                    deciding_qc,
                    ..
                } = &event.event
                else {
                    return Ok(());
                };

                // `qc` justifies the first (most recent) leaf...
                let qcs = once(committing_qc.qc().clone())
                    // ...and each leaf in the chain justifies the subsequent leaf (its parent)
                    // through `leaf.justify_qc`.
                    .chain(leaf_chain.iter().map(|leaf| leaf.leaf.justify_qc()))
                    // Put the QCs in chronological order.
                    .rev()
                    // The oldest QC is the `justify_qc` of the oldest leaf, which does not justify
                    // any leaf in the new chain, so we don't need it.
                    .skip(1);
                for (
                    qc2,
                    LeafInfo {
                        leaf: leaf2,
                        vid_share,
                        ..
                    },
                ) in qcs.zip(leaf_chain.iter().rev())
                {
                    let height = leaf2.block_header().block_number();

                    let leaf_data = match LeafQueryData::new(leaf2.clone(), qc2.clone()) {
                        Ok(leaf) => leaf,
                        Err(err) => {
                            tracing::error!(
                                height,
                                ?leaf2,
                                ?committing_qc,
                                "inconsistent leaf; cannot append leaf information: {err:#}"
                            );
                            return Err(leaf2.block_header().block_number());
                        },
                    };
                    let block_data = leaf2
                        .block_payload()
                        .map(|payload| BlockQueryData::new(leaf2.block_header().clone(), payload));
                    if block_data.is_none() {
                        tracing::warn!(height, "block payload missing at decide");
                    }

                    let (vid_common, vid_share) = match vid_share {
                        Some(VidDisperseShare::V0(share)) => (
                            Some(VidCommonQueryData::new(
                                leaf2.block_header().clone(),
                                VidCommon::V0(share.common.clone()),
                            )),
                            Some(VidShare::V0(share.share.clone())),
                        ),
                        Some(VidDisperseShare::V1(share)) => (
                            Some(VidCommonQueryData::new(
                                leaf2.block_header().clone(),
                                VidCommon::V1(share.common.clone()),
                            )),
                            Some(VidShare::V1(share.share.clone())),
                        ),
                        Some(VidDisperseShare::V2(share)) => (
                            Some(VidCommonQueryData::new(
                                leaf2.block_header().clone(),
                                VidCommon::V2(share.common.clone()),
                            )),
                            Some(VidShare::V2(share.share.clone())),
                        ),
                        None => {
                            if leaf2.view_number() == ViewNumber::genesis() {
                                // HotShot does not run VID in consensus for the genesis block. In
                                // this case, the block payload is guaranteed to always be empty, so
                                // VID isn't really necessary. But for consistency, we will still
                                // store the VID dispersal data, computing it ourselves based on the
                                // well-known genesis VID commitment.
                                match genesis_vid(leaf2) {
                                    Ok((common, share)) => (Some(common), Some(share)),
                                    Err(err) => {
                                        tracing::warn!("failed to compute genesis VID: {err:#}");
                                        (None, None)
                                    },
                                }
                            } else {
                                (None, None)
                            }
                        },
                    };

                    if vid_common.is_none() {
                        tracing::info!(height, "VID not available at decide");
                    }

                    let mut info = BlockInfo::new(leaf_data, block_data, vid_common, vid_share);
                    if let Some(deciding_qc) = deciding_qc
                        && committing_qc.view_number() == info.leaf.leaf().view_number()
                    {
                        let qc_chain =
                            [committing_qc.as_ref().clone(), deciding_qc.as_ref().clone()];
                        info = info.with_qc_chain(qc_chain);
                    }
                    if let Err(err) = self.append(info).await {
                        tracing::error!(height, "failed to append leaf information: {err:#}");
                        return Err(leaf2.block_header().block_number());
                    }
                }
            },
            CoordinatorEvent::NewDecide {
                leaf_infos,
                cert1,
                cert2,
            } => {
                let Some(first) = leaf_infos.first() else {
                    tracing::error!("new decide event contained no leaves");
                    return Ok(());
                };
                let first_leaf = &first.leaf;

                if let Some(cert2) = cert2
                    && cert2.data.leaf_commit != Committable::commit(first_leaf)
                {
                    tracing::error!(
                        height = first_leaf.height(),
                        cert2_leaf = %cert2.data.leaf_commit,
                        newest_leaf = %Committable::commit(first_leaf),
                        "new decide event cert2 does not certify the newest leaf"
                    );
                    return Err(first_leaf.height());
                }

                // `cert1` certifies the newest leaf; each newer leaf's justify_qc
                // certifies the next older leaf.
                let certifying_qcs = once(cert1.clone())
                    .chain(leaf_infos.iter().map(|info| info.leaf.justify_qc()))
                    .take(leaf_infos.len())
                    .collect::<Vec<_>>();

                for (index, (info, qc)) in leaf_infos.iter().zip(certifying_qcs).enumerate().rev() {
                    let leaf = &info.leaf;
                    let height = leaf.block_header().block_number();

                    let leaf_data = match LeafQueryData::new(leaf.clone(), qc) {
                        Ok(leaf) => leaf,
                        Err(err) => {
                            tracing::error!(
                                height,
                                ?leaf,
                                "inconsistent leaf; cannot append leaf information: {err:#}"
                            );
                            return Err(height);
                        },
                    };

                    let block_data = leaf
                        .block_payload()
                        .map(|payload| BlockQueryData::new(leaf.block_header().clone(), payload));
                    if block_data.is_none() {
                        tracing::warn!(height, "block payload missing at decide");
                    }

                    // Extract VID common data from the new protocol's VidDisperseShare2.
                    let (vid_common, vid_share) = match &info.vid_share {
                        Some(VidDisperseShare::V2(share)) => (
                            Some(VidCommonQueryData::new(
                                leaf.block_header().clone(),
                                VidCommon::V2(share.common.clone()),
                            )),
                            Some(VidShare::V2(share.share.clone())),
                        ),
                        Some(_) => (None, None),
                        None => {
                            if leaf.view_number() == ViewNumber::genesis() {
                                // HotShot does not run VID in consensus for the genesis block. In
                                // this case, the block payload is guaranteed to always be empty, so
                                // VID isn't really necessary. But for consistency, we will still
                                // store the VID dispersal data, computing it ourselves based on the
                                // well-known genesis VID commitment.
                                match genesis_vid(leaf) {
                                    Ok((common, share)) => (Some(common), Some(share)),
                                    Err(err) => {
                                        tracing::warn!("failed to compute genesis VID: {err:#}");
                                        (None, None)
                                    },
                                }
                            } else {
                                (None, None)
                            }
                        },
                    };

                    if vid_common.is_none() {
                        tracing::info!(height, "VID not available at decide");
                    }

                    let mut info = BlockInfo::new(leaf_data, block_data, vid_common, vid_share);

                    // Attach `cert2` only to the newest leaf in the batch (`leaves[0]`, which is
                    // `index == 0` since we iterate in reverse). Under the new protocol a single
                    // `cert2` finalizes that leaf directly
                    // older leaves in the batch are finalized using indirect rule
                    if index == 0
                        && let Some(cert2) = &cert2
                    {
                        info = info.with_cert2(cert2.clone());
                    }

                    if let Err(err) = self.append(info).await {
                        tracing::error!(height, "failed to append leaf information: {err:#}");
                        return Err(height);
                    }
                }
            },
            CoordinatorEvent::BlockPayloadReconstructed {
                header, payload, ..
            } => {
                let block = BlockQueryData::new(header.clone(), payload.clone());
                let height = block.height();
                if let Err(err) = self.append_payload(block).await {
                    tracing::error!(height, "failed to store reconstructed payload: {err:#}");
                    return Err(height);
                }
            },
            _ => {},
        }
        Ok(())
    }
}

fn genesis_vid<Types: NodeType>(
    leaf: &Leaf2<Types>,
) -> anyhow::Result<(VidCommonQueryData<Types>, VidShare)> {
    let payload = Payload::<Types>::empty().0;
    let bytes = payload.encode();

    match leaf.block_header().payload_commitment() {
        VidCommitment::V0(commit) => {
            let mut disperse = advz_scheme(GENESIS_VID_NUM_STORAGE_NODES)
                .disperse(bytes)
                .context("unable to compute VID dispersal for genesis block")?;

            ensure!(
                disperse.commit == commit,
                "computed VID commit {} for genesis block does not match header commit {}",
                disperse.commit,
                commit
            );
            Ok((
                VidCommonQueryData::new(
                    leaf.block_header().clone(),
                    VidCommon::V0(disperse.common),
                ),
                VidShare::V0(disperse.shares.remove(0)),
            ))
        },
        VidCommitment::V1(commit) => {
            let avidm_param = init_avidm_param(GENESIS_VID_NUM_STORAGE_NODES)?;
            let weights = vec![1; GENESIS_VID_NUM_STORAGE_NODES];
            let ns_table = parse_ns_table(bytes.len(), &leaf.block_header().metadata().encode());

            let (calculated_commit, mut shares) =
                AvidMScheme::ns_disperse(&avidm_param, &weights, &bytes, ns_table).unwrap();

            ensure!(
                calculated_commit == commit,
                "computed VID commit {} for genesis block does not match header commit {}",
                calculated_commit,
                commit
            );

            Ok((
                VidCommonQueryData::new(leaf.block_header().clone(), VidCommon::V1(avidm_param)),
                VidShare::V1(shares.remove(0)),
            ))
        },
        VidCommitment::V2(commit) => {
            let avidm_gf2_param = init_avidm_gf2_param(GENESIS_VID_NUM_STORAGE_NODES)?;
            let weights = vec![1; GENESIS_VID_NUM_STORAGE_NODES];
            let ns_table = parse_ns_table(bytes.len(), &leaf.block_header().metadata().encode());

            let (calculated_commit, common, mut shares) =
                AvidmGf2Scheme::ns_disperse(&avidm_gf2_param, &weights, &bytes, ns_table).unwrap();

            ensure!(
                calculated_commit == commit,
                "computed VID commit {} for genesis block does not match header commit {}",
                calculated_commit,
                commit
            );

            Ok((
                VidCommonQueryData::new(leaf.block_header().clone(), VidCommon::V2(common)),
                VidShare::V2(shares.remove(0)),
            ))
        },
    }
}

/// A data source with an atomic transaction-based synchronization interface.
///
/// Changes are made to a versioned data source through a [`Transaction`]. Any changes made in a
/// [`Transaction`] are initially visible only when queried through that same [`Transaction`]. They
/// are not immediately written back to storage, which means that a new data source object opened
/// against the same persistent storage will not reflect the changes. In particular, this means that
/// if the process restarts and reopens its storage, uncommitted changes will be lost.
///
/// Only when a [`Transaction`] is committed are changes written back to storage, synchronized with
/// any concurrent changes, and made visible to other connections to the same data source.
pub trait VersionedDataSource: Send + Sync {
    /// A transaction which can read and modify the data source.
    type Transaction<'a>: Transaction
    where
        Self: 'a;

    type ReadOnly<'a>: Transaction
    where
        Self: 'a;

    /// Start an atomic transaction on the data source.
    fn write(&self) -> impl Future<Output = anyhow::Result<Self::Transaction<'_>>> + Send;

    /// Start a read-only transaction on the data source.
    ///
    /// A read-only transaction allows the owner to string together multiple queries of the data
    /// source, which otherwise would not be atomic with respect to concurrent writes, in an atomic
    /// fashion. Upon returning, [`read`](Self::read) locks in a fully consistent snapshot of the
    /// data source, and any read operations performed upon the transaction thereafter read from the
    /// same consistent snapshot. Concurrent modifications to the data source may occur (for
    /// example, from concurrent [`write`](Self::write) transactions being committed), but their
    /// results will not be reflected in a successful read-only transaction which was opened before
    /// the write was committed.
    ///
    /// Read-only transactions do not need to be committed, and reverting has no effect.
    fn read(&self) -> impl Future<Output = anyhow::Result<Self::ReadOnly<'_>>> + Send;
}

/// A unit of atomicity for updating a shared data source.
///
/// The methods provided by this trait can be used to write such pending changes back to persistent
/// storage ([commit](Self::commit)) so that they become visible to other clients of the same
/// underlying storage, and are saved if the process restarts. It also allows pending changes to be
/// rolled back ([revert](Self::revert)) so that they are never written back to storage and are no
/// longer reflected even through the data source object which was used to make the changes.
pub trait Transaction: Send + Sync + Sized {
    fn commit(self) -> impl Future<Output = anyhow::Result<()>> + Send;
    fn revert(self) -> impl Future + Send;
}