perf: 优化下 `orgid_nodekey_assetsid_mapping`

pull/5717/head
xinwen 4 years ago committed by Jiangjie.Bai
parent 615bcadf62
commit 184e8b31e6

@ -40,7 +40,7 @@ def compute_parent_key(key):
class NodeQuerySet(models.QuerySet):
def delete(self):
raise NotImplementedError
#
class FamilyMixin:
__parents = None
@ -261,6 +261,12 @@ class NodeAllAssetsMappingMixin:
# { org_id: { node_key: [ asset1_id, asset2_id ] } }
orgid_nodekey_assetsid_mapping = defaultdict(dict)
locks_for_get_mapping_from_cache = defaultdict(threading.Lock)
@classmethod
def get_lock(cls, org_id):
lock = cls.locks_for_get_mapping_from_cache[str(org_id)]
return lock
@classmethod
def get_node_all_asset_ids_mapping(cls, org_id):
@ -268,8 +274,22 @@ class NodeAllAssetsMappingMixin:
if _mapping:
return _mapping
_mapping = cls.get_node_all_asset_ids_mapping_from_cache_or_generate_to_cache(org_id)
cls.set_node_all_asset_ids_mapping_to_memory(org_id, mapping=_mapping)
logger.debug(f'Get node asset mapping from memory failed, acquire thread lock: '
f'thread={threading.get_ident()} '
f'org_id={org_id}')
with cls.get_lock(org_id):
logger.debug(f'Acquired thread lock ok. check if mapping is in memory now: '
f'thread={threading.get_ident()} '
f'org_id={org_id}')
_mapping = cls.get_node_all_asset_ids_mapping_from_memory(org_id)
if _mapping:
logger.debug(f'Mapping is already in memory now: '
f'thread={threading.get_ident()} '
f'org_id={org_id}')
return _mapping
_mapping = cls.get_node_all_asset_ids_mapping_from_cache_or_generate_to_cache(org_id)
cls.set_node_all_asset_ids_mapping_to_memory(org_id, mapping=_mapping)
return _mapping
# from memory
@ -295,14 +315,10 @@ class NodeAllAssetsMappingMixin:
return mapping
lock_key = f'KEY_LOCK_GENERATE_ORG_{org_id}_NODE_ALL_ASSET_ids_MAPPING'
logger.info(f'Thread[{threading.get_ident()}] acquiring lock[{lock_key}] ...')
with DistributedLock(lock_key):
logger.info(f'Thread[{threading.get_ident()}] acquire lock[{lock_key}] ok')
# 这里使用无限期锁,原因是如果这里卡住了,就卡在数据库了,说明
# 数据库繁忙,所以不应该再有线程执行这个操作,使数据库忙上加忙
# 这里最好先判断内存中有没有,防止同一进程的多个线程重复从 cache 中获取数据,
# 但逻辑过于繁琐,直接判断 cache 吧
_mapping = cls.get_node_all_asset_ids_mapping_from_cache(org_id)
if _mapping:
return _mapping
@ -315,6 +331,9 @@ class NodeAllAssetsMappingMixin:
def get_node_all_asset_ids_mapping_from_cache(cls, org_id):
cache_key = cls._get_cache_key_for_node_all_asset_ids_mapping(org_id)
mapping = cache.get(cache_key)
logger.info(f'Get node asset mapping from cache {bool(mapping)}: '
f'thread={threading.get_ident()} '
f'org_id={org_id}')
return mapping
@classmethod
@ -335,6 +354,9 @@ class NodeAllAssetsMappingMixin:
def generate_node_all_asset_ids_mapping(cls, org_id):
from .asset import Asset
logger.info(f'Generate node asset mapping: '
f'thread={threading.get_ident()} '
f'org_id={org_id}')
t1 = time.time()
with tmp_to_org(org_id):
node_ids_key = Node.objects.annotate(
@ -366,7 +388,7 @@ class NodeAllAssetsMappingMixin:
mapping[ancestor_key].update(asset_ids)
t3 = time.time()
logger.debug('t1-t2(DB Query): {} s, t3-t2(Generate mapping): {} s'.format(t2-t1, t3-t2))
logger.info('t1-t2(DB Query): {} s, t3-t2(Generate mapping): {} s'.format(t2-t1, t3-t2))
return mapping

@ -198,10 +198,12 @@ logger = get_logger(__name__)
def timeit(func):
def wrapper(*args, **kwargs):
if hasattr(func, '__name__'):
name = func.__name__
else:
name = func
name = func
for attr in ('__qualname__', '__name__'):
if hasattr(func, attr):
name = getattr(func, attr)
break
logger.debug("Start call: {}".format(name))
now = time.time()
result = func(*args, **kwargs)

Loading…
Cancel
Save