def test_get_statement(mocker, server): mocker.patch.object(Statement, 'from_json') client = LivyClient(server) statement = client.get_statement(MOCK_SESSION_ID, MOCK_STATEMENT_ID) assert statement == Statement.from_json.return_value Statement.from_json.assert_called_once_with(MOCK_SESSION_ID, MOCK_STATEMENT_JSON)
def test_get_statement(requests_mock, mocker): requests_mock.get( f"http://example.com/sessions/{MOCK_SESSION_ID}" + f"/statements/{MOCK_STATEMENT_ID}", json=MOCK_STATEMENT_JSON, ) mocker.patch.object(Statement, "from_json") client = LivyClient("http://example.com") statement = client.get_statement(MOCK_SESSION_ID, MOCK_STATEMENT_ID) assert statement == Statement.from_json.return_value Statement.from_json.assert_called_once_with(MOCK_SESSION_ID, MOCK_STATEMENT_JSON)
class LivySession: """Manages a remote Livy session and high-level interactions with it. :param url: The URL of the Livy server. :param session_id: The ID of the Livy session. :param auth: A requests-compatible auth object to use when making requests. :param verify: Either a boolean, in which case it controls whether we verify the server’s TLS certificate, or a string, in which case it must be a path to a CA bundle to use. Defaults to ``True``. :param requests_session: A specific ``requests.Session`` to use, allowing advanced customisation. The caller is responsible for closing the session. :param kind: The kind of session to create. :param echo: Whether to echo output printed in the remote session. Defaults to ``True``. :param check: Whether to raise an exception when a statement in the remote session fails. Defaults to ``True``. """ def __init__( self, url: str, session_id: int, auth: Auth = None, verify: Verify = True, requests_session: requests.Session = None, kind: SessionKind = SessionKind.PYSPARK, echo: bool = True, check: bool = True, ) -> None: self.client = LivyClient(url, auth, verify, requests_session) self.session_id = session_id self.kind = kind self.echo = echo self.check = check @classmethod def create( cls, url: str, auth: Auth = None, verify: Verify = True, requests_session: requests.Session = None, kind: SessionKind = SessionKind.PYSPARK, proxy_user: str = None, jars: List[str] = None, py_files: List[str] = None, files: List[str] = None, driver_memory: str = None, driver_cores: int = None, executor_memory: str = None, executor_cores: int = None, num_executors: int = None, archives: List[str] = None, queue: str = None, name: str = None, spark_conf: Dict[str, Any] = None, heartbeat_timeout: int = None, echo: bool = True, check: bool = True, ) -> "LivySession": """Create a new Livy session. The py_files, files, jars and archives arguments are lists of URLs, e.g. ["s3://bucket/object", "hdfs://path/to/file", ...] and must be reachable by the Spark driver process. If the provided URL has no scheme, it's considered to be relative to the default file system configured in the Livy server. URLs in the py_files argument are copied to a temporary staging area and inserted into Python's sys.path ahead of the standard library paths. This allows you to import .py, .zip and .egg files in Python. URLs for jars, py_files, files and archives arguments are all copied to the same working directory on the Spark cluster. The driver_memory and executor_memory arguments have the same format as JVM memory strings with a size unit suffix ("k", "m", "g" or "t") (e.g. 512m, 2g). See https://spark.apache.org/docs/latest/configuration.html for more information on Spark configuration properties. :param url: The URL of the Livy server. :param auth: A requests-compatible auth object to use when making requests. :param verify: Either a boolean, in which case it controls whether we verify the server’s TLS certificate, or a string, in which case it must be a path to a CA bundle to use. Defaults to ``True``. :param requests_session: A specific ``requests.Session`` to use, allowing advanced customisation. The caller is responsible for closing the session. :param kind: The kind of session to create. :param proxy_user: User to impersonate when starting the session. :param jars: URLs of jars to be used in this session. :param py_files: URLs of Python files to be used in this session. :param files: URLs of files to be used in this session. :param driver_memory: Amount of memory to use for the driver process (e.g. '512m'). :param driver_cores: Number of cores to use for the driver process. :param executor_memory: Amount of memory to use per executor process (e.g. '512m'). :param executor_cores: Number of cores to use for each executor. :param num_executors: Number of executors to launch for this session. :param archives: URLs of archives to be used in this session. :param queue: The name of the YARN queue to which submitted. :param name: The name of this session. :param spark_conf: Spark configuration properties. :param heartbeat_timeout: Optional Timeout in seconds to which session be automatically orphaned if no heartbeat is received. :param echo: Whether to echo output printed in the remote session. Defaults to ``True``. :param check: Whether to raise an exception when a statement in the remote session fails. Defaults to ``True``. """ client = LivyClient(url, auth, verify, requests_session) session = client.create_session( kind, proxy_user, jars, py_files, files, driver_memory, driver_cores, executor_memory, executor_cores, num_executors, archives, queue, name, spark_conf, heartbeat_timeout, ) client.close() return cls( url, session.session_id, auth, verify, requests_session, kind, echo, check, ) def __enter__(self) -> "LivySession": self.wait() return self def __exit__(self, exc_type, exc_value, traceback) -> None: self.close() def wait(self) -> None: """Wait for the session to be ready.""" intervals = polling_intervals([0.1, 0.2, 0.3, 0.5], 1.0) while self.state in SESSION_STATE_NOT_READY: time.sleep(next(intervals)) @property def state(self) -> SessionState: """The state of the managed Spark session.""" session = self.client.get_session(self.session_id) if session is None: raise ValueError("session not found - it may have been shut down") return session.state def close(self) -> None: """Kill the managed Spark session.""" self.client.delete_session(self.session_id) self.client.close() def run(self, code: str) -> Output: """Run some code in the managed Spark session. :param code: The code to run. """ output = self._execute(code) if self.echo and output.text: print(output.text) if self.check: output.raise_for_status() return output def download(self, dataframe_name: str) -> pandas.DataFrame: """Evaluate and download a Spark dataframe from the managed session. :param dataframe_name: The name of the Spark dataframe to download. """ code = _spark_serialise_dataframe_code(dataframe_name, self.kind) output = self._execute(code) output.raise_for_status() if output.text is None: raise RuntimeError("statement had no text output") return _deserialise_dataframe(output.text) def read(self, dataframe_name: str) -> pandas.DataFrame: """Evaluate and retrieve a Spark dataframe in the managed session. :param dataframe_name: The name of the Spark dataframe to read. .. deprecated:: 0.8.0 Use :meth:`download` instead. """ warnings.warn( "LivySession.read is deprecated and will be removed in a future " "version. Use LivySession.download instead.", DeprecationWarning, ) return self.download(dataframe_name) def download_sql(self, query: str) -> pandas.DataFrame: """Evaluate a Spark SQL query and download the result. :param query: The Spark SQL query to evaluate. """ if self.kind != SessionKind.SQL: raise ValueError("not a SQL session") output = self._execute(query) output.raise_for_status() if output.json is None: raise RuntimeError("statement had no JSON output") return _dataframe_from_json_output(output.json) def read_sql(self, code: str) -> pandas.DataFrame: """Evaluate a Spark SQL statement and retrieve the result. :param code: The Spark SQL statement to evaluate. .. deprecated:: 0.8.0 Use :meth:`download_sql` instead. """ warnings.warn( "LivySession.read_sql is deprecated and will be removed in a " "future version. Use LivySession.download_sql instead.", DeprecationWarning, ) return self.download_sql(code) def upload(self, dataframe_name: str, data: pandas.DataFrame) -> None: """Upload a pandas dataframe to a Spark dataframe in the session. :param dataframe_name: The name of the Spark dataframe to create. :param data: The pandas dataframe to upload. """ code = _spark_create_dataframe_code(self.kind, dataframe_name, data) output = self._execute(code) output.raise_for_status() def _execute(self, code: str) -> Output: statement = self.client.create_statement(self.session_id, code) intervals = polling_intervals([0.1, 0.2, 0.3, 0.5], 1.0) def waiting_for_output(statement): not_finished = statement.state in { StatementState.WAITING, StatementState.RUNNING, } available = statement.state == StatementState.AVAILABLE return not_finished or (available and statement.output is None) while waiting_for_output(statement): time.sleep(next(intervals)) statement = self.client.get_statement(statement.session_id, statement.statement_id) if statement.output is None: raise RuntimeError("statement had no output") return statement.output
class LivySession: """Manages a remote Livy session and high-level interactions with it. The py_files, files, jars and archives arguments are lists of URLs, e.g. ["s3://bucket/object", "hdfs://path/to/file", ...] and must be reachable by the Spark driver process. If the provided URL has no scheme, it's considered to be relative to the default file system configured in the Livy server. URLs in the py_files argument are copied to a temporary staging area and inserted into Python's sys.path ahead of the standard library paths. This allows you to import .py, .zip and .egg files in Python. URLs for jars, py_files, files and archives arguments are all copied to the same working directory on the Spark cluster. The driver_memory and executor_memory arguments have the same format as JVM memory strings with a size unit suffix ("k", "m", "g" or "t") (e.g. 512m, 2g). See https://spark.apache.org/docs/latest/configuration.html for more information on Spark configuration properties. :param url: The URL of the Livy server. :param auth: A requests-compatible auth object to use when making requests. :param verify: Either a boolean, in which case it controls whether we verify the server’s TLS certificate, or a string, in which case it must be a path to a CA bundle to use. Defaults to ``True``. :param kind: The kind of session to create. :param proxy_user: User to impersonate when starting the session. :param jars: URLs of jars to be used in this session. :param py_files: URLs of Python files to be used in this session. :param files: URLs of files to be used in this session. :param driver_memory: Amount of memory to use for the driver process (e.g. '512m'). :param driver_cores: Number of cores to use for the driver process. :param executor_memory: Amount of memory to use per executor process (e.g. '512m'). :param executor_cores: Number of cores to use for each executor. :param num_executors: Number of executors to launch for this session. :param archives: URLs of archives to be used in this session. :param queue: The name of the YARN queue to which submitted. :param name: The name of this session. :param spark_conf: Spark configuration properties. :param echo: Whether to echo output printed in the remote session. Defaults to ``True``. :param check: Whether to raise an exception when a statement in the remote session fails. Defaults to ``True``. """ def __init__( self, url: str, auth: Auth = None, verify: Verify = True, kind: SessionKind = SessionKind.PYSPARK, proxy_user: str = None, jars: List[str] = None, py_files: List[str] = None, files: List[str] = None, driver_memory: str = None, driver_cores: int = None, executor_memory: str = None, executor_cores: int = None, num_executors: int = None, archives: List[str] = None, queue: str = None, name: str = None, spark_conf: Dict[str, Any] = None, echo: bool = True, check: bool = True, ) -> None: self.client = LivyClient(url, auth, verify=verify) self.kind = kind self.proxy_user = proxy_user self.jars = jars self.py_files = py_files self.files = files self.driver_memory = driver_memory self.driver_cores = driver_cores self.executor_memory = executor_memory self.executor_cores = executor_cores self.num_executors = num_executors self.archives = archives self.queue = queue self.name = name self.spark_conf = spark_conf self.echo = echo self.check = check self.session_id: Optional[int] = None def __enter__(self) -> "LivySession": self.start() return self def __exit__(self, exc_type, exc_value, traceback) -> None: self.close() def start(self) -> None: """Create the remote Spark session and wait for it to be ready.""" session = self.client.create_session( self.kind, self.proxy_user, self.jars, self.py_files, self.files, self.driver_memory, self.driver_cores, self.executor_memory, self.executor_cores, self.num_executors, self.archives, self.queue, self.name, self.spark_conf, ) self.session_id = session.session_id not_ready = {SessionState.NOT_STARTED, SessionState.STARTING} intervals = polling_intervals([0.1, 0.2, 0.3, 0.5], 1.0) while self.state in not_ready: time.sleep(next(intervals)) @property def state(self) -> SessionState: """The state of the managed Spark session.""" if self.session_id is None: raise ValueError("session not yet started") session = self.client.get_session(self.session_id) if session is None: raise ValueError("session not found - it may have been shut down") return session.state def close(self) -> None: """Kill the managed Spark session.""" if self.session_id is not None: self.client.delete_session(self.session_id) self.client.close() def run(self, code: str) -> Output: """Run some code in the managed Spark session. :param code: The code to run. """ output = self._execute(code) if self.echo and output.text: print(output.text) if self.check: output.raise_for_status() return output def read(self, dataframe_name: str) -> pandas.DataFrame: """Evaluate and retrieve a Spark dataframe in the managed session. :param dataframe_name: The name of the Spark dataframe to read. """ code = serialise_dataframe_code(dataframe_name, self.kind) output = self._execute(code) output.raise_for_status() if output.text is None: raise RuntimeError("statement had no text output") return deserialise_dataframe(output.text) def read_sql(self, code: str) -> pandas.DataFrame: """Evaluate a Spark SQL satatement and retrieve the result. :param code: The Spark SQL statement to evaluate. """ if self.kind != SessionKind.SQL: raise ValueError("not a SQL session") output = self._execute(code) output.raise_for_status() if output.json is None: raise RuntimeError("statement had no JSON output") return dataframe_from_json_output(output.json) def _execute(self, code: str) -> Output: if self.session_id is None: raise ValueError("session not yet started") statement = self.client.create_statement(self.session_id, code) intervals = polling_intervals([0.1, 0.2, 0.3, 0.5], 1.0) def waiting_for_output(statement): not_finished = statement.state in { StatementState.WAITING, StatementState.RUNNING, } available = statement.state == StatementState.AVAILABLE return not_finished or (available and statement.output is None) while waiting_for_output(statement): time.sleep(next(intervals)) statement = self.client.get_statement(statement.session_id, statement.statement_id) if statement.output is None: raise RuntimeError("statement had no output") return statement.output
class LivySession: def __init__(self, url: str, kind: SessionKind = SessionKind.PYSPARK, spark_conf: Dict[str, Any] = None, echo: bool = True, check: bool = True) -> None: self.client = LivyClient(url) self.kind = kind self.echo = echo self.check = check self.session_id: Optional[int] = None self.spark_conf = spark_conf def __enter__(self) -> 'LivySession': self.start() return self def __exit__(self, exc_type, exc_value, traceback) -> None: self.close() def start(self) -> None: session = self.client.create_session(self.kind, self.spark_conf) self.session_id = session.session_id not_ready = {SessionState.NOT_STARTED, SessionState.STARTING} intervals = polling_intervals([0.1, 0.2, 0.3, 0.5], 1.0) while self.state in not_ready: time.sleep(next(intervals)) @property def state(self) -> SessionState: if self.session_id is None: raise ValueError('session not yet started') session = self.client.get_session(self.session_id) if session is None: raise ValueError('session not found - it may have been shut down') return session.state def close(self) -> None: if self.session_id is not None: self.client.delete_session(self.session_id) self.client.close() def run(self, code: str) -> Output: output = self._execute(code) if self.echo and output.text: print(output.text) if self.check: output.raise_for_status() return output def read(self, dataframe_name: str) -> pandas.DataFrame: code = serialise_dataframe_code(dataframe_name, self.kind) output = self._execute(code) output.raise_for_status() if output.text is None: raise RuntimeError('statement had no text output') return deserialise_dataframe(output.text) def read_sql(self, code: str) -> pandas.DataFrame: if self.kind != SessionKind.SQL: raise ValueError('not a SQL session') output = self._execute(code) output.raise_for_status() if output.json is None: raise RuntimeError('statement had no JSON output') return dataframe_from_json_output(output.json) def _execute(self, code: str) -> Output: if self.session_id is None: raise ValueError('session not yet started') statement = self.client.create_statement(self.session_id, code) not_finished = {StatementState.WAITING, StatementState.RUNNING} intervals = polling_intervals([0.1, 0.2, 0.3, 0.5], 1.0) while statement.state in not_finished: time.sleep(next(intervals)) statement = self.client.get_statement(statement.session_id, statement.statement_id) if statement.output is None: raise RuntimeError('statement had no output') return statement.output
class LivySession: """Manages a remote Livy session and high-level interactions with it. :param url: The URL of the Livy server. :param kind: The kind of session to create. :param proxy_user: User to impersonate when starting the session. :param spark_conf: Spark configuration properties. :param echo: Whether to echo output printed in the remote session. Defaults to ``True``. :param check: Whether to raise an exception when a statement in the remote session fails. Defaults to ``True``. """ def __init__( self, url: str, auth: Auth = None, kind: SessionKind = SessionKind.PYSPARK, proxy_user: str = None, spark_conf: Dict[str, Any] = None, echo: bool = True, check: bool = True, ) -> None: self.client = LivyClient(url, auth) self.kind = kind self.proxy_user = proxy_user self.spark_conf = spark_conf self.echo = echo self.check = check self.session_id: Optional[int] = None def __enter__(self) -> "LivySession": self.start() return self def __exit__(self, exc_type, exc_value, traceback) -> None: self.close() def start(self) -> None: """Create the remote Spark session and wait for it to be ready.""" session = self.client.create_session(self.kind, self.proxy_user, self.spark_conf) self.session_id = session.session_id not_ready = {SessionState.NOT_STARTED, SessionState.STARTING} intervals = polling_intervals([0.1, 0.2, 0.3, 0.5], 1.0) while self.state in not_ready: time.sleep(next(intervals)) @property def state(self) -> SessionState: """The state of the managed Spark session.""" if self.session_id is None: raise ValueError("session not yet started") session = self.client.get_session(self.session_id) if session is None: raise ValueError("session not found - it may have been shut down") return session.state def close(self) -> None: """Kill the managed Spark session.""" if self.session_id is not None: self.client.delete_session(self.session_id) self.client.close() def run(self, code: str) -> Output: """Run some code in the managed Spark session. :param code: The code to run. """ output = self._execute(code) if self.echo and output.text: print(output.text) if self.check: output.raise_for_status() return output def read(self, dataframe_name: str) -> pandas.DataFrame: """Evaluate and retrieve a Spark dataframe in the managed session. :param dataframe_name: The name of the Spark dataframe to read. """ code = serialise_dataframe_code(dataframe_name, self.kind) output = self._execute(code) output.raise_for_status() if output.text is None: raise RuntimeError("statement had no text output") return deserialise_dataframe(output.text) def read_sql(self, code: str) -> pandas.DataFrame: """Evaluate a Spark SQL satatement and retrieve the result. :param code: The Spark SQL statement to evaluate. """ if self.kind != SessionKind.SQL: raise ValueError("not a SQL session") output = self._execute(code) output.raise_for_status() if output.json is None: raise RuntimeError("statement had no JSON output") return dataframe_from_json_output(output.json) def _execute(self, code: str) -> Output: if self.session_id is None: raise ValueError("session not yet started") statement = self.client.create_statement(self.session_id, code) intervals = polling_intervals([0.1, 0.2, 0.3, 0.5], 1.0) def waiting_for_output(statement): not_finished = statement.state in { StatementState.WAITING, StatementState.RUNNING, } available = statement.state == StatementState.AVAILABLE return not_finished or (available and statement.output is None) while waiting_for_output(statement): time.sleep(next(intervals)) statement = self.client.get_statement(statement.session_id, statement.statement_id) if statement.output is None: raise RuntimeError("statement had no output") return statement.output