-#! /usr/bin/env python
+#! /usr/bin/env python3
#-----------------------------------------------------------------------------
# nominatim - [description]
#-----------------------------------------------------------------------------
log = logging.getLogger()
def make_connection(options, asynchronous=False):
- return psycopg2.connect(dbname=options.dbname, user=options.user,
- password=options.password, host=options.host,
- port=options.port, async_=asynchronous)
+ params = {'dbname' : options.dbname,
+ 'user' : options.user,
+ 'password' : options.password,
+ 'host' : options.host,
+ 'port' : options.port,
+ 'async' : asynchronous}
+
+ return psycopg2.connect(**params)
class RankRunner(object):
class DBConnection(object):
- """ A signle non-blocking database connection.
+ """ A single non-blocking database connection.
"""
def __init__(self, options):
+ self.current_query = None
+ self.current_params = None
+
+ self.conn = None
+ self.connect()
+
+ def connect(self):
+ if self.conn is not None:
+ self.cursor.close()
+ self.conn.close()
+
self.conn = make_connection(options, asynchronous=True)
self.wait()
self.cursor = self.conn.cursor()
- self.current_query = None
- self.current_params = None
-
def wait(self):
""" Block until any pending operation is done.
"""
- wait_select(self.conn)
- self.current_query = None
+ while True:
+ try:
+ wait_select(self.conn)
+ self.current_query = None
+ return
+ except psycopg2.extensions.TransactionRollbackError as e:
+ if e.pgcode == '40P01':
+ log.info("Deadlock detected (params = {}), retry."
+ .format(self.current_params))
+ self.cursor.execute(self.current_query, self.current_params)
+ else:
+ raise
+ except psycopg2.errors.DeadlockDetected:
+ self.cursor.execute(self.current_query, self.current_params)
def perform(self, sql, args=None):
""" Send SQL query to the server. Returns immediately without
return True
except psycopg2.extensions.TransactionRollbackError as e:
if e.pgcode == '40P01':
- log.info("Deadlock detected, retry.")
+ log.info("Deadlock detected (params = {}), retry.".format(self.current_params))
self.cursor.execute(self.current_query, self.current_params)
else:
raise
+ except psycopg2.errors.DeadlockDetected:
+ self.cursor.execute(self.current_query, self.current_params)
return False
def run(self):
""" Run indexing over the entire database.
"""
- log.info("Starting indexing rank ({} to {}) using {} threads".format(
+ log.warning("Starting indexing rank ({} to {}) using {} threads".format(
self.minrank, self.maxrank, len(self.threads)))
for rank in range(self.minrank, self.maxrank):
if self.maxrank == 30:
self.index(InterpolationRunner())
- self.index(RankRunner(30))
+
+ self.index(RankRunner(self.maxrank))
def index(self, obj):
""" Index a single rank or table. `obj` describes the SQL to use
for indexing.
"""
- log.info("Starting {}".format(obj.name()))
+ log.warning("Starting {}".format(obj.name()))
cur = self.conn.cursor(name='main')
cur.execute(obj.sql_index_sectors())
index_sql = obj.sql_index_place()
min_grouped_tuples = total_tuples - len(self.threads) * 1000
- next_info = 100
+ next_info = 100 if log.isEnabledFor(logging.INFO) else total_tuples + 1
for r in cur:
sector = r[0]
now = datetime.now()
done_time = (now - rank_start_time).total_seconds()
tuples_per_sec = done_tuples / done_time
- print("Done {} in {} @ {:.3f} per second - {} ETA (seconds): {:.2f}"
+ log.info("Done {} in {} @ {:.3f} per second - {} ETA (seconds): {:.2f}"
.format(done_tuples, int(done_time),
tuples_per_sec, obj.name(),
(total_tuples - done_tuples)/tuples_per_sec))
rank_end_time = datetime.now()
diff_seconds = (rank_end_time-rank_start_time).total_seconds()
- log.info("Done {}/{} in {} @ {:.3f} per second - FINISHED {}\n".format(
+ log.warning("Done {}/{} in {} @ {:.3f} per second - FINISHED {}\n".format(
done_tuples, total_tuples, int(diff_seconds),
done_tuples/diff_seconds, obj.name()))
sending a query.
"""
ready = self.threads
+ command_stat = 0
while True:
for thread in ready:
if thread.is_done():
+ command_stat += 1
yield thread
- ready, _, _ = select.select(self.threads, [], [])
+ # refresh the connections occasionaly to avoid potential
+ # memory leaks in Postgresql.
+ if command_stat > 100000:
+ for t in self.threads:
+ while not t.is_done():
+ wait_select(t.conn)
+ t.connect()
+ command_stat = 0
+ ready = self.threads
+ else:
+ ready, _, _ = select.select(self.threads, [], [])
assert(False, "Unreachable code")