octo
/
Zebra
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Zebra/book/src/dev/rfcs/0005-state-updates.md

48 KiB
Raw Blame History

State Updates

Summary

Zebra manages chain state in the zebra-state crate, which allows state queries via asynchronous RPC (in the form of a Tower service). The state system is responsible for contextual verification in the sense of RFC2, checking that new blocks are consistent with the existing chain state before committing them. This RFC describes how the state is represented internally, and how state updates are performed.

Motivation

We need to be able to access and modify the chain state, and we want to have a description of how this happens and what guarantees are provided by the state service.

Definitions

  • state data: Any data the state service uses to represent chain state.

  • structural/semantic/contextual verification: as defined in RFC2.

  • block chain: A sequence of valid blocks linked by inclusion of the previous block hash in the subsequent block. Chains are rooted at the genesis block and extend to a tip.

  • chain state: The state of the ledger after application of a particular sequence of blocks (state transitions).

  • block work: The approximate amount of work required for a miner to generate a block hash that passes the difficulty filter. The number of block header attempts and the mining time are proportional to the work value. Numerically higher work values represent longer processing times.

  • cumulative work: The sum of the block work of all blocks in a chain, from genesis to the chain tip.

  • best chain: The chain with the greatest cumulative work. This chain represents the consensus state of the Zcash network and transactions.

  • side chain: A chain which is not contained in the best chain. Side chains are pruned at the reorg limit, when they are no longer connected to the finalized state.

  • chain reorganization: Occurs when a new best chain is found and the previous best chain becomes a side chain.

  • reorg limit: The longest reorganization accepted by zcashd, 100 blocks.

  • orphaned block: A block which is no longer included in the best chain.

  • non-finalized state: State data corresponding to blocks above the reorg limit. This data can change in the event of a chain reorg.

  • finalized state: State data corresponding to blocks below the reorg limit. This data cannot change in the event of a chain reorg.

  • non-finalized tips: The highest blocks in each non-finalized chain. These tips might be at different heights.

  • finalized tip: The highest block in the finalized state. The tip of the best chain is usually 100 blocks (the reorg limit) above the finalized tip. But it can be lower during the initial sync, and after a chain reorganization, if the new best chain is at a lower height.

  • relevant chain: The relevant chain for a block starts at the previous block, and extends back to genesis.

  • relevant tip: The tip of the relevant chain.

Guide-level explanation

The zebra-state crate provides an implementation of the chain state storage logic in a Zcash consensus node. Its main responsibility is to store chain state, validating new blocks against the existing chain state in the process, and to allow later querying of said chain state. zebra-state provides this interface via a tower::Service based on the actor model with a request/response interface for passing messages back and forth between the state service and the rest of the application.

The main entry point for the zebra-state crate is the init function. This function takes a zebra_state::Config and constructs a new state service, which it returns wrapped by a tower::Buffer. This service is then interacted with via the tower::Service trait.

use tower::{Service, ServiceExt};

let state = zebra_state::on_disk::init(state_config, network);
let request = zebra_state::Request::BlockLocator;
let response = state.ready_and().await?.call(request).await?;

assert!(matches!(response, zebra_state::Response::BlockLocator(_)));

Note: The tower::Service API requires that ready is always called exactly once before each call. It is up to users of the zebra state service to uphold this contract.

The tower::Buffer wrapper is Cloneable, allowing shared access to a common state service. This allows different tasks to share access to the chain state.

The set of operations supported by zebra-state are encoded in its Request enum. This enum has one variant for each supported operation.

pub enum Request {
    CommitBlock {
        block: Arc<Block>,
    },
    CommitFinalizedBlock {
        block: Arc<Block>,
    },
    Depth(Hash),
    Tip,
    BlockLocator,
    Transaction(Hash),
    Block(HashOrHeight),

    // .. some variants omitted
}

zebra-state breaks down its requests into two categories and provides different guarantees for each category: requests that modify the state, and requests that do not. Requests that update the state are guaranteed to run sequentially and will never race against each other. Requests that read state are done asynchronously and are guaranteed to read at least the state present at the time the request was processed by the service, or a later state present at the time the request future is executed. The state service avoids race conditions between the read state and the written state by doing all contextual verification internally.

Reference-level explanation

State Components

Zcash (as implemented by zcashd) differs from Bitcoin in its treatment of transaction finality. If a new best chain is detected that does not extend the previous best chain, blocks at the end of the previous best chain become orphaned (no longer included in the best chain). Their state updates are therefore no longer included in the best chain's chain state. The process of rolling back orphaned blocks and applying new blocks is called a chain reorganization. Bitcoin allows chain reorganizations of arbitrary depth, while zcashd limits chain reorganizations to 100 blocks. (In zcashd, the new best chain must be a side-chain that forked within 100 blocks of the tip of the current best chain.)

This difference means that in Bitcoin, chain state only has probabilistic finality, while in Zcash, chain state is final once it is beyond the reorg limit. To simplify our implementation, we split the representation of the state data at the finality boundary provided by the reorg limit.

State data from blocks above the reorg limit (non-finalized state) is stored in-memory and handles multiple chains. State data from blocks below the reorg limit (finalized state) is stored persistently using rocksdb and only tracks a single chain. This allows a simplification of our state handling, because only finalized data is persistent and the logic for finalized data handles less invariants.

One downside of this design is that restarting the node loses the last 100 blocks, but node restarts are relatively infrequent and a short re-sync is cheap relative to the cost of additional implementation complexity.

Another downside of this design is that we do not achieve exactly the same behavior as zcashd in the event of a 51% attack: zcashd limits each chain reorganization to 100 blocks, but permits multiple reorgs, while Zebra limits all chain reorgs to 100 blocks. In the event of a successful 51% attack on Zcash, this could be resolved by wiping the rocksdb state and re-syncing the new chain, but in this scenario there are worse problems.

Service Interface

The state is accessed asynchronously through a Tower service interface. Determining what guarantees the state service can and should provide to the rest of the application requires considering two sets of behaviors:

  1. behaviors related to the state's external API (a Buffered tower::Service);
  2. behaviors related to the state's internal implementation (using rocksdb).

Making this distinction helps us to ensure we don't accidentally leak "internal" behaviors into "external" behaviors, which would violate encapsulation and make it more difficult to replace rocksdb.

In the first category, our state is presented to the rest of the application as a Buffered tower::Service. The Buffer wrapper allows shared access to a service using an actor model, moving the service to be shared into a worker task and passing messages to it over an multi-producer single-consumer (mpsc) channel. The worker task receives messages and makes Service::calls. The Service::call method returns a Future, and the service is allowed to decide how much work it wants to do synchronously (in call) and how much work it wants to do asynchronously (in the Future it returns).

This means that our external API ensures that the state service sees a linearized sequence of state requests, although the exact ordering is unpredictable when there are multiple senders making requests.

Because the state service has exclusive access to the rocksdb database, and the state service sees a linearized sequence of state requests, we have an easy way to opt in to asynchronous database access. We can perform rocksdb operations synchronously in the Service::call, waiting for them to complete, and be sure that all future requests will see the resulting rocksdb state. Or, we can perform rocksdb operations asynchronously in the future returned by Service::call.

If we perform all writes synchronously and allow reads to be either synchronous or asynchronous, we ensure that writes cannot race each other. Asynchronous reads are guaranteed to read at least the state present at the time the request was processed, or a later state.

Summary

  • rocksdb reads may be done synchronously (in call) or asynchronously (in the Future), depending on the context;

  • rocksdb writes must be done synchronously (in call)

In-memory data structures

At a high level, the in-memory data structures store a collection of chains, each rooted at the highest finalized block. Each chain consists of a map from heights to blocks. Chains are stored using an ordered map from cumulative work to chains, so that the map ordering is the ordering of worst to best chains.

The Chain type

The Chain type represents a chain of blocks. Each block represents an incremental state update, and the Chain type caches the cumulative state update from its root to its tip.

The Chain type is used to represent the non-finalized portion of a complete chain of blocks rooted at the genesis block. The parent block of the root of a Chain is the tip of the finalized portion of the chain. As an exception, the finalized portion of the chain is initially empty, until the genesis block has been finalized.

The Chain type supports several operations to manipulate chains, push, pop_root, and fork. push is the most fundamental operation and handles contextual validation of chains as they are extended. pop_root is provided for finalization, and is how we move blocks from the non-finalized portion of the state to the finalized portion. fork on the other hand handles creating new chains for push when new blocks arrive whose parent isn't a tip of an existing chain.

Note: The Chain type's API is only designed to handle non-finalized data. The genesis block and all pre canopy blocks are always considered to be finalized blocks and should not be handled via the Chain type through CommitBlock. They should instead be committed directly to the finalized state with CommitFinalizedBlock. This is particularly important with the genesis block since the Chain will panic if used while the finalized state is completely empty.

The Chain type is defined by the following struct and API:

#[derive(Debug, Default, Clone)]
struct Chain {
    blocks: BTreeMap<block::Height, Arc<Block>>,
    height_by_hash: HashMap<block::Hash, block::Height>,
    tx_by_hash: HashMap<transaction::Hash, (block::Height, usize)>,

    created_utxos: HashSet<transparent::OutPoint>,
    spent_utxos: HashSet<transparent::OutPoint>,
    sprout_anchors: HashSet<sprout::tree::Root>,
    sapling_anchors: HashSet<sapling::tree::Root>,
    sprout_nullifiers: HashSet<sprout::Nullifier>,
    sapling_nullifiers: HashSet<sapling::Nullifier>,
    orchard_nullifiers: HashSet<orchard::Nullifier>,
    partial_cumulative_work: PartialCumulativeWork,
}

pub fn push(&mut self, block: Arc<Block>)

Push a block into a chain as the new tip

  1. Update cumulative data members

    • Add the block's hash to height_by_hash
    • Add work to self.partial_cumulative_work
    • For each transaction in block
      • Add key: transaction.hash and value: (height, tx_index) to tx_by_hash
      • Add created utxos to self.created_utxos
      • Add spent utxos to self.spent_utxos
      • Add nullifiers to the appropriate self.<version>_nullifiers

  2. Add block to self.blocks

pub fn pop_root(&mut self) -> Arc<Block>

Remove the lowest height block of the non-finalized portion of a chain.

  1. Remove the lowest height block from self.blocks

  2. Update cumulative data members

    • Remove the block's hash from self.height_by_hash
    • Subtract work from self.partial_cumulative_work
    • For each transaction in block
      • Remove transaction.hash from tx_by_hash
      • Remove created utxos from self.created_utxos
      • Remove spent utxos from self.spent_utxos
      • Remove the nullifiers from the appropriate self.<version>_nullifiers

  3. Return the block

pub fn fork(&self, new_tip: block::Hash) -> Option<Self>

Fork a chain at the block with the given hash, if it is part of this chain.

  1. If self does not contain new_tip return None

  2. Clone self as forked

  3. While the tip of forked is not equal to new_tip

    • call forked.pop_tip() and discard the old tip

  4. Return forked

fn pop_tip(&mut self)

Remove the highest height block of the non-finalized portion of a chain.

  1. Remove the highest height block from self.blocks

  2. Update cumulative data members

    • Remove the corresponding hash from self.height_by_hash
    • Subtract work from self.partial_cumulative_work
    • for each transaction in block
      • remove transaction.hash from tx_by_hash
      • Remove created utxos from self.created_utxos
      • Remove spent utxos from self.spent_utxos
      • Remove the nullifiers from the appropriate self.<version>_nullifiers

Ord

The Chain type implements Ord for reorganizing chains. First chains are compared by their partial_cumulative_work. Ties are then broken by comparing block::Hashes of the tips of each chain. (This tie-breaker means that all Chains in the NonFinalizedState must have at least one block.)

Note: Unlike zcashd, Zebra does not use block arrival times as a tie-breaker for the best tip. Since Zebra downloads blocks in parallel, download times are not guaranteed to be unique. Using the block::Hash provides a consistent tip order. (As a side-effect, the tip order is also consistent after a node restart, and between nodes.)

Default

The Chain type implements Default for constructing new chains whose parent block is the tip of the finalized state. This implementation should be handled by #[derive(Default)].

  1. initialise cumulative data members
    • Construct an empty self.blocks, height_by_hash, tx_by_hash, self.created_utxos, self.spent_utxos, self.<version>_anchors, self.<version>_nullifiers
    • Zero self.partial_cumulative_work

Note: The ChainState can be empty after a restart, because the non-finalized state is empty.

NonFinalizedState Type

The NonFinalizedState type represents the set of all non-finalized state. It consists of a set of non-finalized but verified chains and a set of unverified blocks which are waiting for the full context needed to verify them to become available.

NonFinalizedState is defined by the following structure and API:

/// The state of the chains in memory, including queued blocks.
#[derive(Debug, Default)]
pub struct NonFinalizedState {
    /// Verified, non-finalized chains.
    chain_set: BTreeSet<Chain>,
    /// Blocks awaiting their parent blocks for contextual verification.
    contextual_queue: QueuedBlocks,
}

pub fn finalize(&mut self) -> Arc<Block>

Finalize the lowest height block in the non-finalized portion of the best chain and updates all side chains to match.

  1. Extract the best chain from self.chain_set into best_chain

  2. Extract the rest of the chains into a side_chains temporary variable, so they can be mutated

  3. Remove the lowest height block from the best chain with let finalized_block = best_chain.pop_root();

  4. Add best_chain back to self.chain_set if best_chain is not empty

  5. For each remaining chain in side_chains

    • remove the lowest height block from chain
    • If that block is equal to finalized_block and chain is not empty add chain back to self.chain_set
    • Else, drop chain

  6. Return finalized_block

fn commit_block(&mut self, block: Arc<Block>)

Commit block to the non-finalized state.

  1. If the block is a pre-Canopy block, or the canopy activation block, panic.

  2. If any chains tip hash equal block.header.previous_block_hash remove that chain from self.chain_set

  3. Else Find the first chain that contains block.parent and fork it with block.parent as the new tip

    • let fork = self.chain_set.iter().find_map(|chain| chain.fork(block.parent));

  4. Else panic, this should be unreachable because commit_block is only called when block is ready to be committed.

  5. Push block into parent_chain

  6. Insert parent_chain into self.chain_set

pub(super) fn commit_new_chain(&mut self, block: Arc<Block>)

Construct a new chain starting with block.

  1. Construct a new empty chain

  2. push block into that new chain

  3. Insert the new chain into self.chain_set

The QueuedBlocks type

The queued blocks type represents the non-finalized blocks that were committed before their parent blocks were. It is responsible for tracking which blocks are queued by their parent so they can be committed immediately after the parent is committed. It also tracks blocks by their height so they can be discarded if they ever end up below the reorg limit.

NonFinalizedState is defined by the following structure and API:

/// A queue of blocks, awaiting the arrival of parent blocks.
#[derive(Debug, Default)]
struct QueuedBlocks {
    /// Blocks awaiting their parent blocks for contextual verification.
    blocks: HashMap<block::Hash, QueuedBlock>,
    /// Hashes from `queued_blocks`, indexed by parent hash.
    by_parent: HashMap<block::Hash, Vec<block::Hash>>,
    /// Hashes from `queued_blocks`, indexed by block height.
    by_height: BTreeMap<block::Height, Vec<block::Hash>>,
}

pub fn queue(&mut self, new: QueuedBlock)

Add a block to the queue of blocks waiting for their requisite context to become available.

  1. extract the parent_hash, new_hash, and new_height from new.block

  2. Add new to self.blocks using new_hash as the key

  3. Add new_hash to the set of hashes in self.by_parent.entry(parent_hash).or_default()

  4. Add new_hash to the set of hashes in self.by_height.entry(new_height).or_default()

pub fn dequeue_children(&mut self, parent: block::Hash) -> Vec<QueuedBlock>

Dequeue the set of blocks waiting on parent.

  1. Remove the set of hashes waiting on parent from self.by_parent

  2. Remove and collect each block in that set of hashes from self.blocks as queued_children

  3. For each block in queued_children remove the associated block.hash from self.by_height

  4. Return queued_children

pub fn prune_by_height(&mut self, finalized_height: block::Height)

Prune all queued blocks whose height are less than or equal to finalized_height.

  1. Split the by_height list at the finalized height, removing all heights that are below finalized_height

  2. for each hash in the removed values of by_height

    • remove the corresponding block from self.blocks
    • remove the block's hash from the list of blocks waiting on block.header.previous_block_hash from self.by_parent

Summary

  • Chain represents the non-finalized portion of a single chain

  • NonFinalizedState represents the non-finalized portion of all chains

  • QueuedBlocks represents all unverified blocks that are waiting for context to be available.

The state service uses the following entry points:

  • commit_block when it receives new blocks.

  • finalize to prevent chains in NonFinalizedState from growing beyond the reorg limit.

  • FinalizedState.queue_and_commit_finalized_blocks on the blocks returned by finalize, to commit those finalized blocks to disk.

Committing non-finalized blocks

New non-finalized blocks are committed as follows:

pub(super) fn queue_and_commit_non_finalized_blocks(&mut self, new: Arc<Block>) -> tokio::sync::oneshot::Receiver<block::Hash>

  1. If a duplicate block hash exists in a non-finalized chain, or the finalized chain, it has already been successfully verified:

    • create a new oneshot channel
    • immediately send Err(DuplicateBlockHash) drop the sender
    • return the receiver

  2. If a duplicate block hash exists in the queue:

    • Find the QueuedBlock for that existing duplicate block
    • create a new channel for the new request
    • replace the old sender in queued_block with the new sender
    • send Err(DuplicateBlockHash) through the old sender channel
    • continue to use the new receiver

  3. Else create a QueuedBlock for block:

    • Create a tokio::sync::oneshot channel
    • Use that channel to create a QueuedBlock for block
    • Add block to self.queued_blocks
    • continue to use the new receiver

  4. If block.header.previous_block_hash is not present in the finalized or non-finalized state:

    • Return the receiver for the block's channel

  5. Else iteratively attempt to process queued blocks by their parent hash starting with block.header.previous_block_hash

  6. While there are recently committed parent hashes to process

    • Dequeue all blocks waiting on parent with let queued_children = self.queued_blocks.dequeue_children(parent);
    • for each queued block
      • Run contextual validation on block
        • contextual validation should check that the block height is equal to the previous block height plus 1. This check will reject blocks with invalid heights.
      • If the block fails contextual validation send the result to the associated channel
      • Else if the block's previous hash is the finalized tip add to the non-finalized state with self.mem.commit_new_chain(block)
      • Else add the new block to an existing non-finalized chain or new fork with self.mem.commit_block(block);
      • Send Ok(hash) over the associated channel to indicate the block was successfully committed
      • Add block.hash to the set of recently committed parent hashes to process

  7. While the length of the non-finalized portion of the best chain is greater than the reorg limit

    • Remove the lowest height block from the non-finalized state with self.mem.finalize();
    • Commit that block to the finalized state with self.disk.commit_finalized_direct(finalized);

  8. Prune orphaned blocks from self.queued_blocks with self.queued_blocks.prune_by_height(finalized_height);

  9. Return the receiver for the block's channel

rocksdb data structures

rocksdb provides a persistent, thread-safe BTreeMap<&[u8], &[u8]>. Each map is a distinct "tree". Keys are sorted using lex order on byte strings, so integer values should be stored using big-endian encoding (so that the lex order on byte strings is the numeric ordering).

We use the following rocksdb column families:

Column Family Keys Values Updates
Blocks
hash_by_height block::Height block::Hash Never
height_tx_count_by_hash block::Hash HeightTransactionCount Never
block_header_by_height block::Height block::Header Never
Transactions
tx_by_loc TransactionLocation Transaction Never
hash_by_tx_loc TransactionLocation transaction::Hash Never
tx_loc_by_hash transaction::Hash TransactionLocation Never
Transparent
utxo_by_out_loc OutputLocation Output || AddressLocation Delete
balance_by_transparent_addr transparent::Address Amount || AddressLocation Update
utxo_by_transparent_addr_loc AddressLocation AtLeastOne<OutputLocation> Up/Del
tx_by_transparent_addr_loc AddressLocation AtLeastOne<TransactionLocation> Append
Sprout
sprout_nullifiers sprout::Nullifier () Never
sprout_anchors sprout::tree::Root sprout::tree::NoteCommitmentTree Never
sprout_note_commitment_tree block::Height sprout::tree::NoteCommitmentTree Delete
Sapling
sapling_nullifiers sapling::Nullifier () Never
sapling_anchors sapling::tree::Root () Never
sapling_note_commitment_tree block::Height sapling::tree::NoteCommitmentTree Never
Orchard
orchard_nullifiers orchard::Nullifier () Never
orchard_anchors orchard::tree::Root () Never
orchard_note_commitment_tree block::Height orchard::tree::NoteCommitmentTree Never
Chain
history_tree block::Height NonEmptyHistoryTree Delete
tip_chain_value_pool () ValueBalance Update

Zcash structures are encoded using ZcashSerialize/ZcashDeserialize. Other structures are encoded using IntoDisk/FromDisk.

Block and Transaction Data:

  • Height: 24 bits, big-endian, unsigned (allows for ~30 years worth of blocks)
  • TransactionIndex: 16 bits, big-endian, unsigned (max ~23,000 transactions in the 2 MB block limit)
  • TransactionCount: same as TransactionIndex
  • TransactionLocation: Height \|\| TransactionIndex
  • HeightTransactionCount: Height \|\| TransactionCount
  • OutputIndex: 24 bits, big-endian, unsigned (max ~223,000 transfers in the 2 MB block limit)
  • transparent and shielded input indexes, and shielded output indexes: 16 bits, big-endian, unsigned (max ~49,000 transfers in the 2 MB block limit)
  • OutputLocation: TransactionLocation \|\| OutputIndex
  • AddressLocation: the first OutputLocation used by a transparent::Address. Always has the same value for each address, even if the first output is spent.
  • Utxo: Output, derives extra fields from the OutputLocation key
  • AtLeastOne<T>: [T; AtLeastOne::len()] (for known-size T)

We use big-endian encoding for keys, to allow database index prefix searches.

Amounts:

  • Amount: 64 bits, little-endian, signed
  • ValueBalance: [Amount; 4]

Derived Formats:

  • *::NoteCommitmentTree: bincode using serde
  • NonEmptyHistoryTree: bincode using serde, using zcash_history's serde implementation

Implementing consensus rules using rocksdb

Each column family handles updates differently, based on its specific consensus rules:

  • Never: Keys are never deleted, values are never updated. The value for each key is inserted once.
  • Delete: Keys can be deleted, but values are never updated. The value for each key is inserted once.
    • Code called by ReadStateService must ignore deleted keys, or use a read lock.
    • TODO: should we prevent re-inserts of keys that have been deleted?
  • Update: Keys are never deleted, but values can be updated.
    • Code called by ReadStateService must accept old or new values, or use a read lock.
  • Append: Keys are never deleted, existing values are never updated, but sets of values can be extended with more entries.
    • Code called by ReadStateService must accept truncated or extended sets, or use a read lock.
  • Up/Del: Keys can be deleted, and values can be added or removed from sets.
    • Code called by ReadStateService must ignore deleted keys and values, accept truncated or extended sets, and accept old or new values. Or it should use a read lock.

RocksDB read locks

The read-only ReadStateService needs to handle concurrent writes and deletes of the finalized column families it reads. It must also handle overlaps between the cached non-finalized Chain, and the current finalized state database.

The StateService uses RocksDB transactions for each block write. So ReadStateService queries that only access a single key or value will always see a consistent view of the database.

If a ReadStateService query only uses column families that have keys and values appended (Never in the Updates table above), it should ignore extra appended values. Most queries do this by default.

For more complex queries, there are several options:

Reading across multiple column families:

  1. Ignore deleted values using custom Rust code
  2. Take a database snapshot - https://docs.rs/rocksdb/latest/rocksdb/struct.DBWithThreadMode.html#method.snapshot

Reading a single column family: 3. multi_get - https://docs.rs/rocksdb/latest/rocksdb/struct.DBWithThreadMode.html#method.multi_get_cf 4. iterator - https://docs.rs/rocksdb/latest/rocksdb/struct.DBWithThreadMode.html#method.iterator_cf

RocksDB also has read transactions, but they don't seem to be exposed in the Rust crate.

Low-Level Implementation Details

RocksDB ignores duplicate puts and deletes, preserving the latest values. If rejecting duplicate puts or deletes is consensus-critical, check db.get_cf(cf, key)? before putting or deleting any values in a batch.

Currently, these restrictions should be enforced by code review:

  • multiple zs_inserts are only allowed on Update column families, and
  • delete_cf is only allowed on Delete column families.

In future, we could enforce these restrictions by:

  • creating traits for Never, Delete, and Update
  • doing different checks in zs_insert depending on the trait
  • wrapping delete_cf in a trait, and only implementing that trait for types that use Delete column families.

As of June 2021, the Rust rocksdb crate ignores the delete callback, and merge operators are unreliable (or have undocumented behaviour). So they should not be used for consensus-critical checks.

Notes on rocksdb column families

  • The hash_by_height and height_tx_count_by_hash column families provide a bijection between block heights and block hashes. (Since the rocksdb state only stores finalized state, they are actually a bijection).

  • Similarly, the tx_by_hash and hash_by_tx column families provide a bijection between transaction locations and transaction hashes.

  • The block_header_by_height column family provides a bijection between block heights and block header data. There is no corresponding height_by_block column family: instead, hash the block, and use the hash from height_tx_count_by_hash. (Since the rocksdb state only stores finalized state, they are actually a bijection). Similarly, there are no column families that go from transaction data to transaction locations: hash the transaction and use tx_by_hash.

  • Block headers and transactions are stored separately in the database, so that individual transactions can be accessed efficiently. Blocks can be re-created on request using the following process:

    • Look up height and tx_count in height_tx_count_by_hash
    • Get the block header for height from block_header_by_height
    • Use prefix_iterator or multi_get to get each transaction from 0..tx_count from tx_by_location

  • Block headers are stored by height, not by hash. This has the downside that looking up a block by hash requires an extra level of indirection. The upside is that blocks with adjacent heights are adjacent in the database, and many common access patterns, such as helping a client sync the chain or doing analysis, access blocks in (potentially sparse) height order. In addition, the fact that we commit blocks in order means we're writing only to the end of the rocksdb column family, which may help save space.

  • Similarly, transaction data is stored in chain order in tx_by_location, and chain order within each vector in tx_by_transparent_address.

  • TransactionLocations are stored as a (height, index) pair referencing the height of the transaction's parent block and the transaction's index in that block. This would more traditionally be a (hash, index) pair, but because we store blocks by height, storing the height saves one level of indirection. Transaction hashes can be looked up using hash_by_tx.

  • Similarly, UTXOs are stored in utxo_by_outpoint by OutputLocation, rather than OutPoint. OutPoints can be looked up using tx_by_hash, and reconstructed using hash_by_tx.

  • The Utxo type can be constructed from the Output data, height: TransactionLocation.height, and is_coinbase: TransactionLocation.index == 0 (coinbase transactions are always the first transaction in a block).

  • balance_by_transparent_addr is the sum of all utxo_by_transparent_addr_locs that are still in utxo_by_outpoint. It is cached to improve performance for addresses with large UTXO sets. It also stores the AddressLocation for each address, which allows for efficient lookups.

  • utxo_by_transparent_addr_loc stores unspent transparent output locations by address. UTXO locations are appended by each block. This list includes the AddressLocation, if it has not been spent. (This duplicate data is small, and helps simplify the code.)

  • When a block write deletes a UTXO from utxo_by_outpoint, that UTXO location should be deleted from utxo_by_transparent_addr_loc. This is an index optimisation.

  • tx_by_transparent_addr_loc stores transaction locations by address. This list includes transactions containing spent UTXOs. It also includes the TransactionLocation from the AddressLocation. (This duplicate data is small, and helps simplify the code.)

  • The sprout_note_commitment_tree stores the note commitment tree state at the tip of the finalized state, for the specific pool. There is always a single entry. Each tree is stored as a "Merkle tree frontier" which is basically a (logarithmic) subset of the Merkle tree nodes as required to insert new items. For each block committed, the old tree is deleted and a new one is inserted by its new height. TODO: store the sprout note commitment tree by (), to avoid ReadStateService concurrent write issues.

  • The {sapling, orchard}_note_commitment_tree stores the note commitment tree state for every height, for the specific pool. Each tree is stored as a "Merkle tree frontier" which is basically a (logarithmic) subset of the Merkle tree nodes as required to insert new items.

  • history_tree stores the ZIP-221 history tree state at the tip of the finalized state. There is always a single entry for it. The tree is stored as the set of "peaks" of the "Merkle mountain range" tree structure, which is what is required to insert new items. TODO: store the history tree by (), to avoid ReadStateService concurrent write issues.

  • Each *_anchors stores the anchor (the root of a Merkle tree) of the note commitment tree of a certain block. We only use the keys since we just need the set of anchors, regardless of where they come from. The exception is sprout_anchors which also maps the anchor to the matching note commitment tree. This is required to support interstitial treestates, which are unique to Sprout.

  • The value pools are only stored for the finalized tip.

  • We do not store the cumulative work for the finalized chain, because the finalized work is equal for all non-finalized chains. So the additional non-finalized work can be used to calculate the relative chain order, and choose the best chain.

Committing finalized blocks

If the parent block is not committed, add the block to an internal queue for future processing. Otherwise, commit the block described below, then commit any queued children. (Although the checkpointer generates verified blocks in order when it completes a checkpoint, the blocks are committed in the response futures, so they may arrive out of order).

Committing a block to the rocksdb state should be implemented as a wrapper around a function also called by Request::CommitBlock, which should:

pub(super) fn queue_and_commit_finalized_blocks(&mut self, queued_block: QueuedBlock)

  1. Obtain the highest entry of hash_by_height as (old_height, old_tip). Check that block's parent hash is old_tip and its height is old_height+1, or panic. This check is performed as defense-in-depth to prevent database corruption, but it is the caller's responsibility (e.g. the zebra-state service's responsibility) to commit finalized blocks in order.

The genesis block does not have a parent block. For genesis blocks, check that block's parent hash is null (all zeroes) and its height is 0.

  1. Insert the block and transaction data into the relevant column families.

  2. If the block is a genesis block, skip any transaction updates.

    (Due to a bug in zcashd, genesis block anchors and transactions are ignored during validation.)

  3. Update the block anchors, history tree, and chain value pools.

  4. Iterate over the enumerated transactions in the block. For each transaction, update the relevant column families.

Note: The Sprout and Sapling anchors are the roots of the Sprout and Sapling note commitment trees that have already been calculated for the last transaction(s) in the block that have JoinSplits in the Sprout case and/or Spend/Output descriptions in the Sapling case. These should be passed as fields in the Commit*Block requests.

Due to the coinbase maturity rules, the Sprout root is the empty root for the first 100 blocks. (These rules are already implemented in contextual validation and the anchor calculations.)

Hypothetically, if Sapling were activated from genesis, the specification requires a Sapling anchor, but zcashd would ignore that anchor.

These updates can be performed in a batch or without necessarily iterating over all transactions, if the data is available by other means; they're specified this way for clarity.

Accessing previous blocks for contextual validation

The state service performs contextual validation of blocks received via the CommitBlock request. Since CommitBlock is synchronous, contextual validation must also be performed synchronously.

The relevant chain for a block starts at its previous block, and follows the chain of previous blocks back to the genesis block.

Relevant chain iterator

The relevant chain can be retrieved from the state service as follows:

  • if the previous block is the finalized tip:
    • get recent blocks from the finalized state
  • if the previous block is in the non-finalized state:
    • get recent blocks from the relevant chain, then
    • get recent blocks from the finalized state, if required

The relevant chain can start at any non-finalized block, or at the finalized tip.

Relevant chain implementation

The relevant chain is implemented as a StateService iterator, which returns Arc<Block>s.

The chain iterator implements ExactSizeIterator, so Zebra can efficiently assert that the relevant chain contains enough blocks to perform each contextual validation check.

impl StateService {
    /// Return an iterator over the relevant chain of the block identified by
    /// `hash`.
    ///
    /// The block identified by `hash` is included in the chain of blocks yielded
    /// by the iterator.
    pub fn chain(&self, hash: block::Hash) -> Iter<'_> { ... }
}

impl Iterator for Iter<'_>  {
    type Item = Arc<Block>;
    ...
}
impl ExactSizeIterator for Iter<'_> { ... }
impl FusedIterator for Iter<'_> {}

For further details, see PR 1271.

Request / Response API

The state API is provided by a pair of Request/Response enums. Each Request variant corresponds to particular Response variants, and it's fine (and encouraged) for caller code to unwrap the expected variants with unreachable! on the unexpected variants. This is slightly inconvenient but it means that we have a unified state interface with unified backpressure.

This API includes both write and read calls. Spotting Commit requests in code review should not be a problem, but in the future, if we need to restrict access to write calls, we could implement a wrapper service that rejects these, and export "read" and "write" frontends to the same inner service.

Request::CommitBlock

CommitBlock {
    block: Arc<Block>,
    sprout_anchor: sprout::tree::Root,
    sapling_anchor: sapling::tree::Root,
}

Performs contextual validation of the given block, committing it to the state if successful. Returns Response::Added(block::Hash) with the hash of the newly committed block or an error.

Request::CommitFinalizedBlock

CommitFinalizedBlock {
    block: Arc<Block>,
    sprout_anchor: sprout::tree::Root,
    sapling_anchor: sapling::tree::Root,
}

Commits a finalized block to the rocksdb state, skipping contextual validation. This is exposed for use in checkpointing, which produces in-order finalized blocks. Returns Response::Added(block::Hash) with the hash of the committed block if successful.

Request::Depth(block::Hash)

Computes the depth in the best chain of the block identified by the given hash, returning

  • Response::Depth(Some(depth)) if the block is in the best chain;
  • Response::Depth(None) otherwise.

Implemented by querying:

  • (non-finalized) the height_by_hash map in the best chain, and
  • (finalized) the height_by_hash tree

Request::Tip

Returns Response::Tip(block::Hash) with the current best chain tip.

Implemented by querying:

  • (non-finalized) the highest height block in the best chain
  • (finalized) the highest height block in the hash_by_height tree, if the non-finalized state is empty

Request::BlockLocator

Returns Response::BlockLocator(Vec<block::Hash>) with hashes starting from the current chain tip and reaching backwards towards the genesis block. The first hash is the best chain tip. The last hash is the tip of the finalized portion of the state. If the finalized and non-finalized states are both empty, the block locator is also empty.

This can be used by the sync component to request hashes of subsequent blocks.

Implemented by querying:

  • (non-finalized) the hash_by_height map in the best chain
  • (finalized) the hash_by_height tree.

Request::Transaction(transaction::Hash)

Returns

  • Response::Transaction(Some(Transaction)) if the transaction identified by the given hash is contained in the state;

  • Response::Transaction(None) if the transaction identified by the given hash is not contained in the state.

Implemented by querying:

  • (non-finalized) the tx_by_hash map (to get the block that contains the transaction) of each chain starting with the best chain, and then find block that chain's blocks (to get the block containing the transaction data)
  • (finalized) the tx_by_hash tree (to get the block that contains the transaction) and then block_by_height tree (to get the block containing the transaction data), if the transaction is not in any non-finalized chain

Request::Block(block::Hash)

Returns

  • Response::Block(Some(Arc<Block>)) if the block identified by the given hash is contained in the state;

  • Response::Block(None) if the block identified by the given hash is not contained in the state;

Implemented by querying:

  • (non-finalized) the height_by_hash of each chain starting with the best chain, then find block that chain's blocks (to get the block data)
  • (finalized) the height_by_hash tree (to get the block height) and then the block_by_height tree (to get the block data), if the block is not in any non-finalized chain

Request::AwaitSpendableUtxo { outpoint: OutPoint, spend_height: Height, spend_restriction: SpendRestriction }

Returns

  • Response::SpendableUtxo(transparent::Output)

Implemented by querying:

  • (non-finalized) if any Chains contain OutPoint in their created_utxos, return the Utxo for OutPoint;
  • (finalized) else if OutPoint is in utxos_by_outpoint, return the Utxo for OutPoint;
  • else wait for OutPoint to be created as described in RFC0004;

Then validating:

  • check the transparent coinbase spend restrictions specified in RFC0004;
  • if the restrictions are satisfied, return the response;
  • if the spend is invalid, drop the request (and the caller will time out).

Drawbacks

  • Restarts can cause zebrad to redownload up to the last one hundred blocks it verified in the best chain, and potentially some recent side-chain blocks.

  • The service interface puts some extra responsibility on callers to ensure it is used correctly and does not verify the usage is correct at compile time.

  • the service API is verbose and requires manually unwrapping enums

  • We do not handle reorgs the same way zcashd does, and could in theory need to delete our entire on disk state and resync the chain in some pathological reorg cases.

  • testnet rollbacks are infrequent, but possible, due to bugs in testnet releases. Each testnet rollback will require additional state service code.