bug fixes and simplifying interfaces

pull/1/head
Steve Nyemba 3 years ago
parent ccc05acc01
commit b239a5149f

@ -8,7 +8,7 @@ def read(fname):
return open(os.path.join(os.path.dirname(__file__), fname)).read()
args = {
"name":"data-transport",
"version":"1.4.1",
"version":"1.4.4",
"author":"The Phi Technology LLC","author_email":"info@the-phi.com",
"license":"MIT",
"packages":["transport"]}

@ -1,45 +1,24 @@
"""
Data Transport - 1.0
Steve L. Nyemba, The Phi Technology LLC
Data Transport, The Phi Technology LLC
Steve L. Nyemba, steve@the-phi.com
This module is designed to serve as a wrapper to a set of supported data stores :
This library is designed to serve as a wrapper to a set of supported data stores :
- couchdb
- mongodb
- Files (character delimited)
- Queues (RabbmitMq)
- Session (Flask)
- s3
- sqlite
The supported operations are read/write and providing meta data to the calling code
Requirements :
pymongo
boto
couldant
The configuration for the data-store is as follows :
couchdb:
{
args:{
url:<url>,
username:<username>,
password:<password>,
dbname:<database>,
doc:<document id>
}
}
RabbitMQ:
{
}
Mongodb:
{
args:{
host:<url>, #localhost:27017
username:<username>,
password:<password>,
dbname:<database>,
doc:<document id>s
}
}
e.g:
mongodb
provider:'mongodb',[port:27017],[host:localhost],db:<name>,doc:<_name>,context:<read|write>
"""
__author__ = 'The Phi Technology'
import pandas as pd
@ -90,9 +69,17 @@ class factory :
PROVIDERS['mongodb'] = PROVIDERS['mongo']
PROVIDERS['couchdb'] = PROVIDERS['couch']
PROVIDERS['sqlite3'] = PROVIDERS['sqlite']
@staticmethod
def instance(**args):
def instance(**_args):
if 'type' in _args :
#
# Legacy code being returned
return factory._instance(**_args);
else:
return instance(**_args)
@staticmethod
def _instance(**args):
"""
This class will create an instance of a transport when providing
:type name of the type we are trying to create
@ -131,7 +118,7 @@ def instance(**_args):
"""
provider = _args['provider']
context = _args['context']
context = _args['context']if 'context' in _args else None
_id = context if context in ['read','write'] else 'read'
if _id :
args = {'provider':_id}
@ -142,6 +129,7 @@ def instance(**_args):
args[key] = value
#
#
args = dict(args,**_args)
# print (provider in factory.PROVIDERS)
@ -149,6 +137,7 @@ def instance(**_args):
pointer = factory.PROVIDERS[provider]['class'][_id]
else:
pointer = sql.SQLReader if _id == 'read' else sql.SQLWriter
return pointer(**args)
return None
return None

@ -111,13 +111,47 @@ class DiskWriter(Writer):
pass
finally:
DiskWriter.THREAD_LOCK.release()
class SQLiteReader (DiskReader):
def __init__(self,**args):
DiskReader.__init__(self,**args)
self.path = args['database'] if 'database' in args else args['path']
self.conn = sqlite3.connect(self.path,isolation_level=None)
class SQLite :
def __init__(self,**_args) :
self.path = _args['database'] if 'database' in _args else _args['path']
self.conn = sqlite3.connect(self.path,isolation_level="IMMEDIATE")
self.conn.row_factory = sqlite3.Row
self.table = args['table']
self.fields = _args['fields'] if 'fields' in _args else []
def has (self,**_args):
found = False
try:
if 'table' in _args :
table = _args['table']
sql = "SELECT * FROM :table limit 1".replace(":table",table)
_df = pd.read_sql(sql,self.conn)
found = _df.columns.size > 0
except Exception as e:
pass
return found
def close(self):
try:
self.conn.close()
except Exception as e :
print(e)
def apply(self,sql):
try:
if not sql.lower().startswith('select'):
cursor = self.conn.cursor()
cursor.execute(sql)
cursor.close()
self.conn.commit()
else:
return pd.read_sql(sql,self.conn)
except Exception as e:
print (e)
class SQLiteReader (SQLite,DiskReader):
def __init__(self,**args):
super().__init__(**args)
# DiskReader.__init__(self,**args)
# self.path = args['database'] if 'database' in args else args['path']
# self.conn = sqlite3.connect(self.path,isolation_level=None)
# self.conn.row_factory = sqlite3.Row
self.table = args['table'] if 'table' in args else None
def read(self,**args):
if 'sql' in args :
sql = args['sql']
@ -135,7 +169,7 @@ class SQLiteReader (DiskReader):
except Exception as e :
pass
class SQLiteWriter(DiskWriter) :
class SQLiteWriter(SQLite,DiskWriter) :
connection = None
LOCK = Lock()
def __init__(self,**args):
@ -143,12 +177,13 @@ class SQLiteWriter(DiskWriter) :
:path
:fields json|csv
"""
DiskWriter.__init__(self,**args)
self.table = args['table']
# DiskWriter.__init__(self,**args)
super().__init__(**args)
self.table = args['table'] if 'table' in args else None
self.conn = sqlite3.connect(self.path,isolation_level="IMMEDIATE")
self.conn.row_factory = sqlite3.Row
self.fields = args['fields'] if 'fields' in args else []
# self.conn = sqlite3.connect(self.path,isolation_level="IMMEDIATE")
# self.conn.row_factory = sqlite3.Row
# self.fields = args['fields'] if 'fields' in args else []
if self.fields and not self.isready():
self.init(self.fields)
@ -185,7 +220,7 @@ class SQLiteWriter(DiskWriter) :
if not self.fields :
self.init(list(info.keys()))
if type(info) == object :
if type(info) == dict :
info = [info]
elif type(info) == pd.DataFrame :
info = info.to_dict(orient='records')
@ -196,6 +231,8 @@ class SQLiteWriter(DiskWriter) :
cursor = self.conn.cursor()
sql = " " .join(["INSERT INTO ",self.table,"(", ",".join(self.fields) ,")", "values(:values)"])
for row in info :
print (row)
print (row.values())
stream =["".join(["",value,""]) if type(value) == str else value for value in row.values()]
stream = json.dumps(stream).replace("[","").replace("]","")

@ -75,7 +75,15 @@ class SQLRW :
_info['securityLevel'] = 0
del _info['dbname']
self.conn = _handler.connect(**_info)
def has(self,**_args):
found = False
try:
table = _args['table']
sql = "SELECT * FROM :table LIMIT 1".replace(":table",table)
found = pd.read_sql(sql,self.conn).shape[0]
except Exception as e:
pass
return found
def isready(self):
_sql = "SELECT * FROM :table LIMIT 1".replace(":table",self.table)
try:
@ -201,8 +209,14 @@ class SQLWriter(SQLRW,Writer):
# values = [ "".join(["'",str(_row[key]),"'"]) if np.nan(_row[key]).isnumeric() else str(_row[key]) for key in _row]
# print (values)
query = _sql.replace(":fields",",".join(fields)).replace(":values",values)
cursor.execute(query,_row.values())
if type(info) == pd.DataFrame :
_values = info.values.tolist()
elif type(info) == list and type(info[0]) == dict:
print ('........')
_values = [tuple(item.values()) for item in info]
else:
_values = info;
cursor.execute(query,_values)
pass
@ -210,14 +224,23 @@ class SQLWriter(SQLRW,Writer):
_fields = ",".join(self.fields)
# _sql = _sql.replace(":fields",_fields)
# _sql = _sql.replace(":values",",".join(["%("+name+")s" for name in self.fields]))
_sql = _sql.replace("(:fields)","")
# _sql = _sql.replace("(:fields)","")
_sql = _sql.replace(":fields",_fields)
values = ", ".join("?"*len(self.fields)) if self._provider == 'netezza' else ",".join(["%s" for name in self.fields])
_sql = _sql.replace(":values",values)
# for row in info :
# values = ["'".join(["",value,""]) if not str(value).isnumeric() else value for value in row.values()]
cursor.executemany(_sql,info)
if type(info) == pd.DataFrame :
_info = info[self.fields].values.tolist()
elif type(info) == dict :
_info = info.values()
else:
# _info = []
_info = pd.DataFrame(info)[self.fields].values.tolist()
# for row in info :
# if type(row) == dict :
# _info.append( list(row.values()))
cursor.executemany(_sql,_info)
# self.conn.commit()
except Exception as e:
@ -250,6 +273,13 @@ class BigQuery:
client = bq.Client.from_service_account_json(self.path)
ref = client.dataset(self.dataset).table(table)
return client.get_table(ref).schema
def has(self,**_args):
found = False
try:
found = self.meta(**_args) is not None
except Exception as e:
pass
return found
class BQReader(BigQuery,Reader) :
def __init__(self,**_args):

Loading…
Cancel
Save