Ms.mempool locking (#9050)

* Prority locking to consensus

* Remove pstats

* Linting

* Do some stuff outside of lock

* Fix startup

* Add log timings

* Try some different locking

* Add limit

* catch excp

* CLVM inside lock

* Try using a semaphore instead

* use events for lock queue

* test

* Add logging for message types

* type

* remove seed

* check new peak waiters

* correct FullNodeAPI self.full_node.new_peak._waiters typo

* correct logging string typos

* only warn about new_peak Waiters if there is at least 1

* remove no-longer-accepted parameter to FullNode.peak_post_processing()

* only warn about respond_transaction Waiters if there is at least 1

* lint

* Change some constants

* Small fix and logging changes

* Put message types outside

* Change some log levels so we can test with info

* More logging

* Increase rate limits but decrease paralelism

* tweaks

* Log dropped tx

* Fix pool rpc test

* Test fixes

* Mempool optimization

* Remove from seen if fails

* Increase queue sizes

* Message types info

* More test and logging

* Small changes to networking just in case

* Decrease logging

* Decrease logging even further

* Decrease logging even further even further

* Decrease logging 3

* Transaction queue

* Don't cancel tasks or close connection

* Cancel tasks on disconnect (for shutdown purposed)

* Fix typo

* Catch cancelled

* Do multiple at a time

* More accurate farmer response time

* More efficiently create tasks

* Increase queue size and priority by fee

* Revert priority

* Don't re-request too many times for dropped TX

* Handle cancelled error so we don't go into a bad state

* Catch cancelled in syncing tasks

* Reduce new_peak_sem to improve performance

* Less bytes conversion

* Missing file, and 2 workers for CLVM

* Validate BLS in a new thread

* tests

* Change semaphore constants

* correct a cancellation triggered exception and assertion

* Fix send_transaction, dont use BaseException, fix tests

* Fix more tests

* only log transaction handler cancellation in debug

* typing in log

* move unfinished validation to diff proc

* it is asyncio.CancelledError

* Add a test for bad signature

* Fix more tests, reduce logging, lint

* One more lint

* blockchain tests, pass bytes directly, single call

* Try to fix rl_wallet failures

* Fix mempool test

* catch everything

* Don't test RL wallet

* Fix more tests and return error code

* Improve error handling in multiprocess

* Add pre-validation time

* Add pre-validation time in logs, and revert pytest.ini changes

* Add log correctly

* Ms.bls cache experiment (#9115)

* Logging for cache

* Less logging

* Return to original plan

* Clean up

* Remove coment

* Remove log

* formalize LockQueue shutdown

* Comments

* Fix blockchain test

* Improve cache

* Remove logs

* Fix sign_coin_spends

* Fix pool wallet

Co-authored-by: Kyle Altendorf <sda@fstab.net>
Co-authored-by: Yostra <straya@chia.net>

chia/clvm/spend_sim.py

@@ -7,7 +7,7 @@
 from chia.types.blockchain_format.program import Program, SerializedProgram
 from chia.util.ints import uint64, uint32
 from chia.util.hash import std_hash
-from chia.util.errors import Err
+from chia.util.errors import Err, ValidationError
 from chia.util.db_wrapper import DBWrapper
 from chia.types.coin_record import CoinRecord
 from chia.types.spend_bundle import SpendBundle
@@ -49,6 +49,7 @@ def __init__(self, rci: List[Coin], height: uint32, timestamp: uint64):
         self.timestamp = timestamp
         self.is_transaction_block = True
         self.header_hash = std_hash(bytes(height))
+        self.prev_transaction_block_hash = std_hash(std_hash(height))
 
 
 class SpendSim:
@@ -78,7 +79,7 @@ async def close(self):
         await self.connection.close()
 
     async def new_peak(self):
-        await self.mempool_manager.new_peak(self.block_records[-1])
+        await self.mempool_manager.new_peak(self.block_records[-1], [])
 
     def new_coin_record(self, coin: Coin, coinbase=False) -> CoinRecord:
         return CoinRecord(
@@ -203,9 +204,12 @@ def __init__(self, service):
         self.service = service
 
     async def push_tx(self, spend_bundle: SpendBundle) -> Tuple[MempoolInclusionStatus, Optional[Err]]:
-        cost_result: NPCResult = await self.service.mempool_manager.pre_validate_spendbundle(
-            spend_bundle, spend_bundle.name()
-        )
+        try:
+            cost_result: NPCResult = await self.service.mempool_manager.pre_validate_spendbundle(
+                spend_bundle, None, spend_bundle.name()
+            )
+        except ValidationError as e:
+            return MempoolInclusionStatus.FAILED, e.code
         cost, status, error = await self.service.mempool_manager.add_spendbundle(
             spend_bundle, cost_result, spend_bundle.name()
         )

chia/consensus/block_body_validation.py

@@ -50,6 +50,7 @@ async def validate_block_body(
     npc_result: Optional[NPCResult],
     fork_point_with_peak: Optional[uint32],
     get_block_generator: Callable,
+    validate_signature=True,
 ) -> Tuple[Optional[Err], Optional[NPCResult]]:
     """
     This assumes the header block has been completely validated.
@@ -457,10 +458,11 @@ async def validate_block_body(
     # However, we force caching of pairings just for unfinished blocks
     # as the cache is likely to be useful when validating the corresponding
     # finished blocks later.
-    force_cache: bool = isinstance(block, UnfinishedBlock)
-    if not cached_bls.aggregate_verify(
-        pairs_pks, pairs_msgs, block.transactions_info.aggregated_signature, force_cache
-    ):
-        return Err.BAD_AGGREGATE_SIGNATURE, None
+    if validate_signature:
+        force_cache: bool = isinstance(block, UnfinishedBlock)
+        if not cached_bls.aggregate_verify(
+            pairs_pks, pairs_msgs, block.transactions_info.aggregated_signature, force_cache
+        ):
+            return Err.BAD_AGGREGATE_SIGNATURE, None
 
     return None, npc_result

chia/consensus/blockchain.py

@@ -17,7 +17,11 @@
 from chia.consensus.difficulty_adjustment import get_next_sub_slot_iters_and_difficulty
 from chia.consensus.find_fork_point import find_fork_point_in_chain
 from chia.consensus.full_block_to_block_record import block_to_block_record
-from chia.consensus.multiprocess_validation import PreValidationResult, pre_validate_blocks_multiprocessing
+from chia.consensus.multiprocess_validation import (
+    PreValidationResult,
+    pre_validate_blocks_multiprocessing,
+    _run_generator,
+)
 from chia.full_node.block_store import BlockStore
 from chia.full_node.coin_store import CoinStore
 from chia.full_node.hint_store import HintStore
@@ -35,7 +39,7 @@
 from chia.types.unfinished_block import UnfinishedBlock
 from chia.types.unfinished_header_block import UnfinishedHeaderBlock
 from chia.types.weight_proof import SubEpochChallengeSegment
-from chia.util.errors import Err
+from chia.util.errors import Err, ConsensusError
 from chia.util.generator_tools import get_block_header, tx_removals_and_additions
 from chia.util.ints import uint16, uint32, uint64, uint128
 from chia.util.streamable import recurse_jsonify
@@ -561,7 +565,7 @@ def get_recent_reward_challenges(self) -> List[Tuple[bytes32, uint128]]:
         return list(reversed(recent_rc))
 
     async def validate_unfinished_block(
-        self, block: UnfinishedBlock, skip_overflow_ss_validation=True
+        self, block: UnfinishedBlock, npc_result: Optional[NPCResult], skip_overflow_ss_validation=True
     ) -> PreValidationResult:
         if (
             not self.contains_block(block.prev_header_hash)
@@ -601,21 +605,6 @@ async def validate_unfinished_block(
             else self.block_record(block.prev_header_hash).height
         )
 
-        npc_result = None
-        if block.transactions_generator is not None:
-            assert block.transactions_info is not None
-            try:
-                block_generator: Optional[BlockGenerator] = await self.get_block_generator(block)
-            except ValueError:
-                return PreValidationResult(uint16(Err.GENERATOR_REF_HAS_NO_GENERATOR.value), None, None)
-            if block_generator is None:
-                return PreValidationResult(uint16(Err.GENERATOR_REF_HAS_NO_GENERATOR.value), None, None)
-            npc_result = get_name_puzzle_conditions(
-                block_generator,
-                min(self.constants.MAX_BLOCK_COST_CLVM, block.transactions_info.cost),
-                cost_per_byte=self.constants.COST_PER_BYTE,
-                safe_mode=False,
-            )
         error_code, cost_result = await validate_block_body(
             self.constants,
             self,
@@ -627,6 +616,7 @@ async def validate_unfinished_block(
             npc_result,
             None,
             self.get_block_generator,
+            False,
         )
 
         if error_code is not None:
@@ -654,6 +644,21 @@ async def pre_validate_blocks_multiprocessing(
             wp_summaries,
         )
 
+    async def run_generator(self, unfinished_block: bytes, generator: BlockGenerator) -> NPCResult:
+        task = asyncio.get_running_loop().run_in_executor(
+            self.pool,
+            _run_generator,
+            self.constants_json,
+            unfinished_block,
+            bytes(generator),
+        )
+        error, npc_result_bytes = await task
+        if error is not None:
+            raise ConsensusError(error)
+        if npc_result_bytes is None:
+            raise ConsensusError(Err.UNKNOWN)
+        return NPCResult.from_bytes(npc_result_bytes)
+
     def contains_block(self, header_hash: bytes32) -> bool:
         """
         True if we have already added this block to the chain. This may return false for orphan blocks

chia/consensus/multiprocess_validation.py

@@ -21,8 +21,9 @@
 from chia.types.full_block import FullBlock
 from chia.types.generator_types import BlockGenerator
 from chia.types.header_block import HeaderBlock
+from chia.types.unfinished_block import UnfinishedBlock
 from chia.util.block_cache import BlockCache
-from chia.util.errors import Err
+from chia.util.errors import Err, ValidationError
 from chia.util.generator_tools import get_block_header, tx_removals_and_additions
 from chia.util.ints import uint16, uint64, uint32
 from chia.util.streamable import Streamable, dataclass_from_dict, streamable
@@ -316,3 +317,34 @@ async def pre_validate_blocks_multiprocessing(
         for batch_result in (await asyncio.gather(*futures))
         for result in batch_result
     ]
+
+
+def _run_generator(
+    constants_dict: bytes,
+    unfinished_block_bytes: bytes,
+    block_generator_bytes: bytes,
+) -> Tuple[Optional[Err], Optional[bytes]]:
+    """
+    Runs the CLVM generator from bytes inputs. This is meant to be called under a ProcessPoolExecutor, in order to
+    validate the heavy parts of a block (clvm program) in a different process.
+    """
+    try:
+        constants: ConsensusConstants = dataclass_from_dict(ConsensusConstants, constants_dict)
+        unfinished_block: UnfinishedBlock = UnfinishedBlock.from_bytes(unfinished_block_bytes)
+        assert unfinished_block.transactions_info is not None
+        block_generator: BlockGenerator = BlockGenerator.from_bytes(block_generator_bytes)
+        assert block_generator.program == unfinished_block.transactions_generator
+        npc_result: NPCResult = get_name_puzzle_conditions(
+            block_generator,
+            min(constants.MAX_BLOCK_COST_CLVM, unfinished_block.transactions_info.cost),
+            cost_per_byte=constants.COST_PER_BYTE,
+            safe_mode=False,
+        )
+        if npc_result.error is not None:
+            return Err(npc_result.error), None
+    except ValidationError as e:
+        return e.code, None
+    except Exception:
+        return Err.UNKNOWN, None
+
+    return None, bytes(npc_result)