def _update_batch(cls, tuples, trx=True):
steemd = SteemClient.instance()
timer = Timer(total=len(tuples), entity='post', laps=['rps', 'wps'])
tuples = sorted(tuples, key=lambda x: x[1]) # enforce ASC id's
for tups in partition_all(1000, tuples):
timer.batch_start()
buffer = []
post_args = [tup[0].split('/') for tup in tups]
posts = steemd.get_content_batch(post_args)
post_ids = [tup[1] for tup in tups]
post_levels = [tup[2] for tup in tups]
for pid, post, level in zip(post_ids, posts, post_levels):
if post['author']:
buffer.append(cls._sql(pid, post, level=level))
else:
print("WARNING: ignoring deleted post {}".format(pid))
cls._bump_last_id(pid)
timer.batch_lap()
cls._batch_queries(buffer, trx)
timer.batch_finish(len(posts))
if len(tuples) >= 1000:
print(timer.batch_status())
def _cache_accounts(cls, accounts, trx=True):
timer = Timer(len(accounts), 'account', ['rps', 'wps'])
for batch in partition_all(1000, accounts):
timer.batch_start()
sqls = cls._generate_cache_sqls(batch)
timer.batch_lap()
cls._batch_update(sqls, trx)
timer.batch_finish(len(batch))
if trx or len(accounts) > 1000:
print(timer.batch_status())
def _cache_accounts(cls, accounts, trx=True):
"""Fetch all `accounts` and write to db."""
timer = Timer(len(accounts), 'account', ['rps', 'wps'])
for batch in partition_all(1000, accounts):
timer.batch_start()
sqls = cls._generate_cache_sqls(batch)
timer.batch_lap()
DB.batch_queries(sqls, trx)
timer.batch_finish(len(batch))
if trx or len(accounts) > 1000:
log.info(timer.batch_status())
def sync_from_steemd():
is_initial_sync = DbState.is_initial_sync()
steemd = get_adapter()
lbound = Blocks.head_num() + 1
ubound = steemd.last_irreversible()
if ubound <= lbound:
return
_abort = False
try:
print("[SYNC] start block %d, +%d to sync" % (lbound, ubound-lbound+1))
timer = Timer(ubound - lbound, entity='block', laps=['rps', 'wps'])
while lbound < ubound:
to = min(lbound + 1000, ubound)
timer.batch_start()
blocks = steemd.get_blocks_range(lbound, to)
timer.batch_lap()
Blocks.process_multi(blocks, is_initial_sync)
timer.batch_finish(len(blocks))
date = blocks[-1]['timestamp']
print(timer.batch_status("[SYNC] Got block %d @ %s" % (to-1, date)))
lbound = to
except KeyboardInterrupt:
traceback.print_exc()
print("\n\n[SYNC] Aborted.. cleaning up..")
_abort = True
if not is_initial_sync:
# Follows flushing may need to be moved closer to core (i.e. moved
# into main block transactions). Important to keep in sync since
# we need to prevent expensive recounts. This will fail if we aborted
# in the middle of a transaction, meaning data loss. Better than
# forcing it, however, since in-memory cache will be out of sync
# with db state.
Follow.flush(trx=True)
# This flush is low importance; accounts are swept regularly.
if not _abort:
Accounts.flush(trx=True)
# If this flush fails, all that could potentially be lost here is
# edits and pre-payout votes. If the post has not been paid out yet,
# then the worst case is it will be synced upon payout. If the post
# is already paid out, worst case is to lose an edit.
CachedPost.flush(trx=True)
if _abort:
print("[SYNC] Aborted")
exit()
def from_steemd(cls, is_initial_sync=False, chunk_size=1000):
"""Fast sync strategy: read/process blocks in batches."""
steemd = SteemClient.instance()
lbound = Blocks.head_num() + 1
ubound = steemd.last_irreversible()
count = ubound - lbound
if count < 1:
return
_abort = False
try:
print("[SYNC] start block %d, +%d to sync" % (lbound, count))
timer = Timer(count, entity='block', laps=['rps', 'wps'])
while lbound < ubound:
timer.batch_start()
# fetch blocks
to = min(lbound + chunk_size, ubound)
blocks = steemd.get_blocks_range(lbound, to)
lbound = to
timer.batch_lap()
# process blocks
Blocks.process_multi(blocks, is_initial_sync)
timer.batch_finish(len(blocks))
date = blocks[-1]['timestamp']
print(
timer.batch_status("[SYNC] Got block %d @ %s" %
(to - 1, date)))
except KeyboardInterrupt:
traceback.print_exc()
print("\n\n[SYNC] Aborted.. cleaning up..")
_abort = True
if not is_initial_sync:
# This flush is low importance; accounts are swept regularly.
if not _abort:
Accounts.flush(trx=True)
# If this flush fails, all that could potentially be lost here is
# edits and pre-payout votes. If the post has not been paid out yet,
# then the worst case is it will be synced upon payout. If the post
# is already paid out, worst case is to lose an edit.
CachedPost.flush(trx=True)
if _abort:
print("[SYNC] Aborted")
exit()
def _update_batch(cls, tuples, trx=True, full_total=None):
"""Fetch, process, and write a batch of posts.
Given a set of posts, fetch from steemd and write them to the
db. The `tuples` arg is the form of `[(url, id, level)*]`
representing posts which are to be fetched from steemd and
updated in cache.
Regarding _bump_last_id: there's a rare edge case when the last
hive_post entry has been deleted "in the future" (ie, we haven't
seen the delete op yet). So even when the post is not found
(i.e. `not post['author']`), it's important to advance _last_id,
because this cursor is used to deduce any missing cache entries.
"""
steemd = SteemClient.instance()
timer = Timer(total=len(tuples),
entity='post',
laps=['rps', 'wps'],
full_total=full_total)
tuples = sorted(tuples, key=lambda x: x[1]) # enforce ASC id's
for tups in partition_all(1000, tuples):
timer.batch_start()
buffer = []
post_args = [tup[0].split('/') for tup in tups]
posts = steemd.get_content_batch(post_args)
post_ids = [tup[1] for tup in tups]
post_levels = [tup[2] for tup in tups]
for pid, post, level in zip(post_ids, posts, post_levels):
if post['author']:
buffer.extend(cls._sql(pid, post, level=level))
else:
# When a post has been deleted (or otherwise DNE),
# steemd simply returns a blank post object w/ all
# fields blank. While it's best to not try to cache
# already-deleted posts, it can happen during missed
# post sweep and while using `trail_blocks` > 0.
pass
cls._bump_last_id(pid)
timer.batch_lap()
DB.batch_queries(buffer, trx)
timer.batch_finish(len(posts))
if len(tuples) >= 1000:
log.info(timer.batch_status())
def _cache_accounts(cls, accounts, steem, trx=True):
"""Fetch all `accounts` and write to db."""
timer = Timer(len(accounts), 'account', ['rps', 'wps'])
for name_batch in partition_all(1000, accounts):
cached_at = datetime.now().strftime('%Y-%m-%dT%H:%M:%S')
timer.batch_start()
batch = steem.get_accounts(name_batch)
timer.batch_lap()
sqls = [cls._sql(acct, cached_at) for acct in batch]
DB.batch_queries(sqls, trx)
timer.batch_finish(len(batch))
if trx or len(accounts) > 1000:
log.info(timer.batch_status())
def _update_batch(cls, tuples, trx=True, full_total=None):
"""Fetch, process, and write a batch of posts.
Given a set of posts, fetch from steemd and write them to the
db. The `tuples` arg is the form of `[(url, id, level)*]`
representing posts which are to be fetched from steemd and
updated in cache.
Regarding _bump_last_id: there's a rare edge case when the last
hive_post entry has been deleted "in the future" (ie, we haven't
seen the delete op yet). So even when the post is not found
(i.e. `not post['author']`), it's important to advance _last_id,
because this cursor is used to deduce any missing cache entries.
"""
steemd = SteemClient.instance()
timer = Timer(total=len(tuples), entity='post',
laps=['rps', 'wps'], full_total=full_total)
tuples = sorted(tuples, key=lambda x: x[1]) # enforce ASC id's
for tups in partition_all(1000, tuples):
timer.batch_start()
buffer = []
post_args = [tup[0].split('/') for tup in tups]
posts = steemd.get_content_batch(post_args)
post_ids = [tup[1] for tup in tups]
post_levels = [tup[2] for tup in tups]
for pid, post, level in zip(post_ids, posts, post_levels):
if post['author']:
buffer.append(cls._sql(pid, post, level=level))
else:
# expected to happen when sweeping missed posts as
# part of initial sync or crash recovery routine,
# otherwise indicates potential bug. TODO: assert?
if not cls._sweeping_missed:
print("WARNING: missing/deleted post %d" % pid)
cls._bump_last_id(pid)
timer.batch_lap()
cls._batch_queries(buffer, trx)
timer.batch_finish(len(posts))
if len(tuples) >= 1000:
print(timer.batch_status())
def from_dpayd(self, is_initial_sync=False, chunk_size=1000):
"""Fast sync strategy: read/process blocks in batches."""
# pylint: disable=no-self-use
dpayd = self._dpay
lbound = Blocks.head_num() + 1
ubound = self._conf.get('test_max_block') or dpayd.last_irreversible()
count = ubound - lbound
if count < 1:
return
log.info("[SYNC] start block %d, +%d to sync", lbound, count)
timer = Timer(count, entity='block', laps=['rps', 'wps'])
while lbound < ubound:
timer.batch_start()
# fetch blocks
to = min(lbound + chunk_size, ubound)
blocks = dpayd.get_blocks_range(lbound, to)
lbound = to
timer.batch_lap()
# process blocks
Blocks.process_multi(blocks, is_initial_sync)
timer.batch_finish(len(blocks))
_prefix = ("[SYNC] Got block %d @ %s" %
(to - 1, blocks[-1]['timestamp']))
log.info(timer.batch_status(_prefix))
if not is_initial_sync:
# This flush is low importance; accounts are swept regularly.
Accounts.flush(dpayd, trx=True)
# If this flush fails, all that could potentially be lost here is
# edits and pre-payout votes. If the post has not been paid out yet,
# then the worst case is it will be synced upon payout. If the post
# is already paid out, worst case is to lose an edit.
CachedPost.flush(dpayd, trx=True)
def _block_consumer(blocks_data_provider, is_initial_sync, lbound, ubound):
from hive.utils.stats import minmax
is_debug = log.isEnabledFor(10)
num = 0
time_start = OPSM.start()
rate = {}
LIMIT_FOR_PROCESSED_BLOCKS = 1000
rate = minmax(rate, 0, 1.0, 0)
sync_type_prefix = "[INITIAL SYNC]" if is_initial_sync else "[FAST SYNC]"
def print_summary():
stop = OPSM.stop(time_start)
log.info("=== TOTAL STATS ===")
wtm = WSM.log_global("Total waiting times")
ftm = FSM.log_global("Total flush times")
otm = OPSM.log_global("All operations present in the processed blocks")
ttm = ftm + otm + wtm
log.info(
f"Elapsed time: {stop :.4f}s. Calculated elapsed time: {ttm :.4f}s. Difference: {stop - ttm :.4f}s"
)
if rate:
log.info(
f"Highest block processing rate: {rate['max'] :.4f} bps. From: {rate['max_from']} To: {rate['max_to']}"
)
log.info(
f"Lowest block processing rate: {rate['min'] :.4f} bps. From: {rate['min_from']} To: {rate['min_to']}"
)
log.info("=== TOTAL STATS ===")
try:
Blocks.set_end_of_sync_lib(ubound)
count = ubound - lbound
timer = Timer(count, entity='block', laps=['rps', 'wps'])
while lbound < ubound:
number_of_blocks_to_proceed = min(
[LIMIT_FOR_PROCESSED_BLOCKS, ubound - lbound])
time_before_waiting_for_data = perf()
blocks = blocks_data_provider.get(number_of_blocks_to_proceed)
if not can_continue_thread():
break
assert len(blocks) == number_of_blocks_to_proceed
to = min(lbound + number_of_blocks_to_proceed, ubound)
timer.batch_start()
block_start = perf()
Blocks.process_multi(blocks, is_initial_sync)
block_end = perf()
timer.batch_lap()
timer.batch_finish(len(blocks))
time_current = perf()
prefix = ("%s Got block %d @ %s" %
(sync_type_prefix, to - 1, blocks[-1].get_date()))
log.info(timer.batch_status(prefix))
log.info("%s Time elapsed: %fs", sync_type_prefix,
time_current - time_start)
log.info("%s Current system time: %s", sync_type_prefix,
datetime.now().strftime("%H:%M:%S"))
log.info(log_memory_usage())
rate = minmax(rate, len(blocks),
time_current - time_before_waiting_for_data, lbound)
if block_end - block_start > 1.0 or is_debug:
otm = OPSM.log_current(
"Operations present in the processed blocks")
ftm = FSM.log_current("Flushing times")
wtm = WSM.log_current("Waiting times")
log.info(f"Calculated time: {otm+ftm+wtm :.4f} s.")
OPSM.next_blocks()
FSM.next_blocks()
WSM.next_blocks()
lbound = to
PC.broadcast(
BroadcastObject('sync_current_block', lbound, 'blocks'))
num = num + 1
if not can_continue_thread():
break
except Exception:
log.exception("Exception caught during processing blocks...")
set_exception_thrown()
print_summary()
raise
print_summary()
return num
def _update_batch(cls, steem, tuples, trx=True, full_total=None):
"""Fetch, process, and write a batch of posts.
Given a set of posts, fetch from steemd and write them to the
db. The `tuples` arg is the form of `[(url, id, level)*]`
representing posts which are to be fetched from steemd and
updated in cache.
Regarding _bump_last_id: there's a rare edge case when the last
hive_post entry has been deleted "in the future" (ie, we haven't
seen the delete op yet). So even when the post is not found
(i.e. `not post['author']`), it's important to advance _last_id,
because this cursor is used to deduce any missing cache entries.
"""
# pylint: disable=too-many-locals
timer = Timer(total=len(tuples), entity='post',
laps=['rps', 'wps'], full_total=full_total)
tuples = sorted(tuples, key=lambda x: x[1]) # enforce ASC id's
for tups in partition_all(1000, tuples):
timer.batch_start()
buffer = []
post_args = [tup[0].split('/') for tup in tups]
posts = steem.get_content_batch(post_args)
post_ids = [tup[1] for tup in tups]
post_levels = [tup[2] for tup in tups]
coremap = cls._get_core_fields(tups)
for pid, post, level in zip(post_ids, posts, post_levels):
if post['author']:
assert pid in coremap, 'pid not in coremap'
if pid in coremap:
core = coremap[pid]
post['category'] = core['category']
post['community_id'] = core['community_id']
post['gray'] = core['is_muted']
post['hide'] = not core['is_valid']
buffer.extend(cls._sql(pid, post, level=level))
else:
# When a post has been deleted (or otherwise DNE),
# steemd simply returns a blank post object w/ all
# fields blank. While it's best to not try to cache
# already-deleted posts, it can happen during missed
# post sweep and while using `trail_blocks` > 0.
# monitor: post not found which should def. exist; see #173
sql = """SELECT id, author, permlink, is_deleted
FROM hive_posts WHERE id = :id"""
row = DB.query_row(sql, id=pid)
if row['is_deleted']:
# rare or impossible -- report if detected
log.error("found deleted post for %s: %s", level, row)
else:
# most likely cause of this condition is that the post
# has been deleted (e.g. sync trails by 2 blocks, post
# was inserted at head-2, deleted at head). another
# possible cause is that a node behind a load balancer
# is behind; we detected a new post but querying a node
# that hasn't seen it yet.
log.warning("post not found -- DEFER %s %s", level, row)
cls._dirty(level, row['author'], row['permlink'], pid)
cls._bump_last_id(pid)
timer.batch_lap()
DB.batch_queries(buffer, trx)
timer.batch_finish(len(posts))
if len(tuples) >= 1000:
log.info(timer.batch_status())
