def read_args():
"""
Read arguments.
"""
parser = argparse.ArgumentParser()
parser.add_argument("-j", "--job",
action="store_true", default=False,
help='Sample running job info with command "bjobs -u all -r -UF".')
parser.add_argument("-q", "--queue",
action="store_true", default=False,
help='Sample queue info with command "bqueues".')
parser.add_argument("-H", "--host",
action="store_true", default=False,
help='Sample host info with command "bhosts".')
parser.add_argument("-l", "--load",
action="store_true", default=False,
help='Sample host load info with command "lsload".')
parser.add_argument("-u", "--user",
action="store_true", default=False,
help='Sample user info with command "busers".')
parser.add_argument("-i", "--interval",
type=int,
default=0,
help='Specify the sampling interval, unit is second. Sampling only once by default".')
args = parser.parse_args()
if args.interval < 0:
common.print_error('*Error*: interval "' + str(args.interval) + '": Cannot be less than "0".')
sys.exit(1)
return(args.job, args.queue, args.host, args.load, args.user, args.interval)
def get_sql_table_count(db_file, orig_conn, table_name):
"""
How many lines of the database table.
"""
count = 'N/A'
(result, conn, curs) = connect_preprocess(db_file, orig_conn)
if result == 'failed':
return (count)
try:
command = "SELECT count(*) FROM '" + str(table_name) + "'"
curs.execute(command)
return_list = curs.fetchall()
return_tuple = return_list[0]
count = return_tuple[0]
curs.close()
if orig_conn == '':
conn.close()
except Exception as error:
common.print_error(
'*Error* (get_sql_table_count) : Failed on getting table count fro table "'
+ str(table_name) + '" on db_file "' + str(db_file) + '": ' +
str(error))
return (count)
def connect_db_file(db_file, mode='read'):
result = 'passed'
conn = ''
if mode == 'write':
journal_db_file = str(db_file) + '-journal'
if os.path.exists(journal_db_file) and (mode == 'write'):
common.print_warning(
'*Warning*: database file "' + str(db_file) +
'" is on another connection, will not connect it.')
result = 'locked'
return (result, conn)
elif mode == 'read':
if not os.path.exists(db_file):
common.print_error('*Error*: "' + str(db_file) +
'" No such database file.')
result = 'failed'
return (result, conn)
try:
conn = sqlite3.connect(db_file)
except Exception as error:
common.print_error('*Error*: Failed on connecting database file "' +
str(db_file) + '": ' + str(error))
result = 'failed'
return (result, conn)
def insert_into_sql_table(db_file,
orig_conn,
table_name,
value_string,
commit=True):
"""
Insert new value into sql table.
"""
(result, conn, curs) = connect_preprocess(db_file, orig_conn, mode='write')
if (result == 'failed') or (result == 'locked'):
return
try:
command = "INSERT INTO '" + str(table_name) + "' VALUES " + str(
value_string)
curs.execute(command)
curs.close()
if commit:
conn.commit()
if orig_conn == '':
conn.close()
except Exception as error:
common.print_error(
'*Error* (insert_into_sql_table) : Failed on inserting specified values into table "'
+ str(table_name) + '" on db file "' + str(db_file) + '": ' +
str(error))
def get_sql_table_list(db_file, orig_conn):
"""
Get all of the tables from the specified db file.
"""
table_list = []
(result, conn, curs) = connect_preprocess(db_file, orig_conn)
if result == 'failed':
return (table_list)
try:
command = "SELECT name FROM sqlite_master WHERE type='table' ORDER BY name"
results = curs.execute(command)
all_items = results.fetchall()
for item in all_items:
(key, ) = item
table_list.append(key)
curs.close()
if orig_conn == '':
conn.close()
except Exception as error:
common.print_error(
'*Error* (get_sql_table_list) : Failed on getting table list on db_file "'
+ str(db_file) + '": ' + str(error))
return (table_list)
def create_sql_table(db_file, orig_conn, table_name, init_string, commit=True):
"""
Create a table if it not exists, initialization the setting.
"""
(result, conn, curs) = connect_preprocess(db_file, orig_conn, mode='write')
if (result == 'failed') or (result == 'locked'):
return
try:
command = "CREATE TABLE IF NOT EXISTS '" + str(
table_name) + "' " + str(init_string)
curs.execute(command)
curs.close()
if commit:
conn.commit()
if orig_conn == '':
conn.close()
except Exception as error:
common.print_error(
'*Error* (create_sql_table) : Failed on creating table "' +
str(table_name) + '" on db file "' + str(db_file) + '": ' +
str(error))
def delete_sql_table_rows(db_file,
orig_conn,
table_name,
row_id,
begin_line,
end_line,
commit=True):
"""
Delete specified table rows (from begin_line to end_line).
"""
(result, conn, curs) = connect_preprocess(db_file, orig_conn, mode='write')
if (result == 'failed') or (result == 'locked'):
return
try:
command = "DELETE FROM '" + str(table_name) + "' WHERE " + str(
row_id) + " IN (SELECT " + str(row_id) + " FROM '" + str(
table_name) + "' ORDER BY " + str(row_id) + " LIMIT " + str(
begin_line) + "," + str(end_line) + ")"
curs.execute(command)
curs.close()
if commit:
conn.commit()
if orig_conn == '':
conn.close()
except Exception as error:
common.print_error(
'*Error* (drop_sql_table) : Failed on deleting table "' +
str(table_name) + '" lines ' + str(begin_line) + '-' +
str(end_line) + ': ' + str(error))
def get_sql_table_key_list(db_file, orig_conn, table_name):
"""
Get all of the tables from the specified db file.
"""
key_list = []
(result, conn, curs) = connect_preprocess(db_file, orig_conn)
if result == 'failed':
return (key_list)
try:
command = "SELECT * FROM '" + str(table_name) + "'"
curs.execute(command)
key_list = [tuple[0] for tuple in curs.description]
curs.close()
if orig_conn == '':
conn.close()
except Exception as error:
common.print_error(
'*Error* (get_sql_table_key_list) : Failed on getting table key list on db_file "'
+ str(db_file) + '": ' + str(error))
return (key_list)
def get_sql_table_data(db_file, orig_conn, table_name, key_list=[], limit=0):
"""
With specified db_file-table_name, get all data from specified key_list.
"""
data_dic = {}
(result, conn, curs) = connect_preprocess(db_file, orig_conn)
if result == 'failed':
return (data_dic)
try:
command = "SELECT * FROM '" + str(table_name) + "'"
if limit != 0:
command = str(command) + ' limit ' + str(limit)
results = curs.execute(command)
all_items = results.fetchall()
table_key_list = [tuple[0] for tuple in curs.description]
curs.close()
if orig_conn == '':
conn.close()
if len(key_list) == 0:
key_list = table_key_list
else:
for key in key_list:
if key not in table_key_list:
common.print_error('*Error* (get_sql_table_data) : "' +
str(key) +
'": invalid key on specified key list.')
return (data_dic)
for item in all_items:
value_list = list(item)
for i in range(len(table_key_list)):
key = table_key_list[i]
if key in key_list:
value = value_list[i]
if key in data_dic.keys():
data_dic[key].append(value)
else:
data_dic[key] = [
value,
]
except Exception as error:
common.print_error(
'*Error* (get_sql_table_data) : Failed on getting table info from table "'
+ str(table_name) + '" of db_file "' + str(db_file) + '": ' +
str(error))
return (data_dic)
def check_job(self, job):
command = 'bjobs -UF ' + str(job)
job_dic = lsf_common.get_lsf_bjobs_uf_info(command)
if job_dic[job]['status'] != 'RUN':
common.print_error('*Error*: Job "' + str(job) + '" is not running, cannot get process status.')
sys.exit(1)
else:
if not job_dic[job]['pids']:
common.print_error('*Error*: Not find PIDs information for job "' + str(job) + '".')
sys.exit(1)
return(job_dic, job_dic[job]['pids'])
def __init__(self, job_sampling, queue_sampling, host_sampling, load_sampling, user_sampling, interval):
self.job_sampling = job_sampling
self.queue_sampling = queue_sampling
self.host_sampling = host_sampling
self.load_sampling = load_sampling
self.user_sampling = user_sampling
self.interval = interval
self.db_path = str(config.db_path) + '/monitor'
job_db_path = str(self.db_path) + '/job'
if not os.path.exists(job_db_path):
try:
os.system('mkdir -p ' + str(job_db_path))
except Exception as error:
common.print_error('*Error*: Failed on creating sqlite job db directory "' + str(job_db_path) + '".')
common.print_error(' ' + str(error))
sys.exit(1)
def check_pid(self, pid):
pid_list = []
command = 'pstree -p ' + str(pid)
(return_code, stdout, stderr) = common.run_command(command)
for line in str(stdout, 'utf-8').split('\n'):
line = line.strip()
if re.findall('\((\d+)\)', line):
tmp_pid_list = re.findall('\((\d+)\)', line)
if tmp_pid_list:
pid_list.extend(tmp_pid_list)
if not pid_list:
common.print_error('*Error*: No valid pid was found.')
sys.exit(1)
return(pid_list)
def read_args():
"""
Read in arguments.
"""
parser = argparse.ArgumentParser()
parser.add_argument('-j', '--job',
default='',
help='Specify the LSF jobid you want to trace on remote host.')
parser.add_argument('-p', '--pid',
default='',
help='Specify the pid you want to trace on local host.')
args = parser.parse_args()
if (not args.job) and (not args.pid):
common.print_error('*Error*: "--job" or "--pid" must be specified.')
sys.exit(1)
return(args.job, args.pid)
def drop_sql_table(db_file, orig_conn, table_name, commit=True):
"""
Drop table if it exists.
"""
(result, conn, curs) = connect_preprocess(db_file, orig_conn, mode='write')
if (result == 'failed') or (result == 'locked'):
return
try:
command = "DROP TABLE IF EXISTS '" + str(table_name) + "'"
curs.execute(command)
curs.close()
if commit:
conn.commit()
if orig_conn == '':
conn.close()
except Exception as error:
common.print_error(
'*Error* (drop_sql_table) : Failed on drop table "' +
str(table_name) + '" from db_file "' + str(db_file) + '": ' +
str(error))