# Licensed under a 3-clause BSD style license - see LICENSE.rst
"""
=============
Gaia TAP plus
=============
@author: Juan Carlos Segovia
@contact: juan.carlos.segovia@sciops.esa.int
European Space Astronomy Centre (ESAC)
European Space Agency (ESA)
Created on 30 jun. 2016
Modified on 18 Ene. 2022 by mhsarmiento
"""
from requests import HTTPError
from astroquery.utils.tap import TapPlus
from astroquery.utils import commons
from astroquery import log
from astropy import units
from astropy.units import Quantity
import zipfile
from astroquery.utils.tap import taputils
from . import conf
import os
from datetime import datetime
import shutil
import astroquery.utils.tap.model.modelutils as modelutils
from astropy.io import votable
from astropy.io import fits
from astropy.table import Table
from astropy import units as u
import warnings
from astroquery.exceptions import InputWarning
[docs]class GaiaClass(TapPlus):
"""
Proxy class to default TapPlus object (pointing to Gaia Archive)
"""
MAIN_GAIA_TABLE = None
MAIN_GAIA_TABLE_RA = conf.MAIN_GAIA_TABLE_RA
MAIN_GAIA_TABLE_DEC = conf.MAIN_GAIA_TABLE_DEC
ROW_LIMIT = conf.ROW_LIMIT
VALID_DATALINK_RETRIEVAL_TYPES = conf.VALID_DATALINK_RETRIEVAL_TYPES
def __init__(self, tap_plus_conn_handler=None,
datalink_handler=None,
gaia_tap_server='https://gea.esac.esa.int/',
gaia_data_server='https://gea.esac.esa.int/',
tap_server_context="tap-server",
data_server_context="data-server",
verbose=False):
super(GaiaClass, self).__init__(url=gaia_tap_server,
server_context=tap_server_context,
tap_context="tap",
upload_context="Upload",
table_edit_context="TableTool",
data_context="data",
datalink_context="datalink",
connhandler=tap_plus_conn_handler,
verbose=verbose)
# Data uses a different TapPlus connection
if datalink_handler is None:
self.__gaiadata = TapPlus(url=gaia_data_server,
server_context=data_server_context,
tap_context="tap",
upload_context="Upload",
table_edit_context="TableTool",
data_context="data",
datalink_context="datalink",
verbose=verbose)
else:
self.__gaiadata = datalink_handler
[docs] def login(self, user=None, password=None, credentials_file=None,
verbose=False):
"""Performs a login.
User and password arguments can be used or a file that contains
username and password
(2 lines: one for username and the following one for the password).
If no arguments are provided, a prompt asking for username and
password will appear.
Parameters
----------
user : str, default None
login name
password : str, default None
user password
credentials_file : str, default None
file containing user and password in two lines
verbose : bool, optional, default 'False'
flag to display information about the process
"""
try:
log.info("Login to gaia TAP server")
TapPlus.login(self, user=user, password=password,
credentials_file=credentials_file,
verbose=verbose)
except HTTPError as err:
log.error("Error logging in TAP server")
return
u = self._TapPlus__user
p = self._TapPlus__pwd
try:
log.info("Login to gaia data server")
TapPlus.login(self.__gaiadata, user=u, password=p,
verbose=verbose)
except HTTPError as err:
log.error("Error logging in data server")
log.error("Logging out from TAP server")
TapPlus.logout(self, verbose=verbose)
[docs] def login_gui(self, verbose=False):
"""Performs a login using a GUI dialog
Parameters
----------
verbose : bool, optional, default 'False'
flag to display information about the process
"""
try:
log.info("Login to gaia TAP server")
TapPlus.login_gui(self, verbose=verbose)
except HTTPError as err:
log.error("Error logging in TAP server")
return
u = self._TapPlus__user
p = self._TapPlus__pwd
try:
log.info("Login to gaia data server")
TapPlus.login(self.__gaiadata, user=u, password=p,
verbose=verbose)
except HTTPError as err:
log.error("Error logging in data server")
log.error("Logging out from TAP server")
TapPlus.logout(self, verbose=verbose)
[docs] def logout(self, verbose=False):
"""Performs a logout
Parameters
----------
verbose : bool, optional, default 'False'
flag to display information about the process
"""
try:
TapPlus.logout(self, verbose=verbose)
except HTTPError as err:
log.error("Error logging out TAP server")
return
log.info("Gaia TAP server logout OK")
try:
TapPlus.logout(self.__gaiadata, verbose=verbose)
log.info("Gaia data server logout OK")
except HTTPError as err:
log.error("Error logging out data server")
[docs] def load_data(self, ids, data_release=None, data_structure='INDIVIDUAL', retrieval_type="ALL", valid_data=True,
band=None, avoid_datatype_check=False, format="votable", output_file=None,
overwrite_output_file=False, verbose=False):
"""Loads the specified table
TAP+ only
Parameters
----------
ids : str list, mandatory
list of identifiers
data_release: str, optional, default None
data release from which data should be taken. E.g. 'Gaia DR2'
By default, it takes the current default one.
data_structure: str, optional, default 'INDIVIDUAL'
it can be 'INDIVIDUAL', 'COMBINED', 'RAW':
'INDIVIDUAL' means products are provided in separate files for each sourceId. All files are zipped in a single
bundle, even if only one source/file is considered
'COMBINED' means products are provided in a single file concatenating the data of all sourceIds together.
How this is organised depends on the chosen format
'RAW' means products are provided following a Data Model similar to that used in the MDB, meaning in
particular that parameters stored as arrays will remain as such. Like in the COMBINED structure, a single
file is provided for the data of all sourceIds together, but in this case there will be always be one
row per sourceId
retrieval_type : str, optional, default 'ALL' to retrieve all data from the list of sources
retrieval type identifier. For GAIA DR2 possible values are ['EPOCH_PHOTOMETRY']
For future GAIA DR3 (Once published), possible values will be ['EPOC_PHOTOMETRY', 'RVS', 'XP_CONTINUOUS',
'XP_SAMPLED', 'MCMC_GSPPHOT' or 'MCMC_MSC']
valid_data : bool, optional, default True
By default, the epoch photometry service returns only valid data,
that is, all data rows where flux is not null and
rejected_by_photometry flag is not true. In order to retrieve
all data associated to a given source without this filter,
this request parameter should be included (valid_data=False)
band : str, optional, default None, valid values: G, BP, RP
By default, the epoch photometry service returns all the
available photometry bands for the requested source.
This parameter allows to filter the output lightcurve by its band.
avoid_datatype_check: boolean, optional, default False.
By default, this value will be set to False. If it is set to 'true'
the Datalink items tags will not be checked.
format : str, optional, default 'votable'
loading format. Other available formats are 'csv', 'ecsv','json','votable_plain' and 'fits'
output_file : string, optional, default None
file where the results are saved.
If it is not provided, the http response contents are returned.
overwrite_output_file : boolean, optional, default False
To overwrite the output_file if it already exists.
verbose : bool, optional, default 'False'
flag to display information about the process
Returns
-------
A table object
"""
if retrieval_type is None:
raise ValueError("Missing mandatory argument 'retrieval_type'")
now = datetime.now()
now_formatted = now.strftime("%Y%m%d_%H%M%S")
temp_dirname = "temp_" + now_formatted
downloadname_formated = "download_" + now_formatted
output_file_specified = False
if output_file is None:
output_file = os.path.join(os.getcwd(), temp_dirname, downloadname_formated)
else:
output_file_specified = True
output_file = os.path.abspath(output_file)
if not overwrite_output_file and os.path.exists(output_file):
raise ValueError(f"{output_file} file already exists. Please use overwrite_output_file='False' to "
f"overwrite output file.")
path = os.path.dirname(output_file)
if ids is None:
raise ValueError("Missing mandatory argument 'ids'")
if avoid_datatype_check is False:
# we need to check params
rt = str(retrieval_type).upper()
if rt != 'ALL' and rt not in self.VALID_DATALINK_RETRIEVAL_TYPES:
raise ValueError(f"Invalid mandatory argument 'retrieval_type'. Found {retrieval_type}, "
f"expected: 'ALL' or any of {self.VALID_DATALINK_RETRIEVAL_TYPES}")
params_dict = {}
if not valid_data or str(retrieval_type) == 'ALL':
params_dict['VALID_DATA'] = "false"
elif valid_data:
params_dict['VALID_DATA'] = "true"
if band is not None:
if band != 'G' and band != 'BP' and band != 'RP':
raise ValueError("Invalid band value '%s' (Valid values: " +
"'G', 'BP' and 'RP)" % band)
else:
params_dict['BAND'] = band
if isinstance(ids, str):
ids_arg = ids
else:
if isinstance(ids, int):
ids_arg = str(ids)
else:
ids_arg = ','.join(str(item) for item in ids)
params_dict['ID'] = ids_arg
if data_release is not None:
params_dict['RELEASE'] = data_release
params_dict['DATA_STRUCTURE'] = data_structure
params_dict['FORMAT'] = str(format)
params_dict['RETRIEVAL_TYPE'] = str(retrieval_type)
params_dict['USE_ZIP_ALWAYS'] = 'true'
if path != '':
try:
os.mkdir(path)
except FileExistsError:
log.error("Path %s already exist" % path)
except OSError:
log.error("Creation of the directory %s failed" % path)
try:
self.__gaiadata.load_data(params_dict=params_dict,
output_file=output_file,
verbose=verbose)
files = Gaia.__get_data_files(output_file=output_file, path=path)
except Exception as err:
raise err
finally:
if not output_file_specified:
shutil.rmtree(path)
if verbose:
if output_file_specified:
log.info("output_file = %s" % output_file)
log.debug("List of products available:")
# for key, value in files.items():
# print("Product =", key)
items = [key for key in files.keys()]
items.sort()
for item in items:
# print(f'* {item}')
if verbose:
log.debug("Product = " + item)
return files
@staticmethod
def __get_data_files(output_file, path):
files = {}
if zipfile.is_zipfile(output_file):
with zipfile.ZipFile(output_file, 'r') as zip_ref:
zip_ref.extractall(os.path.dirname(output_file))
# r=root, d=directories, f = files
for r, d, f in os.walk(path):
for file in f:
if '.fits' in file or '.xml' in file or '.csv' in file:
files[file] = os.path.join(r, file)
for key, value in files.items():
if '.fits' in key:
tables = []
with fits.open(value) as hduList:
# print(hduList)
num_hdus = len(hduList)
for i in range(1, num_hdus):
table = Table.read(hduList[i], format='fits')
Gaia.correct_table_units(table)
tables.append(table)
files[key] = tables
elif '.xml' in key:
tables = []
for table in votable.parse(value).iter_tables():
tables.append(table)
files[key] = tables
elif '.csv' in key:
tables = []
table = Table.read(value, format='ascii.csv',
fast_reader=False)
tables.append(table)
files[key] = tables
return files
[docs] def get_datalinks(self, ids, verbose=False):
"""Gets datalinks associated to the provided identifiers
TAP+ only
Parameters
----------
ids : str list, mandatory
list of identifiers
verbose : bool, optional, default 'False'
flag to display information about the process
Returns
-------
A table object
"""
return self.__gaiadata.get_datalinks(ids=ids, verbose=verbose)
def __query_object(self, coordinate, radius=None, width=None, height=None,
async_job=False, verbose=False, columns=[]):
"""Launches a job
TAP & TAP+
Parameters
----------
coordinate : astropy.coordinate, mandatory
coordinates center point
radius : astropy.units if no 'width' nor 'height' are provided
radius (deg)
width : astropy.units if no 'radius' is provided
box width
height : astropy.units if no 'radius' is provided
box height
async_job : bool, optional, default 'False'
executes the query (job) in asynchronous/synchronous mode (default
synchronous)
verbose : bool, optional, default 'False'
flag to display information about the process
columns: list, optional, default []
if empty, all columns will be selected
Returns
-------
The job results (astropy.table).
"""
coord = self.__getCoordInput(coordinate, "coordinate")
job = None
if radius is not None:
job = self.__cone_search(coord, radius, async_job=async_job,
verbose=verbose, columns=columns)
else:
raHours, dec = commons.coord_to_radec(coord)
ra = raHours * 15.0 # Converts to degrees
widthQuantity = self.__getQuantityInput(width, "width")
heightQuantity = self.__getQuantityInput(height, "height")
widthDeg = widthQuantity.to(units.deg)
heightDeg = heightQuantity.to(units.deg)
if columns:
columns = ','.join(map(str, columns))
else:
columns = "*"
query = """
SELECT
{row_limit}
DISTANCE(
POINT('ICRS', {ra_column}, {dec_column}),
POINT('ICRS', {ra}, {dec})
) as dist,
{columns}
FROM
{table_name}
WHERE
1 = CONTAINS(
POINT('ICRS', {ra_column}, {dec_column}),
BOX(
'ICRS',
{ra},
{dec},
{width},
{height}
)
)
ORDER BY
dist ASC
""".format(**{'row_limit': "TOP {0}".format(self.ROW_LIMIT) if self.ROW_LIMIT > 0 else "",
'ra_column': self.MAIN_GAIA_TABLE_RA, 'dec_column': self.MAIN_GAIA_TABLE_DEC,
'columns': columns, 'table_name': self.MAIN_GAIA_TABLE or conf.MAIN_GAIA_TABLE,
'ra': ra, 'dec': dec,
'width': widthDeg.value, 'height': heightDeg.value})
if async_job:
job = self.launch_job_async(query, verbose=verbose)
else:
job = self.launch_job(query, verbose=verbose)
return job.get_results()
[docs] def query_object(self, coordinate, radius=None, width=None, height=None,
verbose=False, columns=[]):
"""Launches a job
TAP & TAP+
Parameters
----------
coordinate : astropy.coordinates, mandatory
coordinates center point
radius : astropy.units if no 'width'/'height' are provided
radius (deg)
width : astropy.units if no 'radius' is provided
box width
height : astropy.units if no 'radius' is provided
box height
verbose : bool, optional, default 'False'
flag to display information about the process
columns: list, optional, default []
if empty, all columns will be selected
Returns
-------
The job results (astropy.table).
"""
return self.__query_object(coordinate, radius, width, height, async_job=False, verbose=verbose, columns=columns)
[docs] def query_object_async(self, coordinate, radius=None, width=None,
height=None, verbose=False, columns=[]):
"""Launches a job (async)
TAP & TAP+
Parameters
----------
coordinate : astropy.coordinates, mandatory
coordinates center point
radius : astropy.units if no 'width'/'height' are provided
radius
width : astropy.units if no 'radius' is provided
box width
height : astropy.units if no 'radius' is provided
box height
verbose : bool, optional, default 'False'
flag to display information about the process
columns: list, optional, default []
if empty, all columns will be selected
Returns
-------
The job results (astropy.table).
"""
return self.__query_object(coordinate, radius, width, height, async_job=True, verbose=verbose, columns=columns)
def __cone_search(self, coordinate, radius, table_name=None,
ra_column_name=MAIN_GAIA_TABLE_RA,
dec_column_name=MAIN_GAIA_TABLE_DEC,
async_job=False,
background=False,
output_file=None, output_format="votable", verbose=False,
dump_to_file=False,
columns=[]):
"""Cone search sorted by distance
TAP & TAP+
Parameters
----------
coordinate : astropy.coordinate, mandatory
coordinates center point
radius : astropy.units, mandatory
radius
table_name : str, optional, default main gaia table name doing the cone search against
ra_column_name : str, optional, default ra column in main gaia table
ra column doing the cone search against
dec_column_name : str, optional, default dec column in main gaia table
dec column doing the cone search against
async_job : bool, optional, default 'False'
executes the job in asynchronous/synchronous mode (default
synchronous)
background : bool, optional, default 'False'
when the job is executed in asynchronous mode, this flag specifies
whether the execution will wait until results are available
output_file : str, optional, default None
file name where the results are saved if dumpToFile is True.
If this parameter is not provided, the jobid is used instead
output_format : str, optional, default 'votable'
results format
verbose : bool, optional, default 'False'
flag to display information about the process
dump_to_file : bool, optional, default 'False'
if True, the results are saved in a file instead of using memory
columns: list, optional, default []
if empty, all columns will be selected
Returns
-------
A Job object
"""
radiusDeg = None
coord = self.__getCoordInput(coordinate, "coordinate")
raHours, dec = commons.coord_to_radec(coord)
ra = raHours * 15.0 # Converts to degrees
if radius is not None:
radiusQuantity = self.__getQuantityInput(radius, "radius")
radiusDeg = commons.radius_to_unit(radiusQuantity, unit='deg')
if columns:
columns = ','.join(map(str, columns))
else:
columns = "*"
query = """
SELECT
{row_limit}
{columns},
DISTANCE(
POINT('ICRS', {ra_column}, {dec_column}),
POINT('ICRS', {ra}, {dec})
) AS dist
FROM
{table_name}
WHERE
1 = CONTAINS(
POINT('ICRS', {ra_column}, {dec_column}),
CIRCLE('ICRS', {ra}, {dec}, {radius})
)
ORDER BY
dist ASC
""".format(**{'ra_column': ra_column_name,
'row_limit': "TOP {0}".format(self.ROW_LIMIT) if self.ROW_LIMIT > 0 else "",
'dec_column': dec_column_name, 'columns': columns, 'ra': ra, 'dec': dec,
'radius': radiusDeg,
'table_name': table_name or self.MAIN_GAIA_TABLE or conf.MAIN_GAIA_TABLE})
if async_job:
return self.launch_job_async(query=query,
output_file=output_file,
output_format=output_format,
verbose=verbose,
dump_to_file=dump_to_file,
background=background)
else:
return self.launch_job(query=query,
output_file=output_file,
output_format=output_format,
verbose=verbose,
dump_to_file=dump_to_file)
[docs] def cone_search(self, coordinate, radius=None,
table_name=None,
ra_column_name=MAIN_GAIA_TABLE_RA,
dec_column_name=MAIN_GAIA_TABLE_DEC,
output_file=None,
output_format="votable", verbose=False,
dump_to_file=False,
columns=[]):
"""Cone search sorted by distance (sync.)
TAP & TAP+
Parameters
----------
coordinate : astropy.coordinate, mandatory
coordinates center point
radius : astropy.units, mandatory
radius
table_name : str, optional, default main gaia table name doing the cone search against
ra_column_name : str, optional, default ra column in main gaia table
ra column doing the cone search against
dec_column_name : str, optional, default dec column in main gaia table
dec column doing the cone search against
output_file : str, optional, default None
file name where the results are saved if dumpToFile is True.
If this parameter is not provided, the jobid is used instead
output_format : str, optional, default 'votable'
results format
verbose : bool, optional, default 'False'
flag to display information about the process
dump_to_file : bool, optional, default 'False'
if True, the results are saved in a file instead of using memory
columns: list, optional, default []
if empty, all columns will be selected
Returns
-------
A Job object
"""
return self.__cone_search(coordinate,
radius=radius,
table_name=table_name,
ra_column_name=ra_column_name,
dec_column_name=dec_column_name,
async_job=False,
background=False,
output_file=output_file,
output_format=output_format,
verbose=verbose,
dump_to_file=dump_to_file, columns=columns)
[docs] def cone_search_async(self, coordinate, radius=None,
table_name=None,
ra_column_name=MAIN_GAIA_TABLE_RA,
dec_column_name=MAIN_GAIA_TABLE_DEC,
background=False,
output_file=None, output_format="votable",
verbose=False, dump_to_file=False, columns=[]):
"""Cone search sorted by distance (async)
TAP & TAP+
Parameters
----------
coordinate : astropy.coordinate, mandatory
coordinates center point
radius : astropy.units, mandatory
radius
table_name : str, optional, default main gaia table name doing the cone search against
ra_column_name : str, optional, default ra column in main gaia table
ra column doing the cone search against
dec_column_name : str, optional, default dec column in main gaia table
dec column doing the cone search against
background : bool, optional, default 'False'
when the job is executed in asynchronous mode, this flag
specifies whether
the execution will wait until results are available
output_file : str, optional, default None
file name where the results are saved if dumpToFile is True.
If this parameter is not provided, the jobid is used instead
output_format : str, optional, default 'votable'
results format
verbose : bool, optional, default 'False'
flag to display information about the process
dump_to_file : bool, optional, default 'False'
if True, the results are saved in a file instead of using memory
Returns
-------
A Job object
"""
return self.__cone_search(coordinate,
radius=radius,
table_name=table_name,
ra_column_name=ra_column_name,
dec_column_name=dec_column_name,
async_job=True,
background=background,
output_file=output_file,
output_format=output_format,
verbose=verbose,
dump_to_file=dump_to_file, columns=columns)
def __checkQuantityInput(self, value, msg):
if not (isinstance(value, str) or isinstance(value, units.Quantity)):
raise ValueError(f"{msg} must be either a string or astropy coordinates")
def __getQuantityInput(self, value, msg):
if value is None:
raise ValueError(f"Missing required argument: {msg}")
if not (isinstance(value, str) or isinstance(value, units.Quantity)):
raise ValueError(f"{msg} must be either a string or astropy.coordinates")
if isinstance(value, str):
q = Quantity(value)
return q
else:
return value
def __checkCoordInput(self, value, msg):
if not (isinstance(value, str) or isinstance(value,
commons.CoordClasses)):
raise ValueError(f"{msg} must be either a string or astropy.coordinates")
def __getCoordInput(self, value, msg):
if not (isinstance(value, str) or isinstance(value,
commons.CoordClasses)):
raise ValueError(f"{msg} must be either a string or astropy.coordinates")
if isinstance(value, str):
c = commons.parse_coordinates(value)
return c
else:
return value
[docs] @staticmethod
def correct_table_units(table):
for cn in table.colnames:
col = table[cn]
if isinstance(col.unit, u.UnrecognizedUnit):
try:
col.unit = u.Unit(col.unit.name.replace(".", " ").replace("'", ""))
except Exception as err:
pass
elif isinstance(col.unit, str):
col.unit = col.unit.replace(".", " ").replace("'", "")
[docs] def load_user(self, user_id, verbose=False):
"""Loads the specified user
TAP+ only
Parameters
----------
user_id : str, mandatory
user id to load
verbose : bool, optional, default 'False'
flag to display information about the process
Returns
-------
A user
"""
return self.is_valid_user(user_id=user_id,
verbose=verbose)
[docs] def cross_match(self, full_qualified_table_name_a=None,
full_qualified_table_name_b=None,
results_table_name=None,
radius=1.0,
background=False,
verbose=False):
"""Performs a cross-match between the specified tables
The result is a join table (stored in the user storage area)
with the identifies of both tables and the distance.
TAP+ only
Parameters
----------
full_qualified_table_name_a : str, mandatory
a full qualified table name (i.e. schema name and table name)
full_qualified_table_name_b : str, mandatory
a full qualified table name (i.e. schema name and table name)
results_table_name : str, mandatory
a table name without schema. The schema is set to the user one
radius : float (arc. seconds), optional, default 1.0
radius (valid range: 0.1-10.0)
verbose : bool, optional, default 'False'
flag to display information about the process
Returns
-------
Boolean indicating if the specified user is valid
"""
if full_qualified_table_name_a is None:
raise ValueError("Table name A argument is mandatory")
if full_qualified_table_name_b is None:
raise ValueError("Table name B argument is mandatory")
if results_table_name is None:
raise ValueError("Results table name argument is mandatory")
if radius < 0.1 or radius > 10.0:
raise ValueError(f"Invalid radius value. Found {radius}, valid range is: 0.1 to 10.0")
schemaA = taputils.get_schema_name(full_qualified_table_name_a)
if schemaA is None:
raise ValueError(f"Not found schema name in full qualified table A: '{full_qualified_table_name_a}'")
tableA = taputils.get_table_name(full_qualified_table_name_a)
schemaB = taputils.get_schema_name(full_qualified_table_name_b)
if schemaB is None:
raise ValueError(f"Not found schema name in full qualified table B: '{full_qualified_table_name_b}'")
tableB = taputils.get_table_name(full_qualified_table_name_b)
if taputils.get_schema_name(results_table_name) is not None:
raise ValueError("Please, do not specify schema for 'results_table_name'")
query = f"SELECT crossmatch_positional('{schemaA}','{tableA}','{schemaB}','{tableB}',{radius}, " \
f"'{results_table_name}') FROM dual;"
name = str(results_table_name)
return self.launch_job_async(query=query,
name=name,
output_file=None,
output_format="votable",
verbose=verbose,
dump_to_file=False,
background=background,
upload_resource=None,
upload_table_name=None)
[docs] def launch_job(self, query, name=None, output_file=None,
output_format="votable", verbose=False,
dump_to_file=False, upload_resource=None,
upload_table_name=None):
"""Launches a synchronous job
Parameters
----------
query : str, mandatory
query to be executed
name : str, optional, default None
custom name defined by the user for the job that is going to be created
output_file : str, optional, default None
file name where the results are saved if dumpToFile is True.
If this parameter is not provided, the jobid is used instead
output_format : str, optional, default 'votable'
results format. Available formats are: 'votable', 'votable_plain',
'fits', 'csv', 'ecsv' and 'json', default is 'votable'.
Returned results for 'votable' and 'fits' formats are compressed
gzip files.
verbose : bool, optional, default 'False'
flag to display information about the process
dump_to_file : bool, optional, default 'False'
if True, the results are saved in a file instead of using memory
upload_resource : str, optional, default None
resource to be uploaded to UPLOAD_SCHEMA
upload_table_name : str, optional, default None
resource temporary table name associated to the uploaded resource.
This argument is required if upload_resource is provided.
Returns
-------
A Job object
"""
return TapPlus.launch_job(self, query=query, name=name,
output_file=output_file,
output_format=output_format,
verbose=verbose,
dump_to_file=dump_to_file,
upload_resource=upload_resource,
upload_table_name=upload_table_name)
[docs] def launch_job_async(self, query, name=None, output_file=None,
output_format="votable", verbose=False,
dump_to_file=False, background=False,
upload_resource=None, upload_table_name=None,
autorun=True):
"""Launches an asynchronous job
Parameters
----------
query : str, mandatory
query to be executed
name : str, optional, default None
custom name defined by the user for the job that is going to be created
output_file : str, optional, default None
file name where the results are saved if dumpToFile is True.
If this parameter is not provided, the jobid is used instead
output_format : str, optional, default 'votable'
results format. Available formats are: 'votable', 'votable_plain',
'fits', 'csv' and 'json', default is 'votable'.
Returned results for 'votable' 'ecsv' and 'fits' format are compressed
gzip files.
verbose : bool, optional, default 'False'
flag to display information about the process
dump_to_file : bool, optional, default 'False'
if True, the results are saved in a file instead of using memory
background : bool, optional, default 'False'
when the job is executed in asynchronous mode, this flag specifies
whether the execution will wait until results are available
upload_resource : str, optional, default None
resource to be uploaded to UPLOAD_SCHEMA
upload_table_name : str, optional, default None
resource temporary table name associated to the uploaded resource.
This argument is required if upload_resource is provided.
autorun : boolean, optional, default True
if 'True', sets 'phase' parameter to 'RUN',
so the framework can start the job.
Returns
-------
A Job object
"""
return TapPlus.launch_job_async(self, query=query,
name=name,
output_file=output_file,
output_format=output_format,
verbose=verbose,
dump_to_file=dump_to_file,
background=background,
upload_resource=upload_resource,
upload_table_name=upload_table_name,
autorun=autorun)
Gaia = GaiaClass()