Merge remote-tracking branch 'upstream/master'

[nominatim.git] / nominatim / nominatim.py

diff --git a/nominatim/nominatim.py b/nominatim/nominatim.py
old mode 100644 (file)
new mode 100755 (executable)
index 49e2716..b20673d
--- a/nominatim/nominatim.py
+++ b/nominatim/nominatim.py
@@ -1,4 +1,4 @@
-#! /usr/bin/env python
+#! /usr/bin/env python3

 #-----------------------------------------------------------------------------
 # nominatim - [description]
 #-----------------------------------------------------------------------------
 #-----------------------------------------------------------------------------
 # nominatim - [description]
 #-----------------------------------------------------------------------------


@@ -28,17 +28,12 @@ import sys
 import re
 import getpass
 from datetime import datetime
 import re
 import getpass
 from datetime import datetime

-import psycopg2
-from psycopg2.extras import wait_select

 import 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.
 
 class RankRunner(object):
     """ Returns SQL commands for indexing one rank within the placex table.


@@ -50,24 +45,19 @@ class RankRunner(object):
     def name(self):
         return "rank {}".format(self.rank)
 
     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
         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)
 
                   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):
 
 
 class InterpolationRunner(object):


@@ -78,193 +68,146 @@ class InterpolationRunner(object):
     def name(self):
         return "interpolation lines (location_property_osmline)"
 
     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"""
 
         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
         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 signle non-blocking database connection.
+class BoundaryRunner(object):
+    """ Returns SQL commands for indexing the administrative boundaries
+        of a certain rank.

     """
 
     """
 

-    def __init__(self, options):
-        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
-
-    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, retry.")
-                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):
 
 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)]
 
         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.info("Starting indexing rank ({} to {}) using {} threads".format(
+        log.warning("Starting indexing rank ({} to {}) using {} threads".format(

                  self.minrank, self.maxrank, len(self.threads)))
 
                  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(RankRunner(rank))
 
         if self.maxrank == 30:

-            self.index(InterpolationRunner())
-            self.index(RankRunner(30))
+            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
         """ 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.info("Starting {}".format(obj.name()))
-
-        cur = self.conn.cursor(name='main')
-        cur.execute(obj.sql_index_sectors())
+        log.warning("Starting %s (using batch size %s)", obj.name(), batch)

 
 

-        total_tuples = 0
-        for r in cur:
-            total_tuples += r[1]
-        log.debug("Total number of rows; {}".format(total_tuples))
+        cur = self.conn.cursor()
+        cur.execute(obj.sql_count_objects())

 
 

-        cur.scroll(0, mode='absolute')
+        total_tuples = cur.fetchone()[0]
+        log.debug("Total number of rows: {}".format(total_tuples))

 
 

-        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
-
-        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 = next(next_thread)
 

-                thread.perform(index_sql, (place_id,))
-                done_tuples += 1
+                thread.perform(obj.sql_index_place(places))
+                progress.add(len(places))

 
 

-                if done_tuples >= next_info:
-                    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}"
-                           .format(done_tuples, int(done_time),
-                                   tuples_per_sec, obj.name(),
-                                   (total_tuples - done_tuples)/tuples_per_sec))
-                    next_info += int(tuples_per_sec)
+            cur.close()

 
 

-            pcur.close()
+            for t in self.threads:
+                t.wait()

 
 

-            if do_all:
-                break
-
-        cur.close()
-
-        for t in self.threads:
-            t.wait()
-
-        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(
-                 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
             sending a query.
         """
         ready = self.threads
 
     def find_free_thread(self):
         """ Generator that returns the next connection that is free for
             sending a query.
         """
         ready = self.threads

+        command_stat = 0

 
         while True:
             for thread in ready:
                 if thread.is_done():
 
         while True:
             for thread in ready:
                 if thread.is_done():

+                    command_stat += 1

                     yield thread
 
                     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():
+                        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():
 
 
 def nominatim_arg_parser():


@@ -291,6 +234,9 @@ def nominatim_arg_parser():
     p.add_argument('-P', '--port',
                    dest='port', action='store',
                    help='PostgreSQL server port')
     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.')
     p.add_argument('-r', '--minrank',
                    dest='minrank', type=int, metavar='RANK', default=0,
                    help='Minimum/starting rank.')


@@ -318,4 +264,7 @@ if __name__ == '__main__':
         password = getpass.getpass("Database password: ")
         options.password = password
 
         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()