ICU: use normalization from config in PHP

[nominatim.git] / nominatim / indexer / indexer.py

diff --git a/nominatim/indexer/indexer.py b/nominatim/indexer/indexer.py
index 520464565bebc96f04769b336c04604a531bab8f..50bd232e30dc0c341c6d43bbf8cf70eb720d37d0 100644 (file)
--- a/nominatim/indexer/indexer.py
+++ b/nominatim/indexer/indexer.py
@@ -1,191 +1,227 @@
 """
 Main work horse for indexing (computing addresses) the database.
 """
 """
 Main work horse for indexing (computing addresses) the database.
 """

-# pylint: disable=C0111

 import logging
 import logging

-import select
+import time

 
 

-from .progress import ProgressLogger
-from db.async_connection import DBConnection, make_connection
+import psycopg2.extras
+
+from nominatim.indexer.progress import ProgressLogger
+from nominatim.indexer import runners
+from nominatim.db.async_connection import DBConnection, WorkerPool
+from nominatim.db.connection import connect

 
 LOG = logging.getLogger()
 
 
 LOG = logging.getLogger()
 

-class RankRunner:
-    """ Returns SQL commands for indexing one rank within the placex table.
-    """

 
 

-    def __init__(self, rank):
-        self.rank = rank
+class PlaceFetcher:
+    """ Asynchronous connection that fetches place details for processing.
+    """
+    def __init__(self, dsn, setup_conn):
+        self.wait_time = 0
+        self.current_ids = None
+        self.conn = DBConnection(dsn, cursor_factory=psycopg2.extras.DictCursor)
+
+        with setup_conn.cursor() as cur:
+            # need to fetch those manually because register_hstore cannot
+            # fetch them on an asynchronous connection below.
+            hstore_oid = cur.scalar("SELECT 'hstore'::regtype::oid")
+            hstore_array_oid = cur.scalar("SELECT 'hstore[]'::regtype::oid")
+
+        psycopg2.extras.register_hstore(self.conn.conn, oid=hstore_oid,
+                                        array_oid=hstore_array_oid)
+
+    def close(self):
+        """ Close the underlying asynchronous connection.
+        """
+        if self.conn:
+            self.conn.close()
+            self.conn = None

 
 

-    def name(self):
-        return "rank {}".format(self.rank)

 
 

-    def sql_count_objects(self):
-        return """SELECT count(*) FROM placex
-                  WHERE rank_address = {} and indexed_status > 0
-               """.format(self.rank)
+    def fetch_next_batch(self, cur, runner):
+        """ Send a request for the next batch of places.
+            If details for the places are required, they will be fetched
+            asynchronously.

 
 

-    def sql_get_objects(self):
-        return """SELECT place_id FROM placex
-                  WHERE indexed_status > 0 and rank_address = {}
-                  ORDER BY geometry_sector""".format(self.rank)
+            Returns true if there is still data available.
+        """
+        ids = cur.fetchmany(100)

 
 

-    @staticmethod
-    def sql_index_place(ids):
-        return "UPDATE placex SET indexed_status = 0 WHERE place_id IN ({})"\
-               .format(','.join((str(i) for i in ids)))
+        if not ids:
+            self.current_ids = None
+            return False

 
 

+        if hasattr(runner, 'get_place_details'):
+            runner.get_place_details(self.conn, ids)
+            self.current_ids = []
+        else:
+            self.current_ids = ids

 
 

-class InterpolationRunner:
-    """ Returns SQL commands for indexing the address interpolation table
-        location_property_osmline.
-    """
+        return True

 
 

-    @staticmethod
-    def name():
-        return "interpolation lines (location_property_osmline)"
-
-    @staticmethod
-    def sql_count_objects():
-        return """SELECT count(*) FROM location_property_osmline
-                  WHERE indexed_status > 0"""
-
-    @staticmethod
-    def sql_get_objects():
-        return """SELECT place_id FROM location_property_osmline
-                  WHERE indexed_status > 0
-                  ORDER BY geometry_sector"""
-
-    @staticmethod
-    def sql_index_place(ids):
-        return """UPDATE location_property_osmline
-                  SET indexed_status = 0 WHERE place_id IN ({})"""\
-               .format(','.join((str(i) for i in ids)))
-
-class BoundaryRunner:
-    """ Returns SQL commands for indexing the administrative boundaries
-        of a certain rank.
-    """
+    def get_batch(self):
+        """ Get the next batch of data, previously requested with
+            `fetch_next_batch`.
+        """
+        if self.current_ids is not None and not self.current_ids:
+            tstart = time.time()
+            self.conn.wait()
+            self.wait_time += time.time() - tstart
+            self.current_ids = self.conn.cursor.fetchall()

 
 

-    def __init__(self, rank):
-        self.rank = rank
+        return self.current_ids

 
 

-    def name(self):
-        return "boundaries rank {}".format(self.rank)
+    def __enter__(self):
+        return self

 
 

-    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)

 
 

-    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 __exit__(self, exc_type, exc_value, traceback):
+        self.conn.wait()
+        self.close()

 
 

-    @staticmethod
-    def sql_index_place(ids):
-        return "UPDATE placex SET indexed_status = 0 WHERE place_id IN ({})"\
-               .format(','.join((str(i) for i in ids)))

 
 class Indexer:
     """ Main indexing routine.
     """
 
 
 class Indexer:
     """ Main indexing routine.
     """
 

-    def __init__(self, opts):
-        self.minrank = max(1, opts.minrank)
-        self.maxrank = min(30, opts.maxrank)
-        self.conn = make_connection(opts)
-        self.threads = [DBConnection(opts) for _ in range(opts.threads)]
+    def __init__(self, dsn, tokenizer, num_threads):
+        self.dsn = dsn
+        self.tokenizer = tokenizer
+        self.num_threads = num_threads
+
+
+    def has_pending(self):
+        """ Check if any data still needs indexing.
+            This function must only be used after the import has finished.
+            Otherwise it will be very expensive.
+        """
+        with connect(self.dsn) as conn:
+            with conn.cursor() as cur:
+                cur.execute("SELECT 'a' FROM placex WHERE indexed_status > 0 LIMIT 1")
+                return cur.rowcount > 0
+
+
+    def index_full(self, analyse=True):
+        """ Index the complete database. This will first index boundaries
+            followed by all other objects. When `analyse` is True, then the
+            database will be analysed at the appropriate places to
+            ensure that database statistics are updated.
+        """
+        with connect(self.dsn) as conn:
+            conn.autocommit = True
+
+            def _analyze():
+                if analyse:
+                    with conn.cursor() as cur:
+                        cur.execute('ANALYZE')
+
+            self.index_by_rank(0, 4)
+            _analyze()
+
+            self.index_boundaries(0, 30)
+            _analyze()
+
+            self.index_by_rank(5, 25)
+            _analyze()
+
+            self.index_by_rank(26, 30)
+            _analyze()

 
 

-    def index_boundaries(self):
+            self.index_postcodes()
+            _analyze()
+
+
+    def index_boundaries(self, minrank, maxrank):
+        """ Index only administrative boundaries within the given rank range.
+        """

         LOG.warning("Starting indexing boundaries using %s threads",
         LOG.warning("Starting indexing boundaries using %s threads",

-                    len(self.threads))
+                    self.num_threads)

 
 

-        for rank in range(max(self.minrank, 5), min(self.maxrank, 26)):
-            self.index(BoundaryRunner(rank))
+        with self.tokenizer.name_analyzer() as analyzer:
+            for rank in range(max(minrank, 4), min(maxrank, 26)):
+                self._index(runners.BoundaryRunner(rank, analyzer))

 
 

-    def index_by_rank(self):
-        """ Run classic indexing by rank.
+    def index_by_rank(self, minrank, maxrank):
+        """ Index all entries of placex in the given rank range (inclusive)
+            in order of their address rank.
+
+            When rank 30 is requested then also interpolations and
+            places with address rank 0 will be indexed.

         """
         """

+        maxrank = min(maxrank, 30)

         LOG.warning("Starting indexing rank (%i to %i) using %i threads",
         LOG.warning("Starting indexing rank (%i to %i) using %i threads",

-                    self.minrank, self.maxrank, len(self.threads))
+                    minrank, maxrank, self.num_threads)

 
 

-        for rank in range(max(1, self.minrank), self.maxrank):
-            self.index(RankRunner(rank))
+        with self.tokenizer.name_analyzer() as analyzer:
+            for rank in range(max(1, minrank), maxrank):
+                self._index(runners.RankRunner(rank, analyzer))

 
 

-        if self.maxrank == 30:
-            self.index(RankRunner(0))
-            self.index(InterpolationRunner(), 20)
-            self.index(RankRunner(self.maxrank), 20)
-        else:
-            self.index(RankRunner(self.maxrank))
+            if maxrank == 30:
+                self._index(runners.RankRunner(0, analyzer))
+                self._index(runners.InterpolationRunner(analyzer), 20)
+                self._index(runners.RankRunner(30, analyzer), 20)
+            else:
+                self._index(runners.RankRunner(maxrank, analyzer))

 
 

-    def index(self, obj, batch=1):
-        """ Index a single rank or table. `obj` describes the SQL to use
-            for indexing. `batch` describes the number of objects that
-            should be processed with a single SQL statement
+
+    def index_postcodes(self):
+        """Index the entries ofthe location_postcode table.

         """
         """

-        LOG.warning("Starting %s (using batch size %s)", obj.name(), batch)
+        LOG.warning("Starting indexing postcodes using %s threads", self.num_threads)

 
 

-        cur = self.conn.cursor()
-        cur.execute(obj.sql_count_objects())
+        self._index(runners.PostcodeRunner(), 20)

 
 

-        total_tuples = cur.fetchone()[0]
-        LOG.debug("Total number of rows: %i", total_tuples)

 
 

-        cur.close()
+    def update_status_table(self):
+        """ Update the status in the status table to 'indexed'.
+        """
+        with connect(self.dsn) as conn:
+            with conn.cursor() as cur:
+                cur.execute('UPDATE import_status SET indexed = true')

 
 

-        progress = ProgressLogger(obj.name(), total_tuples)
+            conn.commit()

 
 

-        if total_tuples > 0:
-            cur = self.conn.cursor(name='places')
-            cur.execute(obj.sql_get_objects())
+    def _index(self, runner, batch=1):
+        """ Index a single rank or table. `runner` describes the SQL to use
+            for indexing. `batch` describes the number of objects that
+            should be processed with a single SQL statement
+        """
+        LOG.warning("Starting %s (using batch size %s)", runner.name(), batch)

 
 

-            next_thread = self.find_free_thread()
-            while True:
-                places = [p[0] for p in cur.fetchmany(batch)]
-                if not places:
-                    break
+        with connect(self.dsn) as conn:
+            psycopg2.extras.register_hstore(conn)
+            with conn.cursor() as cur:
+                total_tuples = cur.scalar(runner.sql_count_objects())
+                LOG.debug("Total number of rows: %i", total_tuples)

 
 

-                LOG.debug("Processing places: %s", str(places))
-                thread = next(next_thread)
+            conn.commit()

 
 

-                thread.perform(obj.sql_index_place(places))
-                progress.add(len(places))
+            progress = ProgressLogger(runner.name(), total_tuples)

 
 

-            cur.close()
+            if total_tuples > 0:
+                with conn.cursor(name='places') as cur:
+                    cur.execute(runner.sql_get_objects())

 
 

-            for thread in self.threads:
-                thread.wait()
+                    with PlaceFetcher(self.dsn, conn) as fetcher:
+                        with WorkerPool(self.dsn, self.num_threads) as pool:
+                            has_more = fetcher.fetch_next_batch(cur, runner)
+                            while has_more:
+                                places = fetcher.get_batch()

 
 

-        progress.done()
+                                # asynchronously get the next batch
+                                has_more = fetcher.fetch_next_batch(cur, runner)

 
 

-    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():
-                    command_stat += 1
-                    yield thread
-
-            # refresh the connections occasionaly to avoid potential
-            # memory leaks in Postgresql.
-            if command_stat > 100000:
-                for thread in self.threads:
-                    while not thread.is_done():
-                        thread.wait()
-                    thread.connect()
-                command_stat = 0
-                ready = self.threads
-            else:
-                ready, _, _ = select.select(self.threads, [], [])
+                                # And insert the curent batch
+                                for idx in range(0, len(places), batch):
+                                    part = places[idx:idx + batch]
+                                    LOG.debug("Processing places: %s", str(part))
+                                    runner.index_places(pool.next_free_worker(), part)
+                                    progress.add(len(part))
+
+                            LOG.info("Wait time: fetcher: %.2fs,  pool: %.2fs",
+                                     fetcher.wait_time, pool.wait_time)

 
 

-        assert False, "Unreachable code"
+                conn.commit()
+
+        progress.done()