* Add an empty balance_by_transparent_addr column family
* Add an AddressBalanceLocation type for balance_by_transparent_addr
* Add serialization for balance_by_transparent_addr types
* Add round-trip tests for the new serialized types
* Add missing round-trip and serialized equality tests
* Add a network field to DiskWriteBatch
* Refactor confusing all_utxos_spent_by_block argument
It was actually just the UTXOs from the state spent by the block,
excluding the UTXOs created and spent within the block.
But now we need it to contain all the spent outputs,
including the ones created by the block.
* Read and update address balances in the finalized state
* Update raw data snapshots for transparent address balances
* Add test-only deserialization for transparent addresses
* Add high-level snapshot test code for address balances
* Add high-level snapshots for address balances
* Increment the state version after NU5 testnet 2 rollback
* Implement disk serialization for block headers and transactions
* Re-order column family initialization to match the design
* Add new empty transaction column families
* Split writing block header and transaction data
* Re-order column families for consistency
* Update write snapshots for transaction split
* Use split block and transaction data when reading
* Update snapshots to include genesis transaction hash location
* Filter all prefix iterators to make sure they return the correct values
* Test that the new transaction indexes are consistent
* Add some cleanup TODOs
* Increment the database format to version 15
* Remove unused fisk format impls for Block
* Add a missing prefix extractor for transaction locations
* Make the database generic over the thread mode
* Replace prefix iteration with iteration from a key, and a filter
Prefix iteration caused database hangs.
* Manually iterate through transaction locations to re-create blocks
Also:
- re-write disk read API to avoid iterator hangs
- move disk read API to ReadDisk
- re-write impl rocksdb::AsColumnFamilyRef to a where clause, for consistency
* Update the database version so it's larger than the NU5 testnet 2 version
* Increment the database format version
* Update IntoDisk and FromDisk docs
* Rename fixed_byte_len to fixed_disk_byte_len
* Add functions that truncate and extend serialized bytes
* Store heights in 3 bytes on disk
* Update database raw data snapshots for 3-byte heights
* Log an error if we ever get close to the maximum disk height
* Store transaction indexes in 2 bytes on disk
* Update database raw data snapshots for 2-byte transaction indexes
* Make doc comment phrasing consistent
* Replace IntoDiskFixed with fixed constants
* Replace u32 byte length literal with a constant calculation
* Fix off-by-one error in MAX_ON_DISK_HEIGHT
* Add proptest seeds for the MAX_ON_DISK_HEIGHT off-by-one error
* Remove redundant module from a Height type
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
* Add round-trip test for OutputLocation disk serialization
* Make the transaction snapshot tests more accurate
Previously, we were showing the genesis transaction hash at height 0, index 0.
But its hash is actually not stored by location in the database,
because the genesis transaction is skipped due to a consensus rule.
* Update the transaction snapshot data
* Add history tree snapshot tests
At the current test heights, the history trees are all empty.
* Add the history tree snapshot data
* Update comments
* Simplify snapshot test code
* Make some serde::Serialize impls test-only, so we can hex-encode them
This should also speed up release compilation a bit.
* Add snapshot test code for UTXOs
* Add snapshot data for UTXOs
* Add request metrics to ReadStateService
Add a field to distinguish StateService and ReadStateService metrics.
Add missing StateService request metrics.
* Refactor state transaction lookup so it can be shared between services
* Implement ReadState Request::Transaction
* Revert "Drop the Chain watch channel lock before accessing the finalized state"
This reverts commit 8870944d1323fc31e2d009b9938623de6fcbe50f.
* Add a WatchReceiver wrapper that always clones the borrowed watch data
This avoids deadlocks, by holding the read lock for as short a time as possible.
* Drop the shared Arc<Chain>s as quickly as possible
This reduces memory usage.
* Make read::block more flexible, by accepting any AsRef<Chain>
* Make the block method docs consistent
* Avoid livelocks by explicitly dropping the borrow after the clone
* Query Sapling & Orchard trees by height in the finalized state
* Add Sapling & Orchard trees to the non-finalized state
* Add a TODO about concurrent read-only access to Sprout tree
Co-authored-by: teor <teor@riseup.net>
* Update the database format version
* Keep only the most recent Sprout tree in the database
* Check that the database returns empty trees for the genesis block
* Assert that the database returns the highest trees
* Document how to update insta snapshots
* Add note commitment tree insta snapshot tests
* Add comments about cached tree roots in snapshots
* Add snapshot data for sapling and orchard trees
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: teor <teor@riseup.net>
* Add state service module docs and cleanup
* Move and add finalized state methods
* Add chain and non-finalized state methods
* Cleanup methods and imports
* Create a ReadStateService type
* Add a stub service implementation
* Add a TODO
* Update ReadStateService request stubs with RPC names and tickets
* Documentation updates
* Make RPC State generic bounds accept a buffered state and a read-only state
* Doc updates
* Add missing proptest-impl feature in RPC dev dependencies
* Move the legacy chain check to the `check` module
And move `populated_state` to the `arbitrary` module.
* Cleanup imports
* Document the state service struct
* Split state block iter into its own module
* Prepare the finalized state for read-only state
* Add a forced shutdown mode, used in test code before forced exits
* Document the small database drop race condition window
* refactor(test/block): rename large single transaction function
```sh
fastmod single_transaction_block single_transaction_block_many_inputs
```
* rustfmt
* test(block): add a test block with many transparent outputs
* doc(db): explain why we can't just get the UTXOs right before they are deleted
* refactor(db): split out a block data write method
* refactor(block): add a height argument to new_outputs
* test(db): add block and transaction round-trip tests
Including large blocks and transactions.
* test(db): fix large block serialization instability in the tests
* doc(block): add TODOs for generating correct blocks
* Make transparent output functions which take a height test-only
* make sure generated blocks are actually over/under-sized
* replace println!() with an error!() log
* refactor(db): simplify block height serialization
* refactor(db): make height serialization length generic
* refactor(db): create a TransactionIndex type
This changes the names of some snapshot types,
but doesn't change any data.
* refactor(db): create transparent OutputIndex and OutputLocation types
This keeps the same serialization, to avoid changing the database version.
* doc(rfc/db): make transparent database type names consistent
* doc(rfc/db): fix a bug in the Utxo.is_coinbase derivation
* fix(db): use the correct serialized size for OutputLocation
* doc(db): fix some comments
* refactor(db): split disk serialization types into their own module
* refactor(db): split the disk format into modules
* doc(db/test): explain the RON serialization format
* fix: check TEST_FAKE_ACTIVATION_HEIGHTS at runtime
* fix(tests): add TEST_FAKE_ACTIVATION_HEIGHTS variable
This variable ensures the test is activated in the `test-fake-activation-heights` step
* fix(docker): do not run specific tests by default in entrypoint.sh
* fix(test): remove extra TEST_FULL_SYNC argument
* imp(timeout): wait for an average build time
Co-authored-by: Gustavo Valverde <gustavo@iterativo.do>
* test(db): snapshot finalized tips
* test(db): snapshot block heights and hashes
* test(db): snapshot block data
* test(db): snapshot transaction hashes and data
* test(db): assert that blocks and transactions are in chain order
* clippy(db/test): cargo clippy --fix
* clippy(db/test): replace &Vec with &[_]
* refactor(state): split database access into modules by Zebra types
Also split the genesis block check from the genesis note commitment trees.
* test(db): snapshot column family names
* fix(db): assert that the default column family is empty on open and close
* feat(test): apply `cargo insta` settings to all tests
* doc(db/test): improve test docs
* test(db): snapshot column family data for the empty state
* refactor(db/test): split out the raw database snapshot
* test(db): snapshot raw database for blocks 0-2
* test(db): initial serialized snapshot data for raw database
* test(db): tweak snapshot file names
* test(db): rename snapshots for consistency
* test(db): store empty column families in a single snapshot
* test(db): simplify snapshot files by combining empty snapshots
* doc(db/test): put comment in a better place
* refactor(db): fastmod assert_default_is_empty assert_default_cf_is_empty
* doc(test): explain when insta settings are needed
* fix(state/test): use the network to initialize the state
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
Co-authored-by: Janito Vaqueiro Ferreira Filho <janito.vff@gmail.com>
* fix(state): mark DiskWriteBatch as must_use
* doc(state): add TODOs for moving database reads to blocking threads
* doc(state): minor comment tweaks
* refactor(state): split write batch into block, transactions, chain history
* refactor(state): split out a genesis block write method
* refactor(state): just use the empty note commitment trees directly
* refactor(state): split transaction writes into transparent, nullifiers, trees
And change DiskWriteBatch methods to take `&mut self`.
* refactor(state): split chain value pool writes out of history writes
* refactor(state): combine note commitment trees into an agrument struct
* refactor(state): split history and note commitment updates
* refactor(state): calculate current tip height and remove that argument
* refactor(state): move disk_db reads to a new zebra_db module
* refactor(state): make finalized value pool method names consistent
* refactor(state): split database writes into the zebra_db module
* refactor(state): move the block batch method to DiskWriteBatch
* refactor(state): actually add the zebra_db module
Unfortunately, I've lost the interim changes to this file,
so this commit might be the only one that compiles.
* refactor(state): add a newly created file to the cached state CI job
* refactor(state): move RocksDB-specific initialization to a new module
* refactor(state): move RocksDB-specific shutdown to a new module
* refactor(state): temporarily allow RocksDB-specific reads and writes, without a new module
Unlike the last few commits, this one actually compiles.
* refactor(state): add a DiskWriteBatch wrapper for RocksDB writes
* refactor(state): move finalized state test methods to a test module
* refactor(state): split the disk_format module
* refactor(ci): add the new disk_db file to the state CI list
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
* Add validation of ZIP-221 and ZIP-244 commitments
* Apply suggestions from code review
Co-authored-by: teor <teor@riseup.net>
* Add auth commitment check in the finalized state
* Reset the verifier when comitting to state fails
* Add explanation comment
* Add test with fake activation heights
* Add generate_valid_commitments flag
* Enable fake activation heights using env var instead of feature
* Also update initial_tip_hash; refactor into progress_from_tip()
* Improve comments
* Add fake activation heights test to CI
* Fix bug that caused commitment trees to not match when generating partial arbitrary chains
* Add ChainHistoryBlockTxAuthCommitmentHash::from_commitments to organize and deduplicate code
* Remove stale comment, improve readability
* Allow overriding with PROPTEST_CASES
* partial_chain_strategy(): don't update note commitment trees when not needed; add comment
Co-authored-by: teor <teor@riseup.net>
* Refactor HistoryTree into NonEmptyHistoryTree and HistoryTree
* HistoryTree: use Deref instead of AsRef; remove unneeded PartialEq
* ZIP-221: Validate chain history commitments in the non-finalized state (#2301)
* sketch of implementation
* refined implementation; still incomplete
* update librustzcash, change zcash_history to work with it
* simplified code per review; renamed MMR to HistoryTree
* expand HistoryTree implementation
* handle and propagate errors
* simplify check.rs tracing
* add suggested TODO
* add HistoryTree::prune
* fix bug in pruning
* fix compilation of tests; still need to make them pass
* Apply suggestions from code review
Co-authored-by: teor <teor@riseup.net>
* Apply suggestions from code review
Co-authored-by: teor <teor@riseup.net>
* improvements from code review
* improve check.rs comments and variable names
* fix HistoryTree which should use BTreeMap and not HashMap; fix non_finalized_state prop tests
* fix finalized_state proptest
* fix non_finalized_state tests by setting the correct commitments
* renamed mmr.rs to history_tree.rs
* Add HistoryTree struct
* expand non_finalized_state protest
* fix typo
* Add HistoryTree struct
* Update zebra-chain/src/primitives/zcash_history.rs
Co-authored-by: Deirdre Connolly <deirdre@zfnd.org>
* fix formatting
* Apply suggestions from code review
Co-authored-by: Deirdre Connolly <deirdre@zfnd.org>
* history_tree.rs: fixes from code review
* fixes to work with updated HistoryTree
* Improvements from code review
* Add Debug implementations to allow comparing Chains with proptest_assert_eq
Co-authored-by: teor <teor@riseup.net>
Co-authored-by: Deirdre Connolly <deirdre@zfnd.org>
* Apply suggestions from code review
Co-authored-by: teor <teor@riseup.net>
* Improvements from code review
* Restore blocks returned by PreparedChain since other tests broken; adjust tests with history trees
Co-authored-by: teor <teor@riseup.net>
Co-authored-by: Deirdre Connolly <deirdre@zfnd.org>
* add serialization for value balances
* change test names
* change panic messages
* add a deserialization test
* return the errors from `from_bytes()` methods
* add prop test for serialize/deserialize Amount
* Move amount proptests to amount::tests::prop
Co-authored-by: teor <teor@riseup.net>
* Add ZIP-221 history tree to finalized state
* Improve error / panic handling; improve documentation
* Return error again when preparing batch, fix expect messages
* Fix bug when pushing the Heartwood actiation block to the history tree
* Re-increase database version since it was increased in main
Co-authored-by: teor <teor@riseup.net>
* Tidy chain Cargo.toml
* Organize imports
* Add method to get note commitments from all Actions in Orchard shielded data
* Add method to get note commitments from all JoinSplits in Sprout JoinSplitData
* Add Request and Response variants for awaiting anchors
* Add anchors and note commitment trees to finalized state db
* Add (From|Into)Disk impls for tree::Roots and stubs for NoteCommitmentTrees
* Track anchors and note commitment trees in Chain
Append note commitments to their trees when doing update_chain_state_with,
then use the resulting Sapling and Orchard roots to pass to history_tree, and add
new roots to the anchor sets.
* Handle errors when appending to note commitment trees
* Add comments explaining why note commitment are not removed from the tree in revert_chain_state_with
* Implementing note commitments in finalized state
* Finish serialization of Orchard tree; remove old tree when updating finalize state
* Add serialization and finalized state updates for Sprout and Sapling trees
* Partially handle trees in non-finalized state. Use Option for trees in Chain
* Rebuild trees when forking; change finalized state tree getters to not require height
* Pass empty trees to tests; use empty trees by default in Chain
* Also rebuild anchor sets when forking
* Use empty tree as default in finalized state tree getters (for now)
* Use HashMultiSet for anchors in order to make pop_root() work correctly
* Reduce DEFAULT_PARTIAL_CHAIN_PROPTEST_CASES and MAX_PARTIAL_CHAIN_BLOCKS
* Reduce DEFAULT_PARTIAL_CHAIN_PROPTEST_CASES and MAX_PARTIAL_CHAIN_BLOCKS even more
* Apply suggestions from code review
* Add comments about order of note commitments and related methods/fields
* Don't use Option for trees
* Set DEFAULT_PARTIAL_CHAIN_PROPTEST_CASES=1 and restore MAX_PARTIAL_CHAIN_BLOCKS
* Remove unneeded anchor set rebuilding in fork()
* Improve proptest formatting
* Add missing comparisons to eq_internal_state
* Renamed sprout::tree::NoteCommitmentTree::hash() to root()
* Improve comments
* Add asserts, add issues to TODOs
* Remove impl Default for Chain since it was only used by tests
* Improve documentation and assertions; add tree serialization tests
* Remove Sprout code, which will be moved to another branch
* Add todo! in Sprout tree append()
* Remove stub request, response *Anchor* handling for now
* Add test for validating Sapling note commitment tree using test blocks
* Increase database version (new columns added for note commitment trees and anchors)
* Update test to make sure the order of sapling_note_commitments() is being tested
* Improve comments and structure of the test
* Improve variable names again
* Rustfmt
Co-authored-by: Deirdre Connolly <deirdre@zfnd.org>
Co-authored-by: Conrado P. L. Gouvea <conradoplg@gmail.com>
Co-authored-by: Conrado Gouvea <conrado@zfnd.org>
Co-authored-by: teor <teor@riseup.net>
* Add an OrderedUtxo type for validation of spends within a block
This change allows us to check that transparent spends use outputs from
earlier in their block. (But we don't actually do that check yet.)
We need to keep the order of UTXOs when we're contextually verifying
each new block that is added to a chain. But the block order is
irrelevant for UTXOs stored in the state.
* Take ownership in utxos_from_ordered_utxos
* Delete a confusing comment
* Reject duplicate sprout nullifiers in the state
* Improve docs and error messages
* Clarify "must be present" assert logs
* Move nullifier checks to their own module
Also:
* make non-finalized nullifier checks and errors generic over
sprout, sapling, and orchard
* create and update module and function documentation
* Fix a block type name in docs
* Move state assertions or skip them during tests
These changes enable state testing, while still asserting in production.
* Add sprout duplicate nullifier tests
* Improve comments
* Set value balance to 0 to pass future chain value pool checks
* Test finalized state in sprout nullifier accept test
* Replace assert with expect
* Improve assertion messages
* Document the new genesis transaction consensus rule
Zebra previously implemented this rule, but we documented it as a bug in
`zcashd`.
* Document the actual behaviour of zs_insert
* Use the block verifier and non-finalized state in the cached state tests
This substantially increases test coverage.
Previously, the cached state tests were configured with
`checkpoint_sync = true`, which only uses the checkpoint
verifier and the finalized state.
* Log the source of blocks in commit_finalized_direct
This lets us check that we're actually testing the non-finalized state
and block verifier in the cached state tests.
It also improves diagnostics for state errors.
* Fail cached state tests if they're using incorrect heights or configs
This makes sure that the cached state tests actually test the transition
from checkpoint to block verification, and the non-finalized state.
* add legacy chain check and tests
* improve has_network_upgrade check
* add docs to legacy_chain_check()
* change arbitrary module structure
* change the panic message
* move legacy chain acceptance into existing tests
* use a reduced_branch_id_strategy()
* add docs to strategy function
* add argument to check for legacy chain into sync_until()
* add nullifier methods to orchard
* store orchard nullifiers
* bump database version
* update `IntoDisk`
* support V5 in `UpdateWith`
* add a test for finalized state
* Use the latest network upgrade in state proptests
This commit changes the state system and database format to track the
provenance of UTXOs, in addition to the outputs themselves.
Specifically, it tracks the following additional metadata:
- the height at which the UTXO was created;
- whether or not the UTXO was created from a coinbase transaction or
not.
This metadata will allow us to:
- check the coinbase maturity consensus rule;
- check the coinbase inputs => no transparent outputs rule;
- implement lookup of transactions by utxo (using the height to find the
block and then scanning the block) for a future RPC mechanism.
Closes#1342
## Motivation
Prior to this PR we've been using `sled` as our database for storing persistent chain data on the disk between boots. We picked sled over rocksdb to minimize our c++ dependencies despite it being a less mature codebase. The theory was if it worked well enough we'd prefer to have a pure rust codebase, but if we ever ran into problems we knew we could easily swap it out with rocksdb.
Well, we ran into problems. Sled's memory usage was particularly high, and it seemed to be leaking memory. On top of all that, the performance for writes was pretty poor, causing us to become bottle-necked on sled instead of the network.
## Solution
This PR replaces `sled` with `rocksdb`. We've seen a 10x improvement in memory usage out of the box, no more leaking, and much better write performance. With this change writing chain data to disk is no longer a limiting factor in how quickly we can sync the chain.
The code in this pull request has:
- [x] Documentation Comments
- [x] Unit Tests and Property Tests
## Review
@hdevalence
teor
Conrado Gouvea
Marek
Dimitris Apostolou
Alfredo Garcia
Deirdre Connolly
Henry de Valence
Jane Lusby
Jane Lusby