import re
import getpass
from datetime import datetime
-import psycopg2
-from psycopg2.extras import wait_select
import select
-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)
+from indexer.progress import ProgressLogger
+from indexer.db import DBConnection, make_connection
+log = logging.getLogger()
class RankRunner(object):
""" Returns SQL commands for indexing one rank within the placex table.
def name(self):
return "rank {}".format(self.rank)
- def sql_index_sectors(self):
- return """SELECT geometry_sector, count(*) FROM placex
- WHERE rank_search = {} and indexed_status > 0
- GROUP BY geometry_sector
- ORDER BY geometry_sector""".format(self.rank)
+ def sql_count_objects(self):
+ return """SELECT count(*) FROM placex
+ WHERE rank_address = {} and indexed_status > 0
+ """.format(self.rank)
- def sql_nosector_places(self):
+ def sql_get_objects(self):
return """SELECT place_id FROM placex
- WHERE indexed_status > 0 and rank_search = {}
+ WHERE indexed_status > 0 and rank_address = {}
ORDER BY geometry_sector""".format(self.rank)
- def sql_sector_places(self):
- return """SELECT place_id FROM placex
- WHERE indexed_status > 0 and rank_search = {}
- and geometry_sector = %s""".format(self.rank)
-
- def sql_index_place(self):
- return "UPDATE placex SET indexed_status = 0 WHERE place_id = %s"
+ def sql_index_place(self, ids):
+ return "UPDATE placex SET indexed_status = 0 WHERE place_id IN ({})"\
+ .format(','.join((str(i) for i in ids)))
class InterpolationRunner(object):
def name(self):
return "interpolation lines (location_property_osmline)"
- def sql_index_sectors(self):
- return """SELECT geometry_sector, count(*) FROM location_property_osmline
- WHERE indexed_status > 0
- GROUP BY geometry_sector
- ORDER BY geometry_sector"""
+ def sql_count_objects(self):
+ return """SELECT count(*) FROM location_property_osmline
+ WHERE indexed_status > 0"""
- def sql_nosector_places(self):
+ def sql_get_objects(self):
return """SELECT place_id FROM location_property_osmline
WHERE indexed_status > 0
ORDER BY geometry_sector"""
- def sql_sector_places(self):
- return """SELECT place_id FROM location_property_osmline
- WHERE indexed_status > 0 and geometry_sector = %s
- ORDER BY geometry_sector"""
-
- def sql_index_place(self):
+ def sql_index_place(self, ids):
return """UPDATE location_property_osmline
- SET indexed_status = 0 WHERE place_id = %s"""
-
+ SET indexed_status = 0 WHERE place_id IN ({})"""\
+ .format(','.join((str(i) for i in ids)))
-class DBConnection(object):
- """ A single non-blocking database connection.
+class BoundaryRunner(object):
+ """ Returns SQL commands for indexing the administrative boundaries
+ of a certain rank.
"""
- 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()
-
- def wait(self):
- """ Block until any pending operation is done.
- """
- wait_select(self.conn)
- self.current_query = None
-
- def perform(self, sql, args=None):
- """ Send SQL query to the server. Returns immediately without
- blocking.
- """
- self.current_query = sql
- self.current_params = args
- self.cursor.execute(sql, args)
-
- def fileno(self):
- """ File descriptor to wait for. (Makes this class select()able.)
- """
- return self.conn.fileno()
+ def __init__(self, rank):
+ self.rank = rank
- def is_done(self):
- """ Check if the connection is available for a new query.
+ def name(self):
+ return "boundaries rank {}".format(self.rank)
- Also checks if the previous query has run into a deadlock.
- If so, then the previous query is repeated.
- """
- if self.current_query is None:
- return True
-
- try:
- if self.conn.poll() == psycopg2.extensions.POLL_OK:
- self.current_query = None
- return True
- 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
+ def sql_count_objects(self):
+ return """SELECT count(*) FROM placex
+ WHERE indexed_status > 0
+ AND rank_search = {}
+ AND class = 'boundary' and type = 'administrative'""".format(self.rank)
- return False
+ def sql_get_objects(self):
+ return """SELECT place_id FROM placex
+ WHERE indexed_status > 0 and rank_search = {}
+ and class = 'boundary' and type = 'administrative'
+ ORDER BY partition, admin_level""".format(self.rank)
+ def sql_index_place(self, ids):
+ return "UPDATE placex SET indexed_status = 0 WHERE place_id IN ({})"\
+ .format(','.join((str(i) for i in ids)))
class Indexer(object):
""" Main indexing routine.
"""
def __init__(self, options):
- self.minrank = max(0, options.minrank)
+ self.minrank = max(1, options.minrank)
self.maxrank = min(30, options.maxrank)
self.conn = make_connection(options)
self.threads = [DBConnection(options) for i in range(options.threads)]
- def run(self):
- """ Run indexing over the entire database.
+ def index_boundaries(self):
+ log.warning("Starting indexing boundaries using {} threads".format(
+ len(self.threads)))
+
+ for rank in range(max(self.minrank, 5), min(self.maxrank, 26)):
+ self.index(BoundaryRunner(rank))
+
+ def index_by_rank(self):
+ """ Run classic indexing by rank.
"""
log.warning("Starting indexing rank ({} to {}) using {} threads".format(
self.minrank, self.maxrank, len(self.threads)))
- for rank in range(self.minrank, self.maxrank):
+ for rank in range(max(1, self.minrank), self.maxrank):
self.index(RankRunner(rank))
if self.maxrank == 30:
- self.index(InterpolationRunner())
-
- self.index(RankRunner(self.maxrank))
+ self.index(RankRunner(0))
+ self.index(InterpolationRunner(), 20)
+ self.index(RankRunner(self.maxrank), 20)
+ else:
+ self.index(RankRunner(self.maxrank))
- def index(self, obj):
+ def index(self, obj, batch=1):
""" Index a single rank or table. `obj` describes the SQL to use
- for indexing.
+ for indexing. `batch` describes the number of objects that
+ should be processed with a single SQL statement
"""
- log.warning("Starting {}".format(obj.name()))
+ log.warning("Starting %s (using batch size %s)", obj.name(), batch)
- cur = self.conn.cursor(name='main')
- cur.execute(obj.sql_index_sectors())
+ cur = self.conn.cursor()
+ cur.execute(obj.sql_count_objects())
- total_tuples = 0
- for r in cur:
- total_tuples += r[1]
- log.debug("Total number of rows; {}".format(total_tuples))
+ total_tuples = cur.fetchone()[0]
+ log.debug("Total number of rows: {}".format(total_tuples))
- cur.scroll(0, mode='absolute')
-
- next_thread = self.find_free_thread()
- done_tuples = 0
- rank_start_time = datetime.now()
-
- sector_sql = obj.sql_sector_places()
- index_sql = obj.sql_index_place()
- min_grouped_tuples = total_tuples - len(self.threads) * 1000
-
- next_info = 100 if log.isEnabledFor(logging.INFO) else total_tuples + 1
-
- for r in cur:
- sector = r[0]
+ cur.close()
- # Should we do the remaining ones together?
- do_all = done_tuples > min_grouped_tuples
+ progress = ProgressLogger(obj.name(), total_tuples)
- pcur = self.conn.cursor(name='places')
+ if total_tuples > 0:
+ cur = self.conn.cursor(name='places')
+ cur.execute(obj.sql_get_objects())
- if do_all:
- pcur.execute(obj.sql_nosector_places())
- else:
- pcur.execute(sector_sql, (sector, ))
+ next_thread = self.find_free_thread()
+ while True:
+ places = [p[0] for p in cur.fetchmany(batch)]
+ if len(places) == 0:
+ break
- for place in pcur:
- place_id = place[0]
- log.debug("Processing place {}".format(place_id))
+ log.debug("Processing places: {}".format(places))
thread = next(next_thread)
- thread.perform(index_sql, (place_id,))
- done_tuples += 1
-
- if done_tuples >= next_info:
- now = datetime.now()
- done_time = (now - rank_start_time).total_seconds()
- tuples_per_sec = done_tuples / done_time
- 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))
- next_info += int(tuples_per_sec)
-
- pcur.close()
-
- if do_all:
- break
-
- cur.close()
+ thread.perform(obj.sql_index_place(places))
+ progress.add(len(places))
- for t in self.threads:
- t.wait()
+ cur.close()
- rank_end_time = datetime.now()
- diff_seconds = (rank_end_time-rank_start_time).total_seconds()
+ for t in self.threads:
+ t.wait()
- log.warning("Done {}/{} in {} @ {:.3f} per second - FINISHED {}\n".format(
- done_tuples, total_tuples, int(diff_seconds),
- done_tuples/diff_seconds, obj.name()))
+ progress.done()
def find_free_thread(self):
""" Generator that returns the next connection that is free for
if command_stat > 100000:
for t in self.threads:
while not t.is_done():
- wait_select(t.conn)
+ t.wait()
t.connect()
command_stat = 0
ready = self.threads
else:
ready, _, _ = select.select(self.threads, [], [])
- assert(False, "Unreachable code")
+ assert False, "Unreachable code"
def nominatim_arg_parser():
p.add_argument('-P', '--port',
dest='port', action='store',
help='PostgreSQL server port')
+ p.add_argument('-b', '--boundary-only',
+ dest='boundary_only', action='store_true',
+ help='Only index administrative boundaries (ignores min/maxrank).')
p.add_argument('-r', '--minrank',
dest='minrank', type=int, metavar='RANK', default=0,
help='Minimum/starting rank.')
password = getpass.getpass("Database password: ")
options.password = password
- Indexer(options).run()
+ if options.boundary_only:
+ Indexer(options).index_boundaries()
+ else:
+ Indexer(options).index_by_rank()
projects / nominatim.git / blobdiff