def test_finish_same_task_twice():
ti = OrderedTaskPreparation(TwoPrereqs, identity, lambda x: x - 1)
ti.set_finished_dependency(1)
ti.register_tasks((2, ))
ti.finish_prereq(TwoPrereqs.PREREQ1, (2, ))
with pytest.raises(ValidationError):
ti.finish_prereq(TwoPrereqs.PREREQ1, (2, ))
async def test_wait_if_too_many_ready_tasks():
ti = OrderedTaskPreparation(OnePrereq,
identity,
lambda x: x - 1,
max_tasks=1)
ti.set_finished_dependency(3)
ti.register_tasks((4, ))
# This should raise a timeout error because it gets locked waiting for the task to be finished
with pytest.raises(asyncio.TimeoutError):
await wait(ti.wait_add_tasks((5, )))
ti.finish_prereq(OnePrereq.ONE, (4, ))
# This should raise a timeout error because it gets locked waiting for the task to be picked up
with pytest.raises(asyncio.TimeoutError):
await wait(ti.wait_add_tasks((5, )))
completed = await wait(ti.ready_tasks())
assert completed == (4, )
# Now we can add the other task
await wait(ti.wait_add_tasks((5, )))
# ... and finish & pick up the completed one
ti.finish_prereq(OnePrereq.ONE, (5, ))
completed = await wait(ti.ready_tasks())
assert completed == (5, )
async def test_pruning():
# make a number task depend on the mod10, so 4 and 14 both depend on task 3
ti = OrderedTaskPreparation(OnePrereq, identity, lambda x: (x % 10) - 1, max_depth=2)
ti.set_finished_dependency(3)
ti.register_tasks((4, 5, 6))
ti.finish_prereq(OnePrereq.one, (4, 5, 6))
# it's fine to prepare a task that depends up to two back in history
# this depends on 5
ti.register_tasks((16, ))
# this depends on 4
ti.register_tasks((15, ))
# but depending 3 back in history should raise a validation error, because it's pruned
with pytest.raises(MissingDependency):
# this depends on 3
ti.register_tasks((14, ))
# test the same concept, but after pruning more than just the starting task...
ti.register_tasks((7, ))
ti.finish_prereq(OnePrereq.one, (7, ))
ti.register_tasks((26, ))
ti.register_tasks((27, ))
with pytest.raises(MissingDependency):
ti.register_tasks((25, ))
async def test_simplest_path():
ti = OrderedTaskPreparation(TwoPrereqs, identity, lambda x: x - 1)
ti.set_finished_dependency(3)
ti.register_tasks((4, ))
ti.finish_prereq(TwoPrereqs.Prereq1, (4, ))
ti.finish_prereq(TwoPrereqs.Prereq2, (4, ))
ready = await wait(ti.ready_tasks())
assert ready == (4, )
async def test_two_steps_simultaneous_complete():
ti = OrderedTaskPreparation(OnePrereq, identity, lambda x: x - 1)
ti.set_finished_dependency(3)
ti.register_tasks((4, 5))
ti.finish_prereq(OnePrereq.one, (4, ))
ti.finish_prereq(OnePrereq.one, (5, ))
completed = await wait(ti.ready_tasks())
assert completed == (4, 5)
async def test_register_out_of_order():
ti = OrderedTaskPreparation(OnePrereq, identity, lambda x: x - 1, accept_dangling_tasks=True)
ti.set_finished_dependency(1)
ti.register_tasks((4, 5))
ti.finish_prereq(OnePrereq.one, (4, 5))
await assert_nothing_ready(ti)
ti.register_tasks((2, 3))
ti.finish_prereq(OnePrereq.one, (2, 3))
finished = await wait(ti.ready_tasks())
assert finished == (2, 3, 4, 5)
async def test_register_out_of_order():
ti = OrderedTaskPreparation(OnePrereq, identity, lambda x: x - 1, accept_dangling_tasks=True)
ti.set_finished_dependency(1)
ti.register_tasks((4, 5))
ti.finish_prereq(OnePrereq.one, (4, 5))
try:
finished = await wait(ti.ready_tasks())
except asyncio.TimeoutError:
pass
else:
assert False, f"No steps should be ready, but got {finished!r}"
ti.register_tasks((2, 3))
ti.finish_prereq(OnePrereq.one, (2, 3))
finished = await wait(ti.ready_tasks())
assert finished == (2, 3, 4, 5)
async def test_finish_different_entry_at_same_step():
def previous_even_number(num):
return ((num - 1) // 2) * 2
ti = OrderedTaskPreparation(OnePrereq, identity, previous_even_number)
ti.set_finished_dependency(2)
ti.register_tasks((3, 4))
# depends on 2
ti.finish_prereq(OnePrereq.one, (3, ))
# also depends on 2
ti.finish_prereq(OnePrereq.one, (4, ))
completed = await wait(ti.ready_tasks())
assert completed == (3, 4)
async def test_return_original_entry():
# for no particular reason, the id is 3 before the number
ti = OrderedTaskPreparation(OnePrereq, lambda x: x - 3, lambda x: x - 4)
# translates to id -1
ti.set_finished_dependency(2)
ti.register_tasks((3, 4))
# translates to id 0
ti.finish_prereq(OnePrereq.one, (3, ))
# translates to id 1
ti.finish_prereq(OnePrereq.one, (4, ))
entries = await wait(ti.ready_tasks())
# make sure that the original task is returned, not the id
assert entries == (3, 4)
async def test_finished_dependency_midstream():
"""
We need to be able to mark dependencies as finished, after task completion
"""
ti = OrderedTaskPreparation(TwoPrereqs, identity, lambda x: x - 1)
ti.set_finished_dependency(3)
ti.register_tasks((4, ))
ti.finish_prereq(TwoPrereqs.Prereq1, (4, ))
ti.finish_prereq(TwoPrereqs.Prereq2, (4, ))
ready = await wait(ti.ready_tasks())
assert ready == (4, )
# now start in a discontinuous series of tasks
with pytest.raises(MissingDependency):
ti.register_tasks((6, ))
ti.set_finished_dependency(5)
ti.register_tasks((6, ))
ti.finish_prereq(TwoPrereqs.Prereq1, (6, ))
ti.finish_prereq(TwoPrereqs.Prereq2, (6, ))
ready = await wait(ti.ready_tasks())
assert ready == (6, )
async def test_pruning():
# make a number task depend on the mod10, so 4 and 14 both depend on task 3
ti = OrderedTaskPreparation(OnePrereq,
identity,
lambda x: (x % 10) - 1,
max_depth=2)
ti.set_finished_dependency(3)
ti.register_tasks((4, 5, 6, 7, 8))
ti.finish_prereq(OnePrereq.one, (4, 5, 6))
# trigger pruning by requesting the ready tasks through 6, then "finishing" them
# by requesting the next batch of ready tasks (7)
completed = await wait(ti.ready_tasks())
assert completed == (4, 5, 6)
ti.finish_prereq(OnePrereq.one, (7, ))
completed = await wait(ti.ready_tasks())
assert completed == (7, )
# it's fine to prepare a task that depends up to two back in history
# this depends on 5
ti.register_tasks((16, ))
# this depends on 4
ti.register_tasks((15, ))
# but depending 3 back in history should raise a validation error, because it's pruned
with pytest.raises(MissingDependency):
# this depends on 3
ti.register_tasks((14, ))
# test the same concept, but after pruning tasks that weren't the starting tasks
# trigger pruning from the head at 7 by completing the one *after* 7
ti.finish_prereq(OnePrereq.one, (8, ))
completed = await wait(ti.ready_tasks())
assert completed == (8, )
ti.register_tasks((26, ))
ti.register_tasks((27, ))
with pytest.raises(MissingDependency):
ti.register_tasks((25, ))
def test_finish_with_unrecognized_task():
ti = OrderedTaskPreparation(TwoPrereqs, identity, lambda x: x - 1)
ti.set_finished_dependency(1)
with pytest.raises(ValidationError):
ti.finish_prereq('UNRECOGNIZED_TASK', (2, ))
async def test_cannot_finish_before_prepare():
ti = OrderedTaskPreparation(TwoPrereqs, identity, lambda x: x - 1)
ti.set_finished_dependency(3)
with pytest.raises(ValidationError):
ti.finish_prereq(TwoPrereqs.Prereq1, (4, ))
class RegularChainBodySyncer(BaseBodyChainSyncer):
"""
Sync with the Ethereum network by fetching block headers/bodies and importing them.
Here, the run() method will execute the sync loop forever, until our CancelToken is triggered.
"""
def __init__(self,
chain: AsyncChainAPI,
db: BaseAsyncChainDB,
peer_pool: ETHPeerPool,
header_syncer: HeaderSyncerAPI,
block_importer: BaseBlockImporter,
token: CancelToken = None) -> None:
super().__init__(chain, db, peer_pool, header_syncer, token)
# track when block bodies are downloaded, so that blocks can be imported
self._block_import_tracker = OrderedTaskPreparation(
BlockImportPrereqs,
id_extractor=attrgetter('hash'),
# make sure that a block is not imported until the parent block is imported
dependency_extractor=attrgetter('parent_hash'),
# Avoid problems by keeping twice as much data as the import queue size
max_depth=BLOCK_IMPORT_QUEUE_SIZE * 2,
)
self._block_importer = block_importer
# Track if any headers have been received yet
self._got_first_header = asyncio.Event()
# Rate limit the block import logs
self._import_log_limiter = TokenBucket(
0.33, # show about one log per 3 seconds
5, # burst up to 5 logs after a lag
)
# the queue of blocks that are downloaded and ready to be imported
self._import_queue: 'asyncio.Queue[BlockAPI]' = asyncio.Queue(BLOCK_IMPORT_QUEUE_SIZE)
self._import_active = asyncio.Lock()
async def _run(self) -> None:
head = await self.wait(self.db.coro_get_canonical_head())
self._block_import_tracker.set_finished_dependency(head)
self.run_daemon_task(self._launch_prerequisite_tasks())
self.run_daemon_task(self._assign_body_download_to_peers())
self.run_daemon_task(self._import_ready_blocks())
self.run_daemon_task(self._preview_ready_blocks())
self.run_daemon_task(self._display_stats())
await super()._run()
def register_peer(self, peer: BasePeer) -> None:
# when a new peer is added to the pool, add it to the idle peer list
super().register_peer(peer)
self._body_peers.put_nowait(cast(ETHPeer, peer))
async def _should_skip_header(self, header: BlockHeaderAPI) -> bool:
"""
Should we skip trying to import this header?
Return True if the syncing of header appears to be complete.
This is fairly relaxed about the definition, preferring speed over slow precision.
"""
return await self.db.coro_exists(header.state_root)
async def _launch_prerequisite_tasks(self) -> None:
"""
Watch for new headers to be added to the queue, and add the prerequisite
tasks (downloading block bodies) as they become available.
"""
async for headers in self._sync_from_headers(
self._block_import_tracker,
self._should_skip_header):
# Sometimes duplicates are added to the queue, when switching from one sync to another.
# We can simply ignore them.
new_headers = tuple(h for h in headers if h not in self._block_body_tasks)
# if the output queue gets full, hang until there is room
await self.wait(self._block_body_tasks.add(new_headers))
def _mark_body_download_complete(
self,
batch_id: int,
completed_headers: Sequence[BlockHeaderAPI]) -> None:
super()._mark_body_download_complete(batch_id, completed_headers)
self._block_import_tracker.finish_prereq(
BlockImportPrereqs.STORE_BLOCK_BODIES,
completed_headers,
)
async def _preview_ready_blocks(self) -> None:
"""
Wait for block bodies to be downloaded, then compile the blocks and
preview them to the importer.
It's important to do this in a separate step from importing so that
previewing can get ahead of import by a few blocks.
"""
await self.wait(self._got_first_header.wait())
while self.is_operational:
# This tracker waits for all prerequisites to be complete, and returns headers in
# order, so that each header's parent is already persisted.
get_ready_coro = self._block_import_tracker.ready_tasks(1)
completed_headers = await self.wait(get_ready_coro)
if self._block_import_tracker.has_ready_tasks():
# Even after clearing out a big batch, there is no available capacity, so
# pause any coroutines that might wait for capacity
self._db_buffer_capacity.clear()
else:
# There is available capacity, let any waiting coroutines continue
self._db_buffer_capacity.set()
header = completed_headers[0]
block = self._header_to_block(header)
# Put block in short queue for import, wait here if queue is full
await self.wait(self._import_queue.put(block))
# Load the state root of the parent header
try:
parent_state_root = self._block_hash_to_state_root[header.parent_hash]
except KeyError:
# For the very first header that we load, we have to look up the parent's
# state from the database:
parent = await self.chain.coro_get_block_header_by_hash(header.parent_hash)
parent_state_root = parent.state_root
# Emit block for preview
# - look up the addresses referenced by the transaction (eg~ sender and recipient)
# - execute the block ahead of time to start collecting any missing state
# - store the header (for future evm execution that might look up old block hashes)
await self._block_importer.preview_transactions(
header,
block.transactions,
parent_state_root,
)
async def _import_ready_blocks(self) -> None:
"""
Wait for block bodies to be downloaded, then compile the blocks and
preview them to the importer.
"""
await self.wait(self._got_first_header.wait())
while self.is_operational:
if self._import_queue.empty():
if self._import_active.locked():
self._import_active.release()
waiting_for_next_block = Timer()
block = await self.wait(self._import_queue.get())
if not self._import_active.locked():
self.logger.info(
"Waited %.1fs for %s body",
waiting_for_next_block.elapsed,
block.header,
)
await self._import_active.acquire()
await self._import_block(block)
async def _import_block(self, block: BlockAPI) -> None:
timer = Timer()
_, new_canonical_blocks, old_canonical_blocks = await self.wait(
self._block_importer.import_block(block)
)
# how much is the imported block's header behind the current time?
lag = time.time() - block.header.timestamp
humanized_lag = humanize_seconds(lag)
if new_canonical_blocks == (block,):
# simple import of a single new block.
# decide whether to log to info or debug, based on log rate
if self._import_log_limiter.can_take(1):
log_fn = self.logger.info
self._import_log_limiter.take_nowait(1)
else:
log_fn = self.logger.debug
log_fn(
"Imported block %d (%d txs) in %.2f seconds, with %s lag",
block.number,
len(block.transactions),
timer.elapsed,
humanized_lag,
)
elif not new_canonical_blocks:
# imported block from a fork.
self.logger.info(
"Imported non-canonical block %d (%d txs) in %.2f seconds, with %s lag",
block.number,
len(block.transactions),
timer.elapsed,
humanized_lag,
)
elif old_canonical_blocks:
self.logger.info(
"Chain Reorganization: Imported block %d (%d txs) in %.2f seconds, "
"%d blocks discarded and %d new canonical blocks added, with %s lag",
block.number,
len(block.transactions),
timer.elapsed,
len(old_canonical_blocks),
len(new_canonical_blocks),
humanized_lag,
)
else:
raise Exception("Invariant: unreachable code path")
def _header_to_block(self, header: BlockHeaderAPI) -> BlockAPI:
"""
This method converts a header that was queued up for sync into its full block
representation. It may not be called until after the body is marked as fully
downloaded, as tracked by self._block_import_tracker.
"""
vm_class = self.chain.get_vm_class(header)
block_class = vm_class.get_block_class()
if _is_body_empty(header):
transactions: List[SignedTransactionAPI] = []
uncles: List[BlockHeaderAPI] = []
else:
body = self._pending_bodies.pop(header)
tx_class = block_class.get_transaction_class()
transactions = [tx_class.from_base_transaction(tx)
for tx in body.transactions]
uncles = body.uncles
return block_class(header, transactions, uncles)
async def _display_stats(self) -> None:
self.logger.debug("Regular sync waiting for first header to arrive")
await self.wait(self._got_first_header.wait())
self.logger.debug("Regular sync first header arrived")
while self.is_operational:
await self.sleep(5)
self.logger.debug(
"(progress, queued, max) of bodies, receipts: %r. Write capacity? %s Importing? %s",
[(q.num_in_progress(), len(q), q._maxsize) for q in (
self._block_body_tasks,
)],
self._db_buffer_capacity.is_set(),
self._import_active.locked(),
)
class RegularChainBodySyncer(BaseBodyChainSyncer):
"""
Sync with the Ethereum network by fetching block headers/bodies and importing them.
Here, the run() method will execute the sync loop forever, until our CancelToken is triggered.
"""
def __init__(self,
chain: BaseAsyncChain,
db: BaseAsyncChainDB,
peer_pool: ETHPeerPool,
header_syncer: HeaderSyncerAPI,
block_importer: BaseBlockImporter,
token: CancelToken = None) -> None:
super().__init__(chain, db, peer_pool, header_syncer, token)
# track when block bodies are downloaded, so that blocks can be imported
self._block_import_tracker = OrderedTaskPreparation(
BlockImportPrereqs,
id_extractor=attrgetter('hash'),
# make sure that a block is not imported until the parent block is imported
dependency_extractor=attrgetter('parent_hash'),
)
self._block_importer = block_importer
# Track if any headers have been received yet
self._got_first_header = asyncio.Event()
async def _run(self) -> None:
head = await self.wait(self.db.coro_get_canonical_head())
self._block_import_tracker.set_finished_dependency(head)
self.run_daemon_task(self._launch_prerequisite_tasks())
self.run_daemon_task(self._assign_body_download_to_peers())
self.run_daemon_task(self._import_ready_blocks())
self.run_daemon_task(self._display_stats())
await super()._run()
def register_peer(self, peer: BasePeer) -> None:
# when a new peer is added to the pool, add it to the idle peer list
super().register_peer(peer)
self._body_peers.put_nowait(cast(ETHPeer, peer))
async def _should_skip_header(self, header: BlockHeader) -> bool:
"""
Should we skip trying to import this header?
Return True if the syncing of header appears to be complete.
This is fairly relaxed about the definition, preferring speed over slow precision.
"""
return await self.db.coro_exists(header.state_root)
async def _launch_prerequisite_tasks(self) -> None:
"""
Watch for new headers to be added to the queue, and add the prerequisite
tasks (downloading block bodies) as they become available.
"""
async for headers in self._sync_from_headers(
self._block_import_tracker, self._should_skip_header):
# Sometimes duplicates are added to the queue, when switching from one sync to another.
# We can simply ignore them.
new_headers = tuple(h for h in headers
if h not in self._block_body_tasks)
# if the output queue gets full, hang until there is room
await self.wait(self._block_body_tasks.add(new_headers))
def _mark_body_download_complete(
self, batch_id: int, completed_headers: Tuple[BlockHeader,
...]) -> None:
super()._mark_body_download_complete(batch_id, completed_headers)
self._block_import_tracker.finish_prereq(
BlockImportPrereqs.StoreBlockBodies,
completed_headers,
)
async def _import_ready_blocks(self) -> None:
"""
Wait for block bodies to be downloaded, then import the blocks.
"""
await self.wait(self._got_first_header.wait())
while self.is_operational:
timer = Timer()
# wait for block bodies to become ready for execution
completed_headers = await self.wait(
self._block_import_tracker.ready_tasks())
if self._block_import_tracker.has_ready_tasks():
# Even after clearing out a big batch, there is no available capacity, so
# pause any coroutines that might wait for capacity
self._db_buffer_capacity.clear()
else:
# There is available capacity, let any waiting coroutines continue
self._db_buffer_capacity.set()
await self._import_blocks(completed_headers)
head = await self.wait(self.db.coro_get_canonical_head())
self.logger.info(
"Synced chain segment with %d blocks in %.2f seconds, new head: %s",
len(completed_headers),
timer.elapsed,
head,
)
async def _import_blocks(self, headers: Tuple[BlockHeader, ...]) -> None:
"""
Import the blocks for the corresponding headers
:param headers: headers that have the block bodies downloaded
"""
unimported_blocks = self._headers_to_blocks(headers)
for block in unimported_blocks:
timer = Timer()
_, new_canonical_blocks, old_canonical_blocks = await self.wait(
self._block_importer.import_block(block))
if new_canonical_blocks == (block, ):
# simple import of a single new block.
self.logger.info("Imported block %d (%d txs) in %.2f seconds",
block.number, len(block.transactions),
timer.elapsed)
elif not new_canonical_blocks:
# imported block from a fork.
self.logger.info(
"Imported non-canonical block %d (%d txs) in %.2f seconds",
block.number, len(block.transactions), timer.elapsed)
elif old_canonical_blocks:
self.logger.info(
"Chain Reorganization: Imported block %d (%d txs) in %.2f "
"seconds, %d blocks discarded and %d new canonical blocks added",
block.number,
len(block.transactions),
timer.elapsed,
len(old_canonical_blocks),
len(new_canonical_blocks),
)
else:
raise Exception("Invariant: unreachable code path")
@to_tuple
def _headers_to_blocks(
self, headers: Iterable[BlockHeader]) -> Iterable[BaseBlock]:
for header in headers:
vm_class = self.chain.get_vm_class(header)
block_class = vm_class.get_block_class()
if _is_body_empty(header):
transactions: List[BaseTransaction] = []
uncles: List[BlockHeader] = []
else:
body = self._pending_bodies.pop(header)
tx_class = block_class.get_transaction_class()
transactions = [
tx_class.from_base_transaction(tx)
for tx in body.transactions
]
uncles = body.uncles
yield block_class(header, transactions, uncles)
async def _display_stats(self) -> None:
self.logger.debug("Regular sync waiting for first header to arrive")
await self.wait(self._got_first_header.wait())
self.logger.debug("Regular sync first header arrived")
while self.is_operational:
await self.sleep(5)
self.logger.debug(
"(in progress, queued, max size) of bodies, receipts: %r. Write capacity? %s",
[(q.num_in_progress(), len(q), q._maxsize)
for q in (self._block_body_tasks, )],
"yes" if self._db_buffer_capacity.is_set() else "no",
)
async def test_forked_pruning_dangling():
ti = OrderedTaskPreparation(
OnePrereq,
task_id,
fork_prereq,
max_depth=2,
accept_dangling_tasks=True,
)
ti.set_finished_dependency(Task(0, 0, 0))
ti.register_tasks((
Task(2, 0, 0),
Task(2, 1, 0),
))
ti.finish_prereq(OnePrereq.one, (
Task(2, 0, 0),
Task(2, 1, 0),
))
ti.register_tasks((
Task(3, 0, 0),
Task(3, 1, 1),
))
ti.finish_prereq(OnePrereq.one, (
Task(3, 0, 0),
Task(3, 1, 1),
))
ti.register_tasks((
Task(4, 0, 0),
Task(4, 1, 1),
))
ti.finish_prereq(OnePrereq.one, (
Task(4, 0, 0),
Task(4, 1, 1),
))
ti.register_tasks((
Task(5, 0, 0),
Task(6, 0, 0),
Task(7, 0, 0),
Task(8, 0, 0),
Task(9, 0, 0),
Task(10, 0, 0),
))
ti.finish_prereq(OnePrereq.one, (
Task(5, 0, 0),
Task(6, 0, 0),
Task(7, 0, 0),
Task(8, 0, 0),
Task(9, 0, 0),
Task(10, 0, 0),
))
ti.register_tasks((Task(5, 1, 1), ))
ti.finish_prereq(OnePrereq.one, (Task(5, 1, 1), ))
ti.register_tasks((Task(1, 0, 0), ))
# nothing is ready, because the prereq on the first task is incomplete
await assert_nothing_ready(ti)
ti.finish_prereq(OnePrereq.one, (Task(1, 0, 0), ))
ti.register_tasks((
Task(11, 0, 0),
Task(12, 0, 0),
Task(13, 0, 0),
Task(14, 0, 0),
Task(15, 0, 0),
Task(16, 0, 0),
Task(17, 0, 0),
Task(18, 0, 0),
Task(19, 0, 0),
Task(20, 0, 0),
Task(6, 1, 1),
Task(7, 1, 1),
Task(8, 1, 1),
Task(9, 1, 1),
))
ti.finish_prereq(OnePrereq.one, (
Task(11, 0, 0),
Task(12, 0, 0),
Task(13, 0, 0),
Task(14, 0, 0),
Task(15, 0, 0),
Task(16, 0, 0),
Task(17, 0, 0),
Task(18, 0, 0),
Task(19, 0, 0),
Task(20, 0, 0),
Task(6, 1, 1),
Task(7, 1, 1),
Task(8, 1, 1),
Task(9, 1, 1),
))
finished = await ti.ready_tasks()
assert len(finished) == 28
# nothing should be pruned, because caller is still "acting" on all tasks
assert TaskID(1, 0) in ti._tasks
# caller indicates that she is done working on tasks by calling ready_tasks() again
await assert_nothing_ready(ti)
# find the tips of the chains that have and have not been pruned
assert TaskID(6, 1) not in ti._tasks
assert TaskID(7, 1) in ti._tasks
assert TaskID(17, 0) not in ti._tasks
assert TaskID(18, 0) in ti._tasks
class FastChainBodySyncer(BaseBodyChainSyncer):
"""
Sync with the Ethereum network by fetching block headers/bodies and storing them in our DB.
Here, the run() method returns as soon as we complete a sync with the peer that announced the
highest TD, at which point we must run the StateDownloader to fetch the state for our chain
head.
"""
def __init__(self,
chain: BaseAsyncChain,
db: BaseAsyncChainDB,
peer_pool: ETHPeerPool,
header_syncer: HeaderSyncerAPI,
token: CancelToken = None) -> None:
super().__init__(chain, db, peer_pool, token)
# queue up any idle peers, in order of how fast they return receipts
self._receipt_peers: WaitingPeers[ETHPeer] = WaitingPeers(
commands.Receipts)
self._header_syncer = header_syncer
# Track receipt download tasks
# - arbitrarily allow several requests-worth of headers queued up
# - try to get receipts from lower block numbers first
buffer_size = MAX_RECEIPTS_FETCH * REQUEST_BUFFER_MULTIPLIER
self._receipt_tasks = TaskQueue(buffer_size,
attrgetter('block_number'))
# track when both bodies and receipts are collected, so that blocks can be persisted
self._block_persist_tracker = OrderedTaskPreparation(
BlockPersistPrereqs,
id_extractor=attrgetter('hash'),
# make sure that a block is not persisted until the parent block is persisted
dependency_extractor=attrgetter('parent_hash'),
)
# Track whether the fast chain syncer completed its goal
self.is_complete = False
async def _run(self) -> None:
head = await self.wait(self.db.coro_get_canonical_head())
self.tracker = ChainSyncPerformanceTracker(head)
self._block_persist_tracker.set_finished_dependency(head)
self.run_daemon_task(self._launch_prerequisite_tasks())
self.run_daemon_task(self._assign_receipt_download_to_peers())
self.run_daemon_task(self._assign_body_download_to_peers())
self.run_daemon_task(self._persist_ready_blocks())
self.run_daemon_task(self._display_stats())
await super()._run()
def register_peer(self, peer: BasePeer) -> None:
# when a new peer is added to the pool, add it to the idle peer lists
super().register_peer(peer)
peer = cast(ETHPeer, peer)
self._body_peers.put_nowait(peer)
self._receipt_peers.put_nowait(peer)
async def _launch_prerequisite_tasks(self) -> None:
"""
Watch for new headers to be added to the queue, and add the prerequisite
tasks as they become available.
"""
async for headers in self.wait_iter(
self._header_syncer.new_sync_headers()):
try:
# We might end up with duplicates that can be safely ignored.
# Likely scenario: switched which peer downloads headers, and the new peer isn't
# aware of some of the in-progress headers
self._block_persist_tracker.register_tasks(
headers, ignore_duplicates=True)
except MissingDependency:
# The parent of this header is not registered as a dependency yet.
# Some reasons this might happen, in rough descending order of likelihood:
# - a normal fork: the canonical head isn't the parent of the first header synced
# - a bug: the DB has inconsistent state, say saved headers but not block bodies
# - a bug: headers were queued out of order in new_sync_headers
# If the parent header doesn't exist yet, this is a legit bug instead of a fork,
# let the HeaderNotFound exception bubble up
parent_header = await self.wait(
self.db.coro_get_block_header_by_hash(
headers[0].parent_hash))
# This appears to be a fork, since the parent header is persisted,
self.logger.info(
"Fork found while starting fast sync. Canonical head was %s, but the next "
"header %s, has parent %s. Importing fork in case it's the longest chain.",
await self.db.coro_get_canonical_head(),
headers[0],
parent_header,
)
# Set first header's parent as finished
self._block_persist_tracker.set_finished_dependency(
parent_header)
# Re-register the header tasks, which will now succeed
self._block_persist_tracker.register_tasks(headers)
# Sometimes duplicates are added to the queue, when switching from one sync to another.
# We can simply ignore them.
new_body_tasks = tuple(h for h in headers
if h not in self._block_body_tasks)
new_receipt_tasks = tuple(h for h in headers
if h not in self._receipt_tasks)
# if any one of the output queues gets full, hang until there is room
await self.wait(
asyncio.gather(
self._block_body_tasks.add(new_body_tasks),
self._receipt_tasks.add(new_receipt_tasks),
))
async def _display_stats(self) -> None:
while self.is_operational:
await self.sleep(5)
self.logger.debug(
"(in progress, queued, max size) of bodies, receipts: %r",
[(q.num_in_progress(), len(q), q._maxsize) for q in (
self._block_body_tasks,
self._receipt_tasks,
)],
)
stats = self.tracker.report()
utcnow = int(datetime.datetime.utcnow().timestamp())
head_age = utcnow - stats.latest_head.timestamp
self.logger.info(
("blks=%-4d "
"txs=%-5d "
"bps=%-3d "
"tps=%-4d "
"elapsed=%0.1f "
"head=#%d %s "
"age=%s"),
stats.num_blocks,
stats.num_transactions,
stats.blocks_per_second,
stats.transactions_per_second,
stats.elapsed,
stats.latest_head.block_number,
humanize_hash(stats.latest_head.hash),
humanize_elapsed(head_age),
)
async def _persist_ready_blocks(self) -> None:
"""
Persist blocks as soon as all their prerequisites are done: body and receipt downloads.
Persisting must happen in order, so that the block's parent has already been persisted.
Also, determine if fast sync with this peer should end, having reached (or surpassed)
its target hash. If so, shut down this service.
"""
while self.is_operational:
# jhis tracker waits for all prerequisites to be complete, and returns headers in
# order, so that each header's parent is already persisted.
completed_headers = await self.wait(
self._block_persist_tracker.ready_tasks())
await self.wait(self._persist_blocks(completed_headers))
target_hash = self._header_syncer.get_target_header_hash()
if target_hash in [header.hash for header in completed_headers]:
# exit the service when reaching the target hash
self._mark_complete()
break
def _mark_complete(self) -> None:
self.is_complete = True
self.cancel_nowait()
async def _persist_blocks(self, headers: Tuple[BlockHeader, ...]) -> None:
"""
Persist blocks for the given headers, directly to the database
:param headers: headers for which block bodies and receipts have been downloaded
"""
for header in headers:
vm_class = self.chain.get_vm_class(header)
block_class = vm_class.get_block_class()
if _is_body_empty(header):
transactions: List[BaseTransaction] = []
uncles: List[BlockHeader] = []
else:
body = self._pending_bodies.pop(header)
uncles = body.uncles
# transaction data was already persisted in _block_body_bundle_processing, but
# we need to include the transactions for them to be added to the hash->txn lookup
tx_class = block_class.get_transaction_class()
transactions = [
tx_class.from_base_transaction(tx)
for tx in body.transactions
]
# record progress in the tracker
self.tracker.record_transactions(len(transactions))
block = block_class(header, transactions, uncles)
await self.wait(self.db.coro_persist_block(block))
self.tracker.set_latest_head(header)
async def _assign_receipt_download_to_peers(self) -> None:
"""
Loop indefinitely, assigning idle peers to download receipts needed for syncing.
"""
while self.is_operational:
# from all the peers that are not currently downloading receipts, get the fastest
peer = await self.wait(self._receipt_peers.get_fastest())
# get headers for receipts that we need to download, preferring lowest block number
batch_id, headers = await self.wait(
self._receipt_tasks.get(MAX_RECEIPTS_FETCH))
# schedule the receipt download and move on
peer.run_task(
self._run_receipt_download_batch(peer, batch_id, headers))
def _mark_body_download_complete(
self, batch_id: int, completed_headers: Tuple[BlockHeader,
...]) -> None:
super()._mark_body_download_complete(batch_id, completed_headers)
self._block_persist_tracker.finish_prereq(
BlockPersistPrereqs.StoreBlockBodies,
completed_headers,
)
async def _run_receipt_download_batch(
self, peer: ETHPeer, batch_id: int, headers: Tuple[BlockHeader,
...]) -> None:
"""
Given a single batch retrieved from self._receipt_tasks, get as many of the receipt bundles
as possible, and mark them as complete.
"""
# If there is an exception during _process_receipts, prepare to mark the task as finished
# with no headers collected:
completed_headers: Tuple[BlockHeader, ...] = tuple()
try:
completed_headers = await peer.wait(
self._process_receipts(peer, headers))
self._block_persist_tracker.finish_prereq(
BlockPersistPrereqs.StoreReceipts,
completed_headers,
)
except BaseP2PError as exc:
self.logger.info(
"Unexpected p2p perror while downloading receipt from peer: %s",
exc)
self.logger.debug(
"Problem downloading receipt from peer, dropping...",
exc_info=True)
else:
# peer completed successfully, so have it get back in line for processing
if len(completed_headers) > 0:
# peer completed successfully, so have it get back in line for processing
self._receipt_peers.put_nowait(peer)
else:
# peer returned no results, wait a while before trying again
delay = EMPTY_PEER_RESPONSE_PENALTY
self.logger.debug(
"Pausing %s for %.1fs, for sending 0 receipts", peer,
delay)
self.call_later(delay, self._receipt_peers.put_nowait, peer)
finally:
self._receipt_tasks.complete(batch_id, completed_headers)
async def _block_body_bundle_processing(
self, bundles: Tuple[BlockBodyBundle, ...]) -> None:
"""
Fast sync writes all the block body bundle data directly to the database,
in order to make it... fast.
"""
for (_, (_, trie_data_dict), _) in bundles:
await self.wait(self.db.coro_persist_trie_data_dict(trie_data_dict)
)
async def _process_receipts(
self, peer: ETHPeer,
all_headers: Tuple[BlockHeader, ...]) -> Tuple[BlockHeader, ...]:
"""
Downloads and persists the receipts for the given set of block headers.
Some receipts may be trivial, having a blank root hash, and will not be requested.
:param peer: to issue the receipt request to
:param all_headers: attempt to get receipts for as many of these headers as possible
:return: the headers for receipts that were successfully downloaded (or were trivial)
"""
# Post-Byzantium blocks may have identical receipt roots (e.g. when they have the same
# number of transactions and all succeed/failed: ropsten blocks 2503212 and 2503284),
# so we do this to avoid requesting the same receipts multiple times.
# combine headers with the same receipt root, so we can mark them as completed, later
receipt_root_to_headers = groupby(attrgetter('receipt_root'),
all_headers)
# Ignore headers that have an empty receipt root
trivial_headers = tuple(
receipt_root_to_headers.pop(BLANK_ROOT_HASH, tuple()))
# pick one of the headers for each missing receipt root
unique_headers_needed = tuple(
first(headers)
for root, headers in receipt_root_to_headers.items())
if not unique_headers_needed:
return trivial_headers
receipt_bundles = await self._request_receipts(peer,
unique_headers_needed)
if not receipt_bundles:
return trivial_headers
try:
await self._validate_receipts(unique_headers_needed,
receipt_bundles)
except ValidationError as err:
self.logger.info(
"Disconnecting from %s: sent invalid receipt: %s",
peer,
err,
)
await peer.disconnect(DisconnectReason.bad_protocol)
return trivial_headers
# process all of the returned receipts, storing their trie data
# dicts in the database
receipts, trie_roots_and_data_dicts = zip(*receipt_bundles)
receipt_roots, trie_data_dicts = zip(*trie_roots_and_data_dicts)
for trie_data in trie_data_dicts:
await self.wait(self.db.coro_persist_trie_data_dict(trie_data))
# Identify which headers have the receipt roots that are now complete.
completed_header_groups = tuple(
headers for root, headers in receipt_root_to_headers.items()
if root in receipt_roots)
newly_completed_headers = tuple(concat(completed_header_groups))
self.logger.debug(
"Got receipts for %d/%d headers from %s, with %d trivial headers",
len(newly_completed_headers),
len(all_headers) - len(trivial_headers),
peer,
len(trivial_headers),
)
return newly_completed_headers + trivial_headers
async def _validate_receipts(
self, headers: Tuple[BlockHeader, ...],
receipt_bundles: Tuple[ReceiptBundle, ...]) -> None:
header_by_root = {
header.receipt_root: header
for header in headers if not _is_receipts_empty(header)
}
receipts_by_root = {
receipt_root: receipts
for (receipts, (receipt_root, _)) in receipt_bundles
if receipt_root != BLANK_ROOT_HASH
}
for receipt_root, header in header_by_root.items():
if receipt_root not in receipts_by_root:
# this receipt group was not returned by the peer, skip validation
continue
for receipt in receipts_by_root[receipt_root]:
await self.chain.coro_validate_receipt(receipt, header)
async def _request_receipts(
self, peer: ETHPeer,
batch: Tuple[BlockHeader, ...]) -> Tuple[ReceiptBundle, ...]:
"""
Requests the batch of receipts from the given peer, returning the
received receipt data.
"""
self.logger.debug("Requesting receipts for %d headers from %s",
len(batch), peer)
try:
receipt_bundles = await peer.requests.get_receipts(batch)
except TimeoutError as err:
self.logger.debug(
"Timed out requesting receipts for %d headers from %s",
len(batch),
peer,
)
return tuple()
except CancelledError:
self.logger.debug("Pending receipts call to %r future cancelled",
peer)
return tuple()
except OperationCancelled:
self.logger.debug2(
"Pending receipts call to %r operation cancelled", peer)
return tuple()
except PeerConnectionLost:
self.logger.debug(
"Peer went away, cancelling the receipts request and moving on..."
)
return tuple()
except Exception:
self.logger.exception("Unknown error when getting receipts")
raise
if not receipt_bundles:
return tuple()
return receipt_bundles
def test_empty_completion():
ti = OrderedTaskPreparation(TwoPrereqs, identity, lambda x: x - 1)
with pytest.raises(ValidationError):
ti.finish_prereq(TwoPrereqs.Prereq1, tuple())
def test_finish_too_early():
ti = OrderedTaskPreparation(TwoPrereqs, identity, lambda x: x - 1)
ti.set_finished_dependency(3)
with pytest.raises(ValidationError):
ti.finish_prereq(TwoPrereqs.Prereq1, (3, ))
class RegularChainBodySyncer(BaseBodyChainSyncer):
"""
Sync with the Ethereum network by fetching block headers/bodies and importing them.
Here, the run() method will execute the sync loop forever, until our CancelToken is triggered.
"""
def __init__(self,
chain: BaseAsyncChain,
db: BaseAsyncChainDB,
peer_pool: ETHPeerPool,
header_syncer: HeaderSyncerAPI,
token: CancelToken = None) -> None:
super().__init__(chain, db, peer_pool, token)
self._header_syncer = header_syncer
# track when block bodies are downloaded, so that blocks can be imported
self._block_import_tracker = OrderedTaskPreparation(
BlockImportPrereqs,
id_extractor=attrgetter('hash'),
# make sure that a block is not imported until the parent block is imported
dependency_extractor=attrgetter('parent_hash'),
)
async def _run(self) -> None:
head = await self.wait(self.db.coro_get_canonical_head())
self._block_import_tracker.set_finished_dependency(head)
self.run_daemon_task(self._launch_prerequisite_tasks())
self.run_daemon_task(self._assign_body_download_to_peers())
self.run_daemon_task(self._import_ready_blocks())
await super()._run()
def register_peer(self, peer: BasePeer) -> None:
# when a new peer is added to the pool, add it to the idle peer list
super().register_peer(peer)
self._body_peers.put_nowait(cast(ETHPeer, peer))
async def _launch_prerequisite_tasks(self) -> None:
"""
Watch for new headers to be added to the queue, and add the prerequisite
tasks (downloading block bodies) as they become available.
"""
async for headers in self.wait_iter(
self._header_syncer.new_sync_headers()):
self._block_import_tracker.register_tasks(headers)
new_headers = tuple(h for h in headers
if h not in self._block_body_tasks)
# if the output queue gets full, hang until there is room
await self.wait(self._block_body_tasks.add(new_headers))
def _mark_body_download_complete(
self, batch_id: int, completed_headers: Tuple[BlockHeader,
...]) -> None:
super()._mark_body_download_complete(batch_id, completed_headers)
self._block_import_tracker.finish_prereq(
BlockImportPrereqs.StoreBlockBodies,
completed_headers,
)
async def _import_ready_blocks(self) -> None:
"""
Wait for block bodies to be downloaded, then import the blocks.
"""
while self.is_operational:
timer = Timer()
# wait for block bodies to become ready for execution
completed_headers = await self.wait(
self._block_import_tracker.ready_tasks())
await self._import_blocks(completed_headers)
head = await self.wait(self.db.coro_get_canonical_head())
self.logger.info(
"Synced chain segment with %d blocks in %.2f seconds, new head: %s",
len(completed_headers),
timer.elapsed,
head,
)
async def _import_blocks(self, headers: Tuple[BlockHeader, ...]) -> None:
"""
Import the blocks for the corresponding headers
:param headers: headers that have the block bodies downloaded
"""
for header in headers:
vm_class = self.chain.get_vm_class(header)
block_class = vm_class.get_block_class()
if _is_body_empty(header):
transactions: List[BaseTransaction] = []
uncles: List[BlockHeader] = []
else:
body = self._pending_bodies.pop(header)
tx_class = block_class.get_transaction_class()
transactions = [
tx_class.from_base_transaction(tx)
for tx in body.transactions
]
uncles = body.uncles
block = block_class(header, transactions, uncles)
timer = Timer()
_, new_canonical_blocks, old_canonical_blocks = await self.wait(
self.chain.coro_import_block(block, perform_validation=True))
if new_canonical_blocks == (block, ):
# simple import of a single new block.
self.logger.info("Imported block %d (%d txs) in %.2f seconds",
block.number, len(transactions),
timer.elapsed)
elif not new_canonical_blocks:
# imported block from a fork.
self.logger.info(
"Imported non-canonical block %d (%d txs) in %.2f seconds",
block.number, len(transactions), timer.elapsed)
elif old_canonical_blocks:
self.logger.info(
"Chain Reorganization: Imported block %d (%d txs) in %.2f "
"seconds, %d blocks discarded and %d new canonical blocks added",
block.number,
len(transactions),
timer.elapsed,
len(old_canonical_blocks),
len(new_canonical_blocks),
)
else:
raise Exception("Invariant: unreachable code path")
def test_finish_before_setting_start_val():
ti = OrderedTaskPreparation(TwoPrereqs, identity, lambda x: x - 1)
with pytest.raises(ValidationError):
ti.finish_prereq(TwoPrereqs.Prereq1, (2, ))
