def get_data_by_part_id(self, result, part_id):
ParaLiteLog.debug("partition number : %s" % len(self.p_node))
if self.dest != conf.DATA_TO_ANO_OP or self.dest == conf.DATA_TO_ANO_OP and len(self.p_node) == 1:
# part id is the job id
rs = ""
for dataid in self.result:
for data in self.result[dataid]:
if isinstance(data, str):
# data is stored in file
f = open(data, "rb")
rs += f.read()
f.close()
else:
# data is stored in buffer
rs += data.getvalue()
return rs
rs = ""
for part in result[part_id]:
if isinstance(part, str):
# data is stored in file
f = open(part, "rb")
rs += f.read()
f.close()
else:
# data is stored in buffer
rs += part.getvalue()
return rs
def process_ck_info(self, message):
# message --> DATA_PERSIST:CHECKPOINT[:REPLICA_NODE:REPLICA_PORT]
is_ck = message[1]
self.is_checkpoint = is_ck
if is_ck == conf.CHECKPOINT:
replica_addr = (message[2], string.atoi(message[3]))
if self.result_type != conf.MULTI_FILE:
for dataid in self.result:
ds = ""
for data in self.result[dataid]:
ds += data.getvalue()
self.write_data_to_disk(dataid, ds)
f_name = self.get_file_name_by_data_id(gethostname(), dataid)
cmd = "scp %s %s:%s" % (f_name, replica_addr[0], f_name)
ParaLiteLog.debug("CDM: %s" % cmd)
os.system(cmd)
else:
for dataid in self.result:
f = open(self.result[dataid], "rb")
while True:
ds = f.read(self.max_size)
if not ds:
break
msg = conf.SEP_IN_MSG.join(
[conf.DATA_REPLICA, gethostname(), str(dataid), ds])
self.send_data_to_node(msg, AF_INET, replica_addr)
def sql_proc(self):
try:
ParaLiteLog.debug("sql proc : START")
# start local socket server to listen all connections
ch = self.iom.create_server_socket(AF_INET,
SOCK_STREAM, 100, ("", self.my_port))
n, self.my_port = ch.ss.getsockname()
ParaLiteLog.debug("listen on port : %s ..." % str(self.my_port))
# register local port to the master
self.register_to_master(self.cqid, self.opid, gethostname(), self.my_port)
ParaLiteLog.debug("reg to master: FINISH")
while self.is_running:
s_time = time.time()
ev = self.next_event(None)
if isinstance(ev, ioman_base.event_accept):
self.handle_accept(ev)
if isinstance(ev, ioman_base.event_read):
if ev.data != "":
e_time = time.time()
self.handle_read(ev)
for thd in self.threads:
thd.join()
for proc in self.processes:
proc.join()
ParaLiteLog.info("--sql node %s on %s is finished--" % (self.opid,
gethostname()))
#self.notifier.join()
except KeyboardInterrupt, e:
self.report_error("ParaLite receives a interrupt signal and then will close the process\n")
ParaLiteLog.info("--sql node %s on %s is finished--" % (self.opid,
gethostname()))
sys.exit(1)
def distribute_data(self):
whole_data = cStringIO.StringIO()
for i in self.result:
for csio in self.result[i]:
d = string.strip(csio.getvalue())
if len(d) == 0:
continue
whole_data.write(d)
whole_data.write("\n")
del csio
if self.distinct or self.limit != -1:
data_list = whole_data.getvalue().split(self.db_row_sep)
del whole_data
if self.distinct:
data_list = set(data_list)
if self.limit != -1:
data_list = data_list[:self.limit]
data = cStringIO.StringIO()
data.write(self.db_row_sep.join(str(s) for s in data_list))
del data_list
else:
data = whole_data
if self.dest == conf.DATA_TO_ONE_CLIENT:
# send data to a random client
random_num = random.randint(0, len(self.client_sock) - 1)
addr = self.client_sock[random_num]
sock = socket(AF_INET, SOCK_STREAM)
sock.connect(addr)
data_s = data.getvalue()
sock.send("%10s%s" % (len(data_s), data_s))
re = sock.recv(10)
assert re == "OK"
sock.close()
elif self.dest == conf.DATA_TO_DB:
self.data = data
col_sep = self.db_col_sep
row_sep = self.db_row_sep
master = (self.master_name, self.master_port)
ParaLiteLog.debug("Load data start:")
# send request to the master
t_size = len(data.getvalue())
sep = conf.SEP_IN_MSG
tag = conf.LOAD_FROM_API
if row_sep is None or row_sep == "\n":
temp_sep = "NULL"
else:
temp_sep = row_sep
msg = sep.join(
str(s) for s in [conf.REQ, self.cqid, gethostname(),
self.my_port, self.dest_db, self.dest_table,
t_size, tag, self.fashion, temp_sep, "0"])
so_master = socket(AF_INET, SOCK_STREAM)
so_master.connect(master)
so_master.send("%10s%s" % (len(msg),msg))
so_master.close()
def hash_data_file(self, data, key_pos, nodes, row_sep, col_sep, chunk_num, sub_dbs):
sep = conf.SEP_IN_MSG
db_buf = {}
for db in sub_dbs:
db_buf[db] = cStringIO.StringIO()
if col_sep is None: SEP = self.db_col_sep
else: SEP = col_sep
records = data.split(row_sep)
count = 1
for line in records:
if line == "":
continue
key = " ".join(line.strip().split(SEP)[kp] for kp in key_pos)
pnum = abs(hash(key)) % (len(nodes)*chunk_num)
if pnum == 0:
count += 1
db_name = sub_dbs[pnum]
db_buf[db_name].write("%s%s" % (line.strip(), row_sep))
ParaLiteLog.debug("count %s" % count)
for db in db_buf:
ParaLiteLog.debug("%s -- > %s" % (db, len(db_buf[db].getvalue())))
break
return db_buf
def send_to_node(self, db, table, data, addr, row_sep, col_sep, is_replace):
sep = conf.SEP_IN_MSG
req_info = "%s%s%s%s%s%s%s%s%s%s%s%s%s" % (conf.INFO, sep, db, sep, table, sep, self.db_col_sep, sep, col_sep, sep, row_sep, sep,is_replace)
ParaLiteLog.info("sending %s --> %s" % (req_info, addr[0]))
self.really_send(addr, req_info)
# use the first 10 charactors to indicate the database
self.really_send(addr, "%10s%s%s" % (len(db), db, data))
ParaLiteLog.info("sending data : %s --> %s" % (len(data), repr(addr)))
def recv_bytes(self, so, n):
A = []
while n > 0:
x = so.recv(n)
if x == "": break
A.append(x)
ParaLiteLog.debug(len(x))
n = n - len(x)
return string.join(A, "")
def step(self, value):
try:
newvalue = value
if isinstance(value, unicode):
newvalue = value.encode("ascii")
if isinstance(newvalue, str):
newvalue = string.atoi(newvalue)
self.product *= newvalue
except:
ParaLiteLog.info(traceback.format_exc())
raise(Exception(traceback.format_exc()))
def proc_drop(self, exp, target_db):
try:
for db in target_db:
conn = sqlite3.connect(db)
c = conn.cursor()
c.execute(exp)
conn.commit()
conn.close()
except sqlite3.OperationalError, e:
es("%s: %s" % (gethostname(), " ".join(e.args)))
ParaLiteLog.info(traceback.format_exc())
def register_to_master(self, cqid, opid, node, port):
sep = conf.SEP_IN_MSG
msg = sep.join([conf.REG, conf.DATA_NODE, cqid, opid, gethostname(), str(self.my_port), self.local_addr])
ParaLiteLog.debug("MASTER_NODE: %s MASTER_PORT: %s" % (self.master_name, self.master_port))
addr = (self.master_name, self.master_port)
sock = socket(AF_INET, SOCK_STREAM)
try:
sock.connect(addr)
except Exception, e:
ParaLiteLog.error("Error in register_to_master: %s" % traceback.format_exc())
if e.errno == 4:
sock.connect(addr)
def start(self):
try:
# start socket server to listen all connections
ch = self.iom.create_server_socket(AF_INET, SOCK_STREAM, 100, ("", self.my_port))
n, self.my_port = ch.ss.getsockname()
ParaLiteLog.debug("listen on port : %s ..." % str(self.my_port))
# start socket server for local connections
self.local_addr = "/tmp/paralite-local-addr-orderby-%s-%s-%s" % (gethostname(), self.cqid, self.opid)
if os.path.exists(self.local_addr):
os.remove(self.local_addr)
self.iom.create_server_socket(AF_UNIX, SOCK_STREAM, 10, self.local_addr)
# register local port to the master
self.register_to_master(self.cqid, self.opid, gethostname(), self.my_port)
ParaLiteLog.debug("reg to master: FINISH")
while self.is_running:
ev = self.next_event(None)
if isinstance(ev, ioman_base.event_accept):
self.handle_accept(ev)
if isinstance(ev, ioman_base.event_read):
if ev.data != "":
self.handle_read(ev)
ParaLiteLog.info("--orderby node %s on %s is finished--" % (self.opid, gethostname()))
except KeyboardInterrupt, e:
self.report_error("ParaLite receives a interrupt signal and then will close the process\n")
ParaLiteLog.info("--orderby node %s on %s is finished--" % (self.opid, gethostname()))
sys.exit(1)
def proc_create(self, exp, target_db):
try:
# first of all, check if the directory holds database exists or not
for db in target_db:
parent = db[0:db.rfind(os.sep)]
if not os.path.exists(parent):
os.makedirs(parent)
conn = sqlite3.connect(db)
c = conn.cursor()
c.execute(exp)
conn.commit()
conn.close()
except sqlite3.OperationalError, e:
ParaLiteLog.info(traceback.format_exc())
raise(Exception("ERROR: in proc_create: %s: %s" % (gethostname(),
" ".join(e.args))))
def main():
if len(sys.argv) != 7:
sys.exit(1)
proc = OrderbyOp()
proc.master_name = sys.argv[1]
proc.master_port = string.atoi(sys.argv[2])
proc.cqid = sys.argv[3]
proc.opid = sys.argv[4]
proc.my_port = string.atoi(sys.argv[5])
proc.log_dir = sys.argv[6]
if not os.path.exists(proc.log_dir):
os.makedirs(proc.log_dir)
cur_time = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime(time.time()))
ParaLiteLog.init("%s/orderby-%s-%s.log" % (proc.log_dir, gethostname(), cur_time), logging.DEBUG)
ParaLiteLog.info("--orderby node %s on %s is started" % (proc.opid, gethostname()))
proc.start()
def main():
if len(sys.argv) != 7:
sys.exit(1)
proc = SqlOp()
proc.master_name = sys.argv[1]
proc.master_port = string.atoi(sys.argv[2])
proc.cqid = sys.argv[3]
proc.opid = sys.argv[4]
proc.my_port = string.atoi(sys.argv[5])
proc.log_dir = sys.argv[6]
if not os.path.exists(proc.log_dir): os.makedirs(proc.log_dir)
ParaLiteLog.init("%s/sql-%s-%s-%s.log" % (
proc.log_dir, gethostname(), proc.cqid, proc.opid),
logging.DEBUG)
ParaLiteLog.info("--sql node %s on %s is started" % (proc.opid, gethostname()))
proc.sql_proc()
def distinct_data(self, data_list):
try:
csio = cStringIO.StringIO()
whole_data = ""
for data in data_list:
if data.strip() == "":
continue
whole_data += data.strip() + self.db_row_sep
del(data)
if whole_data == "":
return None, None, None
whole_data = set(whole_data.strip().split(self.db_row_sep))
csio.write(self.db_row_sep.join(whole_data))
return conf.SINGLE_BUFFER, [csio], len(csio.getvalue())
except Exception, e:
ParaLiteLog.debug(traceback.format_exc())
self.report_error("ERROR in order_by.py : %s" % traceback.format_exc())
return None, None, None
def get_data_by_blocksize(self, jobid, bksize):
if self.reader is None:
return None
data = self.reader.read(bksize)
if not data:
read_size = 0
else:
read_size = len(data)
if read_size < bksize or read_size == bksize and read_size == self.job_data[jobid]:
# while True:
# if self.reader is not None:
# self.reader.close()
# self.reader = None
# self.reader = self.get_next_reader()
# if self.reader is None:
# break
# new_data = self.reader.read(bksize - read_size)
# data += new_data
# read_size = len(data)
# if read_size >= bksize:
# break
if self.reader is not None:
self.reader.close()
self.reader = None
self.reader = self.get_next_reader()
return data
if self.db_row_sep == "\n":
if not data.endswith("\n"):
extra_data = self.reader.readline()
if extra_data:
data += extra_data
return data
else:
if data:
pos = data.rfind(self.db_row_sep)
ParaLiteLog.info(pos)
send_ds = self.left_ds + data[0:pos]
self.left_ds = data[pos+len(self.db_row_sep):]
return send_ds
else:
return None
def handle_read(self, ev):
data = ev.data
if data == conf.END_TAG:
ParaLiteLog.info("receive: END_TAG")
self.is_running = False
self.queue.put(conf.END_TAG)
elif data.startswith(conf.INFO):
m = data.split(conf.SEP_IN_MSG)
assert len(m) == 7
if self.table == None: self.table = m[2]
if self.db_col_sep == None: self.db_col_sep = m[3]
if self.cmd_col_sep == None: self.cmd_col_sep = m[4]
if self.cmd_row_sep == None: self.cmd_row_sep = m[5]
if self.is_replace == None:
self.is_replace = m[6]
ParaLiteLog.info("DB_COL_SEP = %s CMD_COL_SEP = %s CMD_ROW_SEP = %s is_replace = %s" % (self.db_col_sep, self.cmd_col_sep, self.cmd_row_sep, self.is_replace))
else:
"""
TODO: we can control the buffer size here.
"""
self.queue.put(data)
"""
def scan_data_queue(self):
while True:
data = self.queue.get()
if data == conf.END_TAG:
ParaLiteLog.info("SCAN DATA QUEUE : END")
break
try:
pos = 10+string.atoi(data[0:10].strip())
target_db = data[10:pos]
data = data[pos:]
"""
thd = threading.Thread(target=self.write_to_db, args=(data, len(data)))
thd.setDaemon(True)
thd.start()
self.threads.append(thd)
"""
self.write_to_db(target_db, data, len(data))
del(data)
except Exception, e:
ParaLiteLog.info(traceback.format_exc())
es("in write_to_db: %s" % (traceback.format_exc()))
sys.exit(1)
def start(self, argument):
self.parse(argument)
cur_time = time.strftime('%Y-%m-%d-%H-%M-%S',time.localtime(time.time()))
ParaLiteLog.init("%s/dload-server-%s-%s.log" % (self.log_dir,
gethostname(), cur_time),
logging.DEBUG)
ParaLiteLog.info("START")
ParaLiteLog.info("parse the argumens sucessfully")
ss = time.time()
scan_thd = threading.Thread(target=self.scan_data_queue)
scan_thd.setDaemon(True)
scan_thd.start()
t = time.strftime('%Y-%m-%d-%H-%M-%S',time.localtime(time.time()))
self.local_socket = "%s%s%s-%s-%s" % (self.log_dir, os.sep, gethostname(), t, "UNIX.d")
self.iom.create_server_socket(AF_UNIX, SOCK_STREAM, 5, self.local_socket)
ch = self.iom.create_server_socket(AF_INET, SOCK_STREAM, 5, ("", self.port))
n, self.port = ch.ss.getsockname()
ParaLiteLog.info("global socket addr = %s" % (repr(ch.ss.getsockname())))
self.register_to_master()
try:
while self.is_running:
ev = self.next_event(None)
if isinstance(ev, ioman_base.event_accept):
self.handle_accept(ev)
elif isinstance(ev, ioman_base.event_read):
if ev.data != "":
self.handle_read(ev)
except Exception, e:
es("in dload_server.py : %s" % traceback.format_exc())
ParaLiteLog.info(traceback.format_exc())
sys.exit(1)
def handle_read(self, event):
message = event.data[10:]
m = message.split(conf.SEP_IN_MSG)
try:
if m[0] == conf.JOB_ARGUMENT:
self.parse_args(m[1])
ParaLiteLog.info("parse arguments: FINISH")
elif m[0] == conf.JOB:
ParaLiteLog.debug("MESSAGE: %s" % message)
self.cur_jobid = m[1]
elif m[0] == conf.DATA:
data_id = string.strip(m[1][0:2])
data = m[1][2:]
self.source_data.append(data)
# sort data
if not self.is_data_ready(self.source_data, self.num_of_children):
return
ParaLiteLog.debug("****SORT DATA****: start")
s = 0
for data in self.source_data:
s += len(data)
ParaLiteLog.debug("source data size: %s" % s)
s_time = time.time()
rs_type, rs, t_size = self.sort(self.source_data)
del self.source_data
ParaLiteLog.debug("****SORT DATA****: finish")
if rs_type is None:
self.send_status_to_master(self.cur_jobid, conf.PENDING)
return
self.total_size += t_size
self.source_data = {}
# store the result of one job to the final result
for i in range(len(rs)):
if i not in self.result:
self.result[i] = [rs[i]]
else:
self.result[i].append(rs[i])
if rs_type != conf.MULTI_FILE:
# check if the whole data exceeds the LIMITATION
if self.total_size > self.MAX_SIZE:
self.write_data_to_disk()
self.result_type = conf.MULTI_FILE
e_time = time.time()
self.total_time += e_time - s_time
self.send_status_to_master(self.cur_jobid, conf.PENDING)
elif m[0] == conf.JOB_END:
ParaLiteLog.debug("MESSAGE: %s" % message)
# all jobs are finished
self.send_rs_info_to_master(self.total_size, self.total_time)
# distribute data
if self.dest == conf.DATA_TO_ONE_CLIENT:
ParaLiteLog.debug("dest = %s" % self.dest)
self.distribute_data()
self.send_status_to_master(self.cur_jobid, conf.ACK)
self.is_running = False
elif self.dest == conf.DATA_TO_DB:
self.distribute_data()
elif m[0] == conf.DATA_PERSIST:
# if the data is requried to be persisted or not
if m[1] == conf.CHECKPOINT:
self.write_data_to_disk()
elif m[0] == conf.DLOAD_REPLY:
sep = conf.SEP_IN_MSG
reply = sep.join(m[1:])
ParaLiteLog.info("receive the information from the master")
ParaLiteLog.debug(reply)
if len(self.data.getvalue()) != 0:
dload_client.dload_client().load_internal_buffer(
reply,
self.dest_table,
self.data,
self.fashion,
self.hash_key,
self.hash_key_pos,
self.db_col_sep,
self.db_row_sep,
self.db_col_sep,
False,
"0",
self.log_dir,
)
# send END_TAG to the master
client_id = "0"
msg = sep.join([conf.REQ, conf.END_TAG, gethostname(), client_id])
so_master = socket(AF_INET, SOCK_STREAM)
so_master.connect((self.master_name, self.master_port))
so_master.send("%10s%s" % (len(msg), msg))
so_master.close()
ParaLiteLog.debug("sending to master: %s" % (conf.END_TAG))
ParaLiteLog.debug("----- dload client finish -------")
elif message == conf.DLOAD_END_TAG:
ParaLiteLog.debug("---------import finish---------")
self.send_status_to_master(" ".join(self.cur_jobid), conf.ACK)
self.is_running = False
elif m[0] == conf.EXIT:
self.is_running = False
elif m[0] == conf.NODE_FAIL:
ParaLiteLog.debug("MESSAGE: %s" % message)
# message --> NODE_FAIL:FAILED_NODE:REPLICA_NODE
failed_node, replica_node = m[1:3]
self.failed_node.append(failed_node)
if replica_node != "" and replica_node == gethostname():
# load replica data for the failed node
self.recovery_data(self.replica_result, replica_node)
ParaLiteLog.debug("Finish to handle node failure message")
except Exception, e:
es(traceback.format_exc())
ParaLiteLog.info(traceback.format_exc())
self.is_running = False
self.no_error = False
def process_job(self, jobid, exp, target_db, process_queue):
s_time = time.time()
rs_type, rs, t_size = self.proc_select(jobid, exp, target_db)
e_time = time.time()
process_queue.put((jobid, rs_type, rs, t_size, e_time-s_time))
ParaLiteLog.debug("Job %s cost %s " % (jobid, e_time-s_time))
def sort(self, data_list):
try:
ParaLiteLog.debug("sort data START")
csio = cStringIO.StringIO()
whole_data = ""
for data in data_list:
if data.strip() == "":
continue
whole_data += data.strip() + self.db_row_sep
del (data)
if whole_data == "":
return None, None, None
whole_data = whole_data.strip().split(self.db_row_sep)
ParaLiteLog.debug("order key %s" % (str(self.order_key)))
ParaLiteLog.debug("order type %s" % (str(self.order_type)))
# get the column positions to be sorted in the input
key_pos = [] # (pos:string_or_int:reverse_or_not)
for key in self.order_key:
pos = self.input.index(key)
if self.attrs.has_key(key) == False:
key_type = conf.STRING
else:
key_type = self.attrs[key]
if self.order_type[self.order_key.index(key)] == conf.DESC:
key_pos.append((pos, key_type, True))
else:
key_pos.append((pos, key_type, False))
# check if the ordering type is the same or not
flag = 0
t0 = self.order_type[0]
for t in self.order_type:
if t != t0:
flag = 1
break
if flag == 1:
# for different ordering type, we have to perform sort several times
i = len(key_pos) - 1
while i >= 0:
pos = key_pos[i]
if pos[1] == conf.INT:
whole_data.sort(key=lambda l: string.atoi(l.split(self.db_col_sep)[pos[0]]), reverse=pos[2])
elif pos[1] == conf.FLOAT or pos[1] == conf.REAL:
whole_data.sort(key=lambda l: float(l.split(self.db_col_sep)[pos[0]]), reverse=pos[2])
else:
whole_data.sort(key=lambda l: l.split(self.db_col_sep)[pos[0]], reverse=pos[2])
i -= 1
else:
sort_key = []
col_sep = self.db_col_sep
for pos in key_pos:
if pos[1] == conf.INT:
sort_key.append("string.atoi(l.split('%s')[%s])" % (self.db_col_sep, pos[0]))
elif pos[1] == conf.FLOAT or pos[1] == conf.REAL:
sort_key.append("float(l.split('%s')[%s])" % (self.db_col_sep, pos[0]))
else:
sort_key.append("l.split('%s')[%s]" % (self.db_col_sep, pos[0]))
whole_data.sort(key=lambda l: eval(",".join(sort_key)), reverse=key_pos[0][2])
csio.write(self.db_row_sep.join(whole_data))
# if len(self.db_col_sep) == 1 and self.db_row_sep == "\n":
# ParaLiteLog.debug("sort data: shell")
# self.shell_sort(data_list, csio)
# else:
# ParaLiteLog.debug("sort data: quick sort")
# self.quick_sort(data_list, csio)
return conf.SINGLE_BUFFER, [csio], len(csio.getvalue())
except Exception, e:
ParaLiteLog.debug(traceback.format_exc())
self.report_error("ERROR in order_by.py : %s" % traceback.format_exc())
return None, None, None
while self.is_running:
ev = self.next_event(None)
if isinstance(ev, ioman_base.event_accept):
self.handle_accept(ev)
if isinstance(ev, ioman_base.event_read):
if ev.data != "":
self.handle_read(ev)
ParaLiteLog.info("--orderby node %s on %s is finished--" % (self.opid, gethostname()))
except KeyboardInterrupt, e:
self.report_error("ParaLite receives a interrupt signal and then will close the process\n")
ParaLiteLog.info("--orderby node %s on %s is finished--" % (self.opid, gethostname()))
sys.exit(1)
except Exception, e1:
ParaLiteLog.debug(traceback.format_exc())
self.report_error(traceback.format_exc())
sys.exit(1)
def handle_read(self, event):
message = event.data[10:]
m = message.split(conf.SEP_IN_MSG)
try:
if m[0] == conf.JOB_ARGUMENT:
self.parse_args(m[1])
ParaLiteLog.info("parse arguments: FINISH")
elif m[0] == conf.JOB:
ParaLiteLog.debug("MESSAGE: %s" % message)
self.cur_jobid = m[1]
def parse_func_2(self):
"""
parse each expression to get useful information: input is different
input --> ***[count(*), sum(a), sum(a+b*c)]***
output --> ***[count(*), sum(a) + 1, avg(a+b*c)]***
expression --> ['count(*)', 'sum(a)', 'avg(a+b*c)']
==> [
['count', 1, 0, '*', 1, 0],
['sum', 2, 0, 'a', 2, 1, [1], lambda _a:_a+1 ],
['sum', 3, 0, 'a+b*c', 3, 0]
]
"""
# if there is avg function, convert it to sum and count
avg_pos = []
for fun in self.function:
if fun.find("avg(") != -1:
avg_pos.append(self.function.index(fun))
if avg_pos != []:
if "count(*)" not in self.function:
self.function.append("count(*)")
# sometimes, a result column has more than one function, then the pos in
# the self.expr and self.expression is not the same, this counter is to
# count the pos of a function in self.expr and pos_map shows the mapping
# of these two kinds of postions.
fun_counter = 0
pos_map = {}
for expr_num in range(len(self.function)):
expr = self.function[expr_num]
pos_map[expr_num] = fun_counter
# parse "sum(a) + 1" --> ['sum(a)', '+', '1']
if expr.find("count(*)") != -1:
expr = expr.replace("count(*)", "count(a_a)")
_expr = newparser.parse_column_expr(expr)
# to describe the pos of elements in the outer arithmetic operation
outer_ao_pos = []
new_expr = expr
new_args = []
ParaLiteLog.debug(self.input)
for ele in _expr:
if re.match("(.*)\((.*)\)", ele):
parsed_expr = []
new_expr = new_expr.replace(ele, "_col%s" % str(fun_counter))
new_args.append("_col%s" % str(fun_counter))
func_name = ele[0:ele.find("(")]
if func_name not in conf.GENERAL_FUNC:
return False, "ParaLite cannot support aggregation function %s" % func_name
if func_name == "avg":
ele = ele.replace("avg", "sum")
fun_counter += 1
func_attr = ele[ele.find("(") + 1 : ele.rfind(")")]
parsed_expr.append(func_name)
if func_attr == "a_a":
func_attr = "*"
ele = ele.replace("a_a", "*")
expr = expr.replace("count(a_a)", "count(*)")
opexpr = [func_attr]
pos_in_input = self.input.index(ele)
parsed_expr.append(pos_in_input)
parsed_expr.append(0)
parsed_expr.append(ele)
if expr in self.output: parsed_expr.append(self.output.index(expr))
else: parsed_expr.append(-1)
self.expr.append(parsed_expr)
outer_ao_pos.append(fun_counter - 1)
else:
# other operator element: + - * / ^
continue
cur_pos = pos_map[expr_num]
if cur_pos >= len(self.expr):
# the exception that select sum(a), count(*), avg(a) ... does not need
# to do anything for avg(a)
continue
if len(_expr) == 1:
self.expr[cur_pos].append(0)
else:
self.expr[cur_pos].append(1)
self.expr[cur_pos].append(outer_ao_pos)
tempexpr = new_expr
tempargs = ",".join(new_args)
for eacharg in new_args:
newarg = eacharg.replace(".", "_")
tempexpr = tempexpr.replace(eacharg, newarg)
tempargs = tempargs.replace(eacharg, newarg)
self.expr[cur_pos].append(
eval("lambda %s:%s" % (tempargs, tempexpr)))
self.pos_map = pos_map
return True, None
def load_internal_buffer(self, reply, table, buf, fashion, key, key_pos,
db_col_sep, row_sep, col_sep, is_replace, client_id,
LOG_DIR):
ParaLiteLog.info("load_internal: START")
ParaLiteLog.info("row separator = %s col separator = %s" % (row_sep, col_sep) )
self.db_col_sep = db_col_sep
total_size = len(buf.getvalue())
try:
"""
received message = nodes # sub_dbs # chunk_num # replica_info
nodes should be:
n1:p1:l1 , n2:p2:l2 , ... IF fashion = HASH_FASHION
n1:p1:l1:s1:num , n2:p2:l2:s2:num , ... IF fashion = ROUND_ROBIN
TBD IF fashion = RANGE_FASHION
node_db_info: db_1_1 , db_1_2 , db_2_1, ...
replica_info: db_1_1 db_1_1_r_1 node1 , db_1_2 db_1_2_r_1 node2 , ...
"""
mm = reply.split("#")
ParaLiteLog.info("receive the information from the master %s" % mm)
nodes = mm[0].split(",")
sub_dbs = mm[1].split(",")
chunk_num = string.atoi(mm[2])
replica = mm[3]
replica_info = {} # {db_name : {replica_db_name:node}}
if replica != "":
for ll in replica.split(","):
lll = ll.split(" ")
if lll[0] not in replica_info:
replica_info[lll[0]] = {}
replica_info[lll[0]][lll[1]] = lll[2]
ParaLiteLog.info(nodes)
node_addr = {} # {node:addr}
for node in nodes:
m = node.split(conf.SEP_IN_MSG)
if m[0] == gethostname(): addr = m[2]
else: addr = (m[0], string.atoi(m[1]))
node_addr[m[0]] = addr
ss1 = time.time()
if nodes == []:
ParaLiteLog.info("there is no data to load")
elif fashion == conf.HASH_FASHION:
ParaLiteLog.info(fashion)
# get the data for each sub db
# db_buf = {db_name, buffer_of_data}
db_buf = self.hash_data_buffer(buf, key_pos, nodes, row_sep, col_sep, chunk_num, sub_dbs)
for db in db_buf:
data = db_buf[db].getvalue()
node = db.split("_")[-3]
self.send_to_node(db, table, data, node_addr[node], row_sep, col_sep, is_replace)
if db in replica_info:
for rdb in replica_info[db]:
node = replica_info[db][rdb]
self.send_to_node(rdb, table, data, node_addr[node], row_sep, col_sep, is_replace)
"""
buf_scanner = threading.Thread(target=self.scan_buf,
args=(table, node_buf, node_addr, row_sep, col_sep, is_replace))
buf_scanner.setDaemon(True)
buf_scanner.start()
buf_scanner.join()
"""
elif fashion == conf.REPLICATE_FASHION:
self.replicate_data(table, files, total_size, nodes)
elif fashion == conf.RANGE_FASHION:
self.range_data()
else:
thds = []
num_of_db = len(nodes) * chunk_num
if row_sep is not None and row_sep != "\n":
whole_data = buf.getvalue()
lines = whole_data.split(row_sep)
if lines[len(lines)-1] == "":
lines.pop(len(lines)-1)
l = len(lines)
if l % num_of_db == 0:
num_each = l / num_of_db
else:
num_each = l / num_of_db + 1
i = 0
while i < num_of_db:
db = sub_dbs[i]
node = db.split("_")[-3]
cur_num = i*num_each + num_each
if cur_num > l:
cur_num = l
ds = row_sep.join(lines[i*num_each:cur_num])
thd = threading.Thread(target=self.send_to_node,
args=(db, table, ds, node_addr[node], row_sep,
col_sep, is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
if db in replica_info:
for rdb in replica_info[db]:
node = replica_info[db][rdb]
thd = threading.Thread(target=self.send_to_node,
args=(rdb, table, ds, node_addr[node],
row_sep, col_sep, is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
i += 1
else:
buf.seek(0)
i = 0
while i < num_of_db:
db = sub_dbs[i]
node = db.split("_")[-3]
node_id = i / chunk_num
size = string.atoi(nodes[node_id].split(conf.SEP_IN_MSG)[3]) / chunk_num
ParaLiteLog.info("start to get data as bk: %s" % (size))
ds = buf.read(size)
if ds is None:
ParaLiteLog.info("really get data as bk: 0")
continue
if not ds.endswith("\n"):
ds += buf.readline()
ParaLiteLog.info("really get data as bk: %s" % (len(ds)))
thd = threading.Thread(target=self.send_to_node,
args=(db, table, ds, node_addr[node],
row_sep, col_sep, is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
if db in replica_info:
for rdb in replica_info[db]:
node = replica_info[db][rdb]
thd = threading.Thread(target=self.send_to_node,
args=(rdb, table, ds, node_addr[node],
row_sep, col_sep, is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
i += 1
for thd in thds:
thd.join()
except Exception, e:
raise(e)
def handle_read(self, event):
message = event.data[10:]
sep = conf.SEP_IN_MSG
m = message.split(sep)
try:
if m[0] == conf.DATA_END:
ParaLiteLog.debug("MESSAGE: %s" % message)
# all data is dipatched to the parent nodes
self.send_status_to_master(" ".join(self.job_data), conf.ACK)
ParaLiteLog.debug("notify ACK to master")
self.is_running = False
elif message == conf.END_TAG:
ParaLiteLog.debug("MESSAGE: %s" % message)
self.send_status_to_master(" ".join(self.job_data), conf.ACK)
self.is_running = False
elif message == conf.DLOAD_END_TAG:
ParaLiteLog.debug("---------import finish---------")
self.send_status_to_master(" ".join(self.job_data), conf.ACK)
self.is_running = False
elif message == conf.EXIT:
ParaLiteLog.debug("MESSAGE: %s" % message)
self.is_running = False
elif m[0] == conf.JOB_ARGUMENT:
self.parse_args(m[1])
ParaLiteLog.info("parse arguments: FINISH")
# init the persisted result data
if self.is_checkpoint is not None and self.is_checkpoint == conf.CHECKPOINT:
ParaLiteLog.debug("recovery data: START")
# this is a recovery operator
self.recovery_data(self.result, gethostname())
ParaLiteLog.debug("recovery data: FINISH")
self.send_rs_info_to_master(0, 0)
else:
# delete all temporary files for this operator
os.system("rm -f %s/%s_%s" % (self.temp_dir, "sql", self.opid))
###############################
# scanner = threading.Thread(target=self.scan_process_queue, args=(self.process_queue, ))
# scanner.setDaemon(True)
# scanner.start()
# self.threads.append(scanner)
##########################
elif m[0] == conf.JOB:
self.ex_s_time = time.time()
self.ex_w_time = 0
ParaLiteLog.debug("MESSAGE: %s" % message)
s_time = time.time()
jobid = m[1]
target_db = m[2].split()
exp = self.expression
ParaLiteLog.debug("*****JOB %s******:start" % jobid)
# FAULT TOLERANCE:
if jobid in self.job_data:
# this is a failed job, we should first delete the old result value
if self.dest == conf.DATA_TO_ANO_OP and self.partition_num > 1:
for partid in self.result:
pos = self.job_list.index(jobid)
self.result[partid][pos] = ""
else:
self.result[jobid] = ""
if exp.lower().startswith("select"):
"""
selection task: (1), execute sql (2), notify the result to
the master (3), wait for the DATA_PERSIST message from the
master (4), persist data if so (5), notify ACK to the master
"""
ParaLiteLog.info("proc_select: START")
st_time = time.time()
rs_type, rs, t_size = self.proc_select(jobid, exp, target_db)
et_time = time.time()
ParaLiteLog.debug("Job %s cost time %s second" % (jobid, (et_time - st_time)))
# FAULT TOLERANCE:
if jobid in self.job_data:
# this is a failed job
if self.dest == conf.DATA_TO_ANO_OP and self.partition_num > 1:
for partid in self.result:
pos = self.job_list.index(jobid)
self.result[partid][pos] = rs[partid]
else:
self.result[jobid] = rs
self.send_status_to_master(jobid, conf.PENDING)
return
self.job_data[jobid] = t_size
self.job_list.append(jobid)
self.total_size += t_size
# store the result of one job to the final result
if len(rs) == 1:
if self.dest == conf.DATA_TO_ANO_OP:
# dest is AGGR op or ORDER op, use 0 as the key
if 0 not in self.result:
self.result[0] = rs
else:
self.result[0].append(rs[0])
if self.is_checkpoint == 1:
self.write_data_to_disk(0, rs[0].getvalue())
else:
# dest is UDX op, use jobid as the key
self.result[string.atoi(jobid)] = rs
if self.is_checkpoint == 1:
self.write_data_to_disk(0, rs[0].getvalue())
else:
# use partid as the key
for i in range(len(rs)):
if i not in self.result:
self.result[i] = [rs[i]]
else:
self.result[i].append(rs[i])
if self.is_checkpoint == 1:
for i in range(len(rs)):
self.write_data_to_disk(i, rs[i].getvalue())
# check if the whole data exceeds the LIMITATION
if rs_type != self.MULTI_FILE:
if self.is_checkpoint is not None and self.is_checkpoint == conf.CHECKPOINT or self.total_size > self.MAX_SIZE:
for dataid in self.result:
data = ""
for d in self.result[dataid]:
data += d.getvalue()
self.write_data_to_disk(dataid, data)
self.result_type = self.MULTI_FILE
e_time = time.time()
if self.total_time == 0:
self.total_time = (e_time - s_time)
self.send_status_to_master(jobid, conf.PENDING)
elif exp.lower().startswith("create"):
ParaLiteLog.info("proc_create: START")
ParaLiteLog.info("SQL: %s" % exp)
self.proc_create(exp, target_db)
ParaLiteLog.info("proc_create: START")
self.send_status_to_master(jobid, conf.ACK)
self.is_running = False
elif exp.lower().startswith("drop"):
ParaLiteLog.info("proc_drop: START")
self.proc_drop(exp, target_db)
self.send_status_to_master(jobid, conf.ACK)
self.is_running = False
ParaLiteLog.debug("*****JOB %s******:finish" % jobid)
self.ex_w_time += (time.time() - self.ex_s_time)
self.ex_s_time = 0
elif m[0] == conf.JOB_END:
ParaLiteLog.debug("MESSAGE: %s" % message)
# all jobs are finished
# create a dictionary to store the status of each part of data
data_status = {} # {data_id:[(pos_in_result, status)]}
for dataid in self.result:
if dataid not in data_status:
data_status[dataid] = []
for i in range(len(self.result[dataid])):
data_status[dataid].append((i, 1))
self.data_status = data_status
self.reader = self.get_next_reader()
self.send_rs_info_to_master(self.total_size, self.total_time)
# distribute data
if self.dest == conf.DATA_TO_ONE_CLIENT:
self.distribute_data()
self.send_status_to_master(" ".join(self.job_data), conf.ACK)
self.is_running = False
elif self.dest == conf.DATA_TO_DB:
self.distribute_data()
elif m[0] == conf.DLOAD_REPLY:
reply = sep.join(m[1:])
ParaLiteLog.info("receive the information from the master")
ParaLiteLog.debug(reply)
if len(self.data.getvalue()) != 0:
dload_client.dload_client().load_internal_buffer(
reply, self.dest_table, self.data, self.fashion,
self.hash_key, self.hash_key_pos, self.db_col_sep,
self.db_row_sep, self.db_col_sep, False, "0", self.log_dir)
# send END_TAG to the master
client_id = "0"
msg = sep.join([conf.REQ, conf.END_TAG, gethostname(), client_id])
so_master = socket(AF_INET, SOCK_STREAM)
so_master.connect((self.master_name, self.master_port))
so_master.send("%10s%s" % (len(msg), msg))
so_master.close()
ParaLiteLog.debug("sending to master: %s" % (conf.END_TAG))
ParaLiteLog.debug("----- dload client finish -------")
elif m[0] == conf.DATA_PERSIST:
ParaLiteLog.debug("MESSAGE: %s" % message)
# if the data is requried to be persisted or not
self.process_ck_info(m)
elif m[0] == conf.DATA_DISTRIBUTE:
ParaLiteLog.debug("MESSAGE: %s" % message)
# send a part of data to the next operator
# DATA_DISTRIBUTE:partition_num:destnode
part_id, destnode = m[1:]
data = self.get_data_by_part_id(self.result, string.atoi(part_id))
# DATA message includes: type:id+data
# the first 2 chars represents the opid
msg = sep.join([conf.DATA, "%2s%s" % (self.opid, data)])
if destnode == gethostname():
# use local socket
addr = self.p_node[destnode][1]
t = AF_UNIX
else:
addr = (destnode, self.p_node[destnode][0])
t = AF_INET
self.send_data_to_node(msg, t, addr)
ParaLiteLog.debug("send data susscufully %s %s --> %s" % (self.opid, gethostname(), destnode))
elif m[0] == conf.DATA_DISTRIBUTE_UDX:
ParaLiteLog.debug("MESSAGE: %s" % message)
# send data to udx client
# m[1:] = worker.id:jobid:(node:port | addr):size
if len(m) == 6:
w_id, jobid = m[1:3]
addr = (m[3], string.atoi(m[4]))
t = AF_INET
bk = string.atoi(m[5])
elif len(m) == 5:
w_id, jobid = m[1:3]
addr = m[3]
t = AF_UNIX
bk = string.atoi(m[4])
data = self.get_data_by_blocksize(jobid, bk)
if not data:
# if we don't send something here, udx will not send KAL
# again, and then they will not receive data again, the whole
# process will be blocked for ever
msg = sep.join([conf.DATA, "EMPTY"])
else:
msg = sep.join([conf.DATA, data])
self.send_data_to_node(msg, t, addr)
elif m[0] == conf.DATA_REPLICA:
ParaLiteLog.debug("MESSAGE: %s" % message)
# message --> DATA_REPLICA:DATANODE:DATAID:DATA
datanode, dataid = m[1:3]
f_name = self.get_file_name_by_data_id(gethostname(), dataid)
fr = open(f_name, "wa")
fr.write(m[4])
fr.close()
elif m[0] == conf.NODE_FAIL:
ParaLiteLog.debug("MESSAGE: %s" % message)
# message --> NODE_FAIL:FAILED_NODE:REPLICA_NODE
failed_node, replica_node = m[1:3]
self.failed_node.append(failed_node)
if replica_node == gethostname():
# load replica data for the failed node
self.recovery_data(self.replica_result, replica_node)
except Exception, e:
es("in sql_proc : %s" % traceback.format_exc())
ParaLiteLog.info(traceback.format_exc())
self.is_running = False
self.no_error = False
def range_data(self):
ParaLiteLog.info("Now RANGE FASHION is not supported...")
def proc_select(self, jobid, exp, target_db):
assert len(target_db) == 1
cur_db = target_db[0]
try:
conn = sqlite3.connect(cur_db)
conn.text_factory = str
# register the user-defined aggregate
conn.create_aggregate("mul", 1, mul)
c = conn.cursor()
"""
if self.temp_store != 0:
c.execute('pragma temp_store=%s' % (self.temp_store))
if self.cache_size != -1:
c.execute('pragma cache_size=%s' % (self.cache_size))
"""
# for test
c.execute('pragma temp_store=memory')
c.execute('pragma cache_size=2073741824')
ParaLiteLog.info("start to execute sql: %s" % exp)
col_sep = self.db_col_sep
row_sep = self.db_row_sep
num_of_dest = self.partition_num
if self.dest == conf.DATA_TO_ANO_OP and num_of_dest > 1:
columns = self.output
split_key = self.split_key
assert split_key is not None
# partition data in hash fashion
pos = []
for key in split_key:
pos.append(columns.index(key))
data_part_list = []
for i in range(self.partition_num):
data_part_list.append(cStringIO.StringIO())
size = 0
t_size = 0
for row in c.execute(exp):
part_id = abs(hash(self.db_col_sep.join(str(row[p]) for p in pos))) % num_of_dest
#part_id = abs(hash(row[pos[0]])) % num_of_dest
data = col_sep.join(str(s) for s in row)
"""
size += len(data)
if size > self.MAX_SIZE:
for partid in data_part_list:
fs = self.write_data_to_disk(
partid, data_part_list[partid])
# delete all data in csio
data_part_list[partid].truncate(0)
t_size += size
size = 0
self.result_type = self.MULTI_FILE
"""
data_part_list[part_id].write(data)
data_part_list[part_id].write(row_sep)
for i in range(len(data_part_list)):
t_size += len(data_part_list[i].getvalue())
ParaLiteLog.debug("finish to retrieve the result: %s" % t_size)
if self.result_type == self.MULTI_FILE:
for partid in data_part_list:
self.write_data_to_disk(
partid, data_part_list[partid].getvalue())
del data_part_list
return self.MULTI_FILE, None, t_size
else:
########################
# new_list = []
# for d in data_part_list:
# new_list.append(d.getvalue())
# return self.MULTI_BUFFER, new_list, t_size
###################
return self.MULTI_BUFFER, data_part_list, t_size
else:
csio = cStringIO.StringIO()
t_size = 0
size = 0 # record the size of current data
data_pos = [] # the file name of data if persisted
for row in c.execute(exp):
# NOTE: For aggregation SQL, e.g. "select max(col) from T ..."
# if there is no record in T, (None,) will be returned
if row[0] is None:
continue
data = col_sep.join(str(s) for s in row)
size += len(data)
if size >= self.MAX_SIZE:
result_type = self.MULTI_FILE
self.write_data_to_disk(jobid, csio.getvalue())
# delete all data in csio
csio.truncate(0)
t_size += size
size = 0
csio.write(data)
csio.write(row_sep)
t_size += len(csio.getvalue())
ParaLiteLog.debug("finish to retrieve the result: %s" % t_size)
if self.result_type == conf.MULTI_FILE:
self.write_data_to_disk(jobid, csio.getvalue())
del csio
return conf.MULTI_FILE, None, t_size
else:
return self.SINGLE_BUFFER, [csio], t_size
except sqlite3.OperationalError, e:
ParaLiteLog.info(traceback.format_exc())
raise(Exception("%s: QueryExecutionError: %s" % (gethostname(),
traceback.format_exc())))
def load_internal_file(self, reply, opt, db_col_sep, LOG_DIR):
ParaLiteLog.info("load_internal_file: START")
table = opt.table
files = opt.files
col_sep = opt.col_sep
row_sep = opt.row_sep
fashion = opt.fashion
key = opt.key
key_pos = opt.key_pos
is_replace = opt.replace
self.db_col_sep = db_col_sep
for f in files:
self.files[f] = 1
self.file_reader = open(self.get_next_file(), "rb")
try:
"""
received message = nodes # sub_dbs # chunk_num # replica_info
nodes should be (| is SEP_IN_MSG):
n1 : p1|l1 , n2 : p2|l2 , ... IF fashion = HASH_FASHION
n1 : p1|l1|s1|num , n2 : p2|l2|s2|num , ... IF fashion = ROUND_ROBIN
TBD IF fashion = RANGE_FASHION
node_db_info: node1:[db_1_1] , node2:[db_1_2] , node3:[db_2_1], ...
replica_info: db_1_1 db_1_1_r_1 node1 , db_1_2 db_1_2_r_1 node2 , ...
"""
mm = reply.split("#")
nodes = mm[0].split(",")
sub_dbs = mm[1].split(",")
chunk_num = string.atoi(mm[2])
replica = mm[3]
replica_info = {} # {db_name : {replica_db_name:node}}
if replica != "":
for whole_re in replica.split(","):
lll = whole_re.split(" ")
if lll[0] not in replica_info:
replica_info[lll[0]] = {}
replica_info[lll[0]][lll[1]] = lll[2]
node_addr = {} # {node:addr}
for node in nodes:
m = node.split(conf.SEP_IN_MSG)
if m[0] == gethostname(): addr = m[2]
else: addr = (m[0], string.atoi(m[1]))
node_addr[m[0]] = addr
thds = []
if nodes == []:
ParaLiteLog.info("there is no data to load")
elif fashion == conf.HASH_FASHION:
ParaLiteLog.info(fashion)
if row_sep is not None and row_sep != "\n":
while True:
dst = self.get_data_as_bk(DATA_MAX_SIZE)
if dst is None:
ParaLiteLog.info("really get data as bk: 0")
break
ParaLiteLog.info("really get data as bk: %s" % (len(dst)))
pos = dst.rfind(row_sep)
ds = left_ds + dst[0:pos]
left_ds = dst[pos+len(row_sep):]
del dst
db_buf = self.hash_data_file(ds, key_pos, nodes,
row_sep, col_sep,
chunk_num, sub_dbs)
ParaLiteLog.debug("hash data finish %s" % len(ds))
del ds
for db in db_buf:
data = db_buf[db].getvalue()
node = db.split("_")[-3]
thd = threading.Thread(target=self.send_to_node,
args=(db, table, data, node_addr[node],
row_sep,col_sep,
is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
if db in replica_info:
for rdb in replica_info[db]:
node = replica_info[db][rdb]
self.send_to_node(rdb, table, data,
node_addr[node],
row_sep, col_sep, is_replace)
else:
while True:
ds = self.get_data_as_bk(DATA_MAX_SIZE)
if ds is None:
ParaLiteLog.info("really get data as bk: 0")
break
ParaLiteLog.info("really get data as bk: %s" % (len(ds)))
db_buf = self.hash_data_file(ds, key_pos, nodes,
"\n", col_sep, chunk_num, sub_dbs)
for db in db_buf:
ParaLiteLog.debug(
"%s -- > %s" % (db, len(db_buf[db].getvalue())))
break
for db in db_buf:
data = db_buf[db].getvalue()
node = db.split("_")[-3]
thd = threading.Thread(target=self.send_to_node,
args=(db, table, data,
node_addr[node],
row_sep,col_sep,
is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
if db in replica_info:
for rdb in replica_info[db]:
node = replica_info[db][rdb]
self.send_to_node(rdb, table, data,
node_addr[node],
row_sep, col_sep, is_replace)
del db_buf
del ds
elif fashion == conf.REPLICATE_FASHION:
self.replicate_data(table, files, total_size, nodes)
elif fashion == conf.RANGE_FASHION:
self.range_data()
else:
num_of_db = len(nodes) * chunk_num
if row_sep is not None and row_sep != "\n":
i = 0
left_ds = ""
while True:
db = sub_dbs[i % num_of_db]
#m = nodes[(i % num_of_db) / chunk_num].split(conf.SEP_IN_MSG)
#node = m[0]
node = db.split("_")[-3]
size = string.atoi(m[3]) / chunk_num + 1
if size > DATA_MAX_SIZE:
ParaLiteLog.info("start to get data as bk: %s" % (DATA_MAX_SIZE))
ds = self.get_data_as_bk(DATA_MAX_SIZE)
else:
ParaLiteLog.info("start to get data as bk: %s" % (size))
ds = self.get_data_as_bk(size)
if ds is None:
ParaLiteLog.info("really get data as bk: 0")
break
ParaLiteLog.info("really get data as bk: %s" % (len(ds)))
pos = ds.rfind(row_sep)
send_ds = left_ds + ds[0:pos]
left_ds = ds[pos+len(row_sep):]
thd = threading.Thread(
target=self.send_to_node,
args=(db, table, send_ds, node_addr[node],
row_sep, col_sep, is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
if db in replica_info:
for rdb in replica_info[db]:
node = replica_info[db][rdb]
thd = threading.Thread(
target=self.send_to_node,
args=(rdb, table, ds, node_addr[node],
row_sep, col_sep, is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
i += 1
else:
i = 0
while True:
db = sub_dbs[i % num_of_db]
#m = nodes[(i % num_of_db)/chunk_num].split(conf.SEP_IN_MSG)
#node = m[0]
node = db.split("_")[-3]
size = string.atoi(m[3]) / chunk_num + 1
if size > DATA_MAX_SIZE:
ParaLiteLog.info(
"start to get data as bk: %s" % (DATA_MAX_SIZE))
ds = self.get_data_as_bk(DATA_MAX_SIZE)
else:
ParaLiteLog.info("start to get data as bk: %s" % (size))
ds = self.get_data_as_bk(size)
if ds is None:
ParaLiteLog.info("really get data as bk: 0")
break
ParaLiteLog.info("really get data as bk: %s" % (len(ds)))
thd = threading.Thread(
target=self.send_to_node,
args=(db, table, ds, node_addr[node],
row_sep,col_sep, is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
if db in replica_info:
for rdb in replica_info[db]:
ParaLiteLog.info(rdb)
node = replica_info[db][rdb]
thd = threading.Thread(
target=self.send_to_node,
args=(rdb, table, ds, node_addr[node],
row_sep, col_sep, is_replace))
thd.setDaemon(True)
thd.start()
thds.append(thd)
i += 1
del ds
for thd in thds:
thd.join()
except Exception, e:
ParaLiteLog.debug(traceback.format_exc())
raise(Exception(traceback.format_exc()))
def distribute_data(self):
# handle the limit condition: get the first N records
# E.g. select ... limit 10, the master firstly decides the limit
# number for each process and set the limit value for each process
# to be the post-limit
whole_data = cStringIO.StringIO()
for i in self.result:
for csio in self.result[i]:
d = string.strip(csio.getvalue())
if len(d) == 0:
continue
whole_data.write(d)
whole_data.write("\n")
del csio
if self.distinct or self.limit != -1:
data_list = whole_data.getvalue().split(self.db_row_sep)
del whole_data
if self.distinct:
data_list = set(data_list)
if self.limit != -1:
data_list = data_list[:self.limit]
data = cStringIO.StringIO()
data.write(self.db_row_sep.join(str(s) for s in data_list))
del data_list
else:
data = whole_data
if self.dest == conf.DATA_TO_DB:
self.data = data
col_sep = self.db_col_sep
row_sep = self.db_row_sep
master = (self.master_name, self.master_port)
ParaLiteLog.info("proc_select: load data start")
# send request to the master
t_size = len(data.getvalue())
sep = conf.SEP_IN_MSG
tag = conf.LOAD_FROM_API
if row_sep is None or row_sep == "\n":
temp_sep = "NULL"
else:
temp_sep = row_sep
msg = sep.join(
str(s) for s in [conf.REQ, self.cqid, gethostname(),
self.my_port, self.dest_db, self.dest_table,
t_size, tag, self.fashion, temp_sep, "0"])
so_master = socket(AF_INET, SOCK_STREAM)
so_master.connect(master)
so_master.send("%10s%s" % (len(msg),msg))
so_master.close()
# dload_client.dload_client().load_internal_buffer(
# master, self.cqid, gethostname(), self.my_port, self.dest_db,
# self.dest_table, data, conf.LOAD_FROM_API, self.fashion,
# self.hash_key, self.hash_key_pos, self.db_col_sep, row_sep,
# col_sep, False, "0", self.log_dir)
elif self.dest == conf.DATA_TO_ONE_CLIENT:
random_num = random.randint(0, len(self.client_sock) - 1)
addr = self.client_sock[random_num]
sock = socket(AF_INET, SOCK_STREAM)
sock.connect(addr)
data_s = data.getvalue()
ParaLiteLog.info("DATA SIZE = %s" % len(data_s))
sock.send("%10s%s" % (len(data_s), data_s))
re = sock.recv(10)
assert re == "OK"
sock.close()
