Merge pull request #3342 from lbryio/bug_flush_counter

[resync required] Avoid flush counter overflows on long running hubs by increasing it to 32 bits
This commit is contained in:
Jack Robison 2021-07-07 23:45:47 -04:00 committed by GitHub
commit f05dc46432
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 101 additions and 8 deletions

View file

@ -436,7 +436,8 @@ class RPCSession(SessionBase):
except CancelledError:
raise
except Exception:
self.logger.exception(f'exception handling {request}')
reqstr = str(request)
self.logger.exception(f'exception handling {reqstr[:16_000]}')
result = RPCError(JSONRPC.INTERNAL_ERROR,
'internal server error')
if isinstance(request, Request):

View file

@ -16,13 +16,13 @@ from collections import defaultdict
from functools import partial
from lbry.wallet.server import util
from lbry.wallet.server.util import pack_be_uint16, unpack_be_uint16_from
from lbry.wallet.server.util import pack_be_uint32, unpack_be_uint32_from, unpack_be_uint16_from
from lbry.wallet.server.hash import hash_to_hex_str, HASHX_LEN
class History:
DB_VERSIONS = [0]
DB_VERSIONS = [0, 1]
def __init__(self):
self.logger = util.class_logger(__name__, self.__class__.__name__)
@ -32,9 +32,34 @@ class History:
self.unflushed_count = 0
self.db = None
@property
def needs_migration(self):
return self.db_version != max(self.DB_VERSIONS)
def migrate(self):
# 0 -> 1: flush_count from 16 to 32 bits
self.logger.warning("HISTORY MIGRATION IN PROGRESS. Please avoid shutting down before it finishes.")
with self.db.write_batch() as batch:
for key, value in self.db.iterator(prefix=b''):
if len(key) != 13:
continue
flush_id, = unpack_be_uint16_from(key[-2:])
new_key = key[:-2] + pack_be_uint32(flush_id)
batch.put(new_key, value)
self.logger.warning("history migration: new keys added, removing old ones.")
for key, value in self.db.iterator(prefix=b''):
if len(key) == 13:
batch.delete(key)
self.logger.warning("history migration: writing new state.")
self.db_version = 1
self.write_state(batch)
self.logger.warning("history migration: done.")
def open_db(self, db_class, for_sync, utxo_flush_count, compacting):
self.db = db_class('hist', for_sync)
self.read_state()
if self.needs_migration:
self.migrate()
self.clear_excess(utxo_flush_count)
# An incomplete compaction needs to be cancelled otherwise
# restarting it will corrupt the history
@ -81,7 +106,7 @@ class History:
keys = []
for key, hist in self.db.iterator(prefix=b''):
flush_id, = unpack_be_uint16_from(key[-2:])
flush_id, = unpack_be_uint32_from(key[-4:])
if flush_id > utxo_flush_count:
keys.append(key)
@ -126,7 +151,7 @@ class History:
def flush(self):
start_time = time.time()
self.flush_count += 1
flush_id = pack_be_uint16(self.flush_count)
flush_id = pack_be_uint32(self.flush_count)
unflushed = self.unflushed
with self.db.write_batch() as batch:
@ -250,7 +275,7 @@ class History:
write_size = 0
keys_to_delete.update(hist_map)
for n, chunk in enumerate(util.chunks(full_hist, max_row_size)):
key = hashX + pack_be_uint16(n)
key = hashX + pack_be_uint32(n)
if hist_map.get(key) == chunk:
keys_to_delete.remove(key)
else:
@ -301,8 +326,8 @@ class History:
# Loop over 2-byte prefixes
cursor = self.comp_cursor
while write_size < limit and cursor < 65536:
prefix = pack_be_uint16(cursor)
while write_size < limit and cursor < (1 << 32):
prefix = pack_be_uint32(cursor)
write_size += self._compact_prefix(prefix, write_items,
keys_to_delete)
cursor += 1

View file

@ -189,3 +189,13 @@ class TestHubDiscovery(CommandTestCase):
('127.0.0.1', 9988): {}
}
)
class TestStressFlush(CommandTestCase):
async def test_flush_over_66_thousand(self):
history = self.conductor.spv_node.server.db.history
history.flush_count = 66_000
history.flush()
self.assertEqual(history.flush_count, 66_001)
await self.generate(1)
self.assertEqual(history.flush_count, 66_002)

View file

@ -0,0 +1,57 @@
import unittest
from shutil import rmtree
from tempfile import mkdtemp
from lbry.wallet.server.history import History
from lbry.wallet.server.storage import LevelDB
# dumped from a real history database. Aside from the state, all records are <hashX><flush_count>: <value>
STATE_RECORD = (b'state\x00\x00', b"{'flush_count': 21497, 'comp_flush_count': -1, 'comp_cursor': -1, 'db_version': 0}")
UNMIGRATED_RECORDS = {
'00538b2cbe4a5f1be2dc320241': 'f5ed500142ee5001',
'00538b48def1904014880501f2': 'b9a52a01baa52a01',
'00538cdcf989b74de32c5100ca': 'c973870078748700',
'00538d42d5df44603474284ae1': 'f5d9d802',
'00538d42d5df44603474284ae2': '75dad802',
'00538ebc879dac6ddbee9e0029': '3ca42f0042a42f00',
'00538ed1d391327208748200bc': '804e7d00af4e7d00',
'00538f3de41d9e33affa0300c2': '7de8810086e88100',
'00539007f87792d98422c505a5': '8c5a7202445b7202',
'0053902cf52ee9682d633b0575': 'eb0f64026c106402',
'005390e05674571551632205a2': 'a13d7102e13d7102',
'0053914ef25a9ceed927330584': '78096902960b6902',
'005391768113f69548f37a01b1': 'a5b90b0114ba0b01',
'005391a289812669e5b44c02c2': '33da8a016cdc8a01',
}
class TestHistoryDBMigration(unittest.TestCase):
def test_migrate_flush_count_from_16_to_32_bits(self):
self.history = History()
tmpdir = mkdtemp()
self.addCleanup(lambda: rmtree(tmpdir))
LevelDB.import_module()
db = LevelDB(tmpdir, 'hist', True)
with db.write_batch() as batch:
for key, value in UNMIGRATED_RECORDS.items():
batch.put(bytes.fromhex(key), bytes.fromhex(value))
batch.put(*STATE_RECORD)
self.history.db = db
self.history.read_state()
self.assertEqual(21497, self.history.flush_count)
self.assertEqual(0, self.history.db_version)
self.assertTrue(self.history.needs_migration)
self.history.migrate()
self.assertFalse(self.history.needs_migration)
self.assertEqual(1, self.history.db_version)
for idx, (key, value) in enumerate(sorted(db.iterator())):
if key == b'state\x00\x00':
continue
key, counter = key[:-4], key[-4:]
expected_value = UNMIGRATED_RECORDS[key.hex() + counter.hex()[-4:]]
self.assertEqual(value.hex(), expected_value)
if __name__ == '__main__':
unittest.main()