aboutsummaryrefslogtreecommitdiff
path: root/wqflask
diff options
context:
space:
mode:
authorAlexander_Kabui2023-01-24 18:27:13 +0300
committerAlexander_Kabui2023-01-24 18:27:13 +0300
commit47c19fcbd6c3e477285c765b4c02b5d177cd6206 (patch)
treeb77a9cc9941949e207d276558b979fbc860b3965 /wqflask
parent7012b68b1ee054004a6de8252e752f3d67612346 (diff)
downloadgenenetwork2-47c19fcbd6c3e477285c765b4c02b5d177cd6206.tar.gz
error handling for read and writee
Diffstat (limited to 'wqflask')
-rw-r--r--wqflask/wqflask/correlation/correlation_file_generator.py62
1 files changed, 35 insertions, 27 deletions
diff --git a/wqflask/wqflask/correlation/correlation_file_generator.py b/wqflask/wqflask/correlation/correlation_file_generator.py
index 3b6a0984..4d8a2f23 100644
--- a/wqflask/wqflask/correlation/correlation_file_generator.py
+++ b/wqflask/wqflask/correlation/correlation_file_generator.py
@@ -2,12 +2,13 @@ from urllib.parse import urlparse
import pymysql as mdb
import os
import csv
-import lmdb
+import lmdb
import os
-import tempfile
-import numpy as np
-from io import BytesIO
+import functools
+import tempfile
+import numpy as np
+from io import BytesIO
def parse_db_url():
@@ -16,7 +17,6 @@ def parse_db_url():
always expected"""
parsed_db = urlparse(SQL_URI)
-
return (
parsed_db.hostname, parsed_db.username, parsed_db.password,
parsed_db.path[1:], 3306)
@@ -26,7 +26,7 @@ def parse_db_url():
def database_connector():
"""function to create db connector"""
host, user, passwd, db_name, db_port = parse_db_url()
- return mdb.connect(host=host,user=user,password=passwd,database=db_name)
+ return mdb.connect(host=host, user=user, password=passwd, database=db_name)
def get_probesetfreezes(conn, inbredsetid=1):
@@ -88,8 +88,8 @@ def get_probesetfreeze(conn, probes):
def query_for_last_modification():
- pass
- """
+ pass
+ """
SELECT database_name,table_name,last_update FROM
mysql.innodb_table_stats a,
@@ -111,18 +111,18 @@ WHERE a.database_name = b.db_last_update_name
def parse_dataset(results):
- ids = ["ID"]
- data = {}
- for (trait, strain,val) in results:
- if strain not in ids:
- ids.append(strain)
+ ids = ["ID"]
+ data = {}
+ for (trait, strain,val) in results:
+ if strain not in ids:
+ ids.append(strain)
- if trait in data:
- data[trait].append(val)
- else:
- data[trait] = [trait,val]
+ if trait in data:
+ data[trait].append(val)
+ else:
+ data[trait] = [trait,val]
- return (data,ids)
+ return (data,ids)
# above refactor the code
@@ -164,27 +164,35 @@ def bytes_to_array(b: bytes) -> np.ndarray:
def lmdb_file_generator(file_name,data:np.ndarray):
- #create unique filename
+ # create unique filename
# pass
+
+def lmdb_error_handler(func):
+ @functools.wraps(func)
+ def inner_function(*args, **kwargs):
+ try:
+ return func(*args, **kwargs)
+ except lmdb.Error as error:
+ print(f"{func.__name__} >>>> error . {str(error)}")
+ return None
+ return inner_function
+
+
def create_dataset(file_name, db_name, dataset: np.ndarray):
# map size int(ie12)
- with lmdb.open(file_name, map_size=(dataset.nbytes*10)) as env:
+ with lmdb.open(file_name, map_size=dataset.nbytes*10) as env:
with env.begin(write=True) as txn:
txn.put(db_name.encode(), array_to_bytes(dataset))
-
return (file_name, db_name)
+@lmdb_error_handler
def read_dataset(file_name, db_name):
with lmdb.open(file_name, readonly=True, create=False) as env:
with env.begin() as txn:
- ndaray_bytes = txn.get(db_name.encode())
- # error handling
- return bytes_to_array(ndaray_bytes)
-
-
-#simple lmdb file \ No newline at end of file
+ results = txn.get(db_name.encode())
+ return bytes_to_array(results) if results else None \ No newline at end of file