def prepare_engine_args(disable_connection_pool=False):
"""Prepare SQLAlchemy engine args"""
default_args = {}
for dialect, default in DEFAULT_ENGINE_ARGS.items():
if SQL_ALCHEMY_CONN.startswith(dialect):
default_args = default.copy()
break
engine_args: dict = conf.getjson('core', 'sql_alchemy_engine_args', fallback=default_args) # type: ignore
if disable_connection_pool or not conf.getboolean('core', 'SQL_ALCHEMY_POOL_ENABLED'):
engine_args['poolclass'] = NullPool
log.debug("settings.prepare_engine_args(): Using NullPool")
elif not SQL_ALCHEMY_CONN.startswith('sqlite'):
# Pool size engine args not supported by sqlite.
# If no config value is defined for the pool size, select a reasonable value.
# 0 means no limit, which could lead to exceeding the Database connection limit.
pool_size = conf.getint('core', 'SQL_ALCHEMY_POOL_SIZE', fallback=5)
# The maximum overflow size of the pool.
# When the number of checked-out connections reaches the size set in pool_size,
# additional connections will be returned up to this limit.
# When those additional connections are returned to the pool, they are disconnected and discarded.
# It follows then that the total number of simultaneous connections
# the pool will allow is pool_size + max_overflow,
# and the total number of “sleeping” connections the pool will allow is pool_size.
# max_overflow can be set to -1 to indicate no overflow limit;
# no limit will be placed on the total number
# of concurrent connections. Defaults to 10.
max_overflow = conf.getint('core', 'SQL_ALCHEMY_MAX_OVERFLOW', fallback=10)
# The DB server already has a value for wait_timeout (number of seconds after
# which an idle sleeping connection should be killed). Since other DBs may
# co-exist on the same server, SQLAlchemy should set its
# pool_recycle to an equal or smaller value.
pool_recycle = conf.getint('core', 'SQL_ALCHEMY_POOL_RECYCLE', fallback=1800)
# Check connection at the start of each connection pool checkout.
# Typically, this is a simple statement like “SELECT 1”, but may also make use
# of some DBAPI-specific method to test the connection for liveness.
# More information here:
# https://docs.sqlalchemy.org/en/13/core/pooling.html#disconnect-handling-pessimistic
pool_pre_ping = conf.getboolean('core', 'SQL_ALCHEMY_POOL_PRE_PING', fallback=True)
log.debug(
"settings.prepare_engine_args(): Using pool settings. pool_size=%d, max_overflow=%d, "
"pool_recycle=%d, pid=%d",
pool_size,
max_overflow,
pool_recycle,
os.getpid(),
)
engine_args['pool_size'] = pool_size
engine_args['pool_recycle'] = pool_recycle
engine_args['pool_pre_ping'] = pool_pre_ping
engine_args['max_overflow'] = max_overflow
# The default isolation level for MySQL (REPEATABLE READ) can introduce inconsistencies when
# running multiple schedulers, as repeated queries on the same session may read from stale snapshots.
# 'READ COMMITTED' is the default value for PostgreSQL.
# More information here:
# https://dev.mysql.com/doc/refman/8.0/en/innodb-transaction-isolation-levels.html"
# Similarly MSSQL default isolation level should be set to READ COMMITTED.
# We also make sure that READ_COMMITTED_SNAPSHOT option is on, in order to avoid deadlocks when
# Select queries are running. This is by default enforced during init/upgrade. More information:
# https://docs.microsoft.com/en-us/sql/t-sql/statements/set-transaction-isolation-level-transact-sql
if SQL_ALCHEMY_CONN.startswith(('mysql', 'mssql')):
engine_args['isolation_level'] = 'READ COMMITTED'
# Allow the user to specify an encoding for their DB otherwise default
# to utf-8 so jobs & users with non-latin1 characters can still use us.
engine_args['encoding'] = conf.get('core', 'SQL_ENGINE_ENCODING', fallback='utf-8')
return engine_args
def __init__(self):
configuration_dict = conf.as_dict(display_sensitive=True)
self.core_configuration = configuration_dict[self.core_section]
self.airflow_home = AIRFLOW_HOME
self.dags_folder = conf.get(self.core_section, 'dags_folder')
self.parallelism = conf.getint(self.core_section, 'parallelism')
self.pod_template_file = conf.get(self.kubernetes_section,
'pod_template_file',
fallback=None)
self.delete_worker_pods = conf.getboolean(self.kubernetes_section,
'delete_worker_pods')
self.delete_worker_pods_on_failure = conf.getboolean(
self.kubernetes_section, 'delete_worker_pods_on_failure')
self.worker_pods_creation_batch_size = conf.getint(
self.kubernetes_section, 'worker_pods_creation_batch_size')
self.worker_container_repository = conf.get(
self.kubernetes_section, 'worker_container_repository')
self.worker_container_tag = conf.get(self.kubernetes_section,
'worker_container_tag')
if self.worker_container_repository and self.worker_container_tag:
self.kube_image = f'{self.worker_container_repository}:{self.worker_container_tag}'
else:
self.kube_image = None
# The Kubernetes Namespace in which the Scheduler and Webserver reside. Note
# that if your
# cluster has RBAC enabled, your scheduler may need service account permissions to
# create, watch, get, and delete pods in this namespace.
self.kube_namespace = conf.get(self.kubernetes_section, 'namespace')
self.multi_namespace_mode = conf.getboolean(self.kubernetes_section,
'multi_namespace_mode')
# The Kubernetes Namespace in which pods will be created by the executor. Note
# that if your
# cluster has RBAC enabled, your workers may need service account permissions to
# interact with cluster components.
self.executor_namespace = conf.get(self.kubernetes_section,
'namespace')
self.worker_pods_pending_timeout = conf.getint(
self.kubernetes_section, 'worker_pods_pending_timeout')
self.worker_pods_pending_timeout_check_interval = conf.getint(
self.kubernetes_section,
'worker_pods_pending_timeout_check_interval')
self.worker_pods_pending_timeout_batch_size = conf.getint(
self.kubernetes_section, 'worker_pods_pending_timeout_batch_size')
self.worker_pods_queued_check_interval = conf.getint(
self.kubernetes_section, 'worker_pods_queued_check_interval')
self.kube_client_request_args = conf.getjson(
self.kubernetes_section, 'kube_client_request_args', fallback={})
if not isinstance(self.kube_client_request_args, dict):
raise AirflowConfigException(
f"[{self.kubernetes_section}] 'kube_client_request_args' expected a JSON dict, got "
+ type(self.kube_client_request_args).__name__)
if self.kube_client_request_args:
if '_request_timeout' in self.kube_client_request_args and isinstance(
self.kube_client_request_args['_request_timeout'], list):
self.kube_client_request_args['_request_timeout'] = tuple(
self.kube_client_request_args['_request_timeout'])
self.delete_option_kwargs = conf.getjson(self.kubernetes_section,
'delete_option_kwargs',
fallback={})
if not isinstance(self.delete_option_kwargs, dict):
raise AirflowConfigException(
f"[{self.kubernetes_section}] 'delete_option_kwargs' expected a JSON dict, got "
+ type(self.delete_option_kwargs).__name__)
