Merge remote-tracking branch 'upstream/master'

[nominatim.git] / nominatim / tools / database_import.py

diff --git a/nominatim/tools/database_import.py b/nominatim/tools/database_import.py
index 00ec95c03ab49553f5eb3bd41b693f25b319efad..50938c19abf2e02fa1ad7647d64a911064278799 100644 (file)
--- a/nominatim/tools/database_import.py
+++ b/nominatim/tools/database_import.py
@@ -1,25 +1,48 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# This file is part of Nominatim. (https://nominatim.org)
+#
+# Copyright (C) 2022 by the Nominatim developer community.
+# For a full list of authors see the git log.

 """
 Functions for setting up and importing a new Nominatim database.
 """
 import logging
 import os
 """
 Functions for setting up and importing a new Nominatim database.
 """
 import logging
 import os

+import selectors

 import subprocess
 import subprocess

-import shutil

 from pathlib import Path
 
 import psutil
 from pathlib import Path
 
 import psutil

+from psycopg2 import sql as pysql

 
 

-from ..db.connection import connect, get_pg_env
-from ..db import utils as db_utils
-from .exec_utils import run_osm2pgsql
-from ..errors import UsageError
-from ..version import POSTGRESQL_REQUIRED_VERSION, POSTGIS_REQUIRED_VERSION
+from nominatim.db.connection import connect, get_pg_env
+from nominatim.db.async_connection import DBConnection
+from nominatim.db.sql_preprocessor import SQLPreprocessor
+from nominatim.tools.exec_utils import run_osm2pgsql
+from nominatim.errors import UsageError
+from nominatim.version import POSTGRESQL_REQUIRED_VERSION, POSTGIS_REQUIRED_VERSION

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

-def create_db(dsn, rouser=None):
-    """ Create a new database for the given DSN. Fails when the database
-        already exists or the PostgreSQL version is too old.
+def _require_version(module, actual, expected):
+    """ Compares the version for the given module and raises an exception
+        if the actual version is too old.
+    """
+    if actual < expected:
+        LOG.fatal('Minimum supported version of %s is %d.%d. '
+                  'Found version %d.%d.',
+                  module, expected[0], expected[1], actual[0], actual[1])
+        raise UsageError(f'{module} is too old.')
+
+
+def setup_database_skeleton(dsn, rouser=None):
+    """ Create a new database for Nominatim and populate it with the
+        essential extensions.
+
+        The function fails when the database already exists or Postgresql or
+        PostGIS versions are too old.
+

         Uses `createdb` to create the database.
 
         If 'rouser' is given, then the function also checks that the user
         Uses `createdb` to create the database.
 
         If 'rouser' is given, then the function also checks that the user


@@ -33,13 +56,9 @@ def create_db(dsn, rouser=None):
         raise UsageError('Creating new database failed.')
 
     with connect(dsn) as conn:
         raise UsageError('Creating new database failed.')
 
     with connect(dsn) as conn:

-        postgres_version = conn.server_version_tuple()
-        if postgres_version < POSTGRESQL_REQUIRED_VERSION:
-            LOG.fatal('Minimum supported version of Postgresql is %d.%d. '
-                      'Found version %d.%d.',
-                      POSTGRESQL_REQUIRED_VERSION[0], POSTGRESQL_REQUIRED_VERSION[1],
-                      postgres_version[0], postgres_version[1])
-            raise UsageError('PostgreSQL server is too old.')
+        _require_version('PostgreSQL server',
+                         conn.server_version_tuple(),
+                         POSTGRESQL_REQUIRED_VERSION)

 
         if rouser is not None:
             with conn.cursor() as cur:
 
         if rouser is not None:
             with conn.cursor() as cur:


@@ -50,109 +69,174 @@ def create_db(dsn, rouser=None):
                               "\n      createuser %s", rouser, rouser)
                     raise UsageError('Missing read-only user.')
 
                               "\n      createuser %s", rouser, rouser)
                     raise UsageError('Missing read-only user.')
 

+        # Create extensions.
+        with conn.cursor() as cur:
+            cur.execute('CREATE EXTENSION IF NOT EXISTS hstore')
+            cur.execute('CREATE EXTENSION IF NOT EXISTS postgis')
+        conn.commit()
+
+        _require_version('PostGIS',
+                         conn.postgis_version_tuple(),
+                         POSTGIS_REQUIRED_VERSION)

 
 
 
 

-def setup_extensions(conn):
-    """ Set up all extensions needed for Nominatim. Also checks that the
-        versions of the extensions are sufficient.
+def import_osm_data(osm_files, options, drop=False, ignore_errors=False):
+    """ Import the given OSM files. 'options' contains the list of
+        default settings for osm2pgsql.

     """
     """

-    with conn.cursor() as cur:
-        cur.execute('CREATE EXTENSION IF NOT EXISTS hstore')
-        cur.execute('CREATE EXTENSION IF NOT EXISTS postgis')
-    conn.commit()
+    options['import_file'] = osm_files
+    options['append'] = False
+    options['threads'] = 1

 
 

-    postgis_version = conn.postgis_version_tuple()
-    if postgis_version < POSTGIS_REQUIRED_VERSION:
-        LOG.fatal('Minimum supported version of PostGIS is %d.%d. '
-                  'Found version %d.%d.',
-                  POSTGIS_REQUIRED_VERSION[0], POSTGIS_REQUIRED_VERSION[1],
-                  postgis_version[0], postgis_version[1])
-        raise UsageError('PostGIS version is too old.')
+    if not options['flatnode_file'] and options['osm2pgsql_cache'] == 0:
+        # Make some educated guesses about cache size based on the size
+        # of the import file and the available memory.
+        mem = psutil.virtual_memory()
+        fsize = 0
+        if isinstance(osm_files, list):
+            for fname in osm_files:
+                fsize += os.stat(str(fname)).st_size
+        else:
+            fsize = os.stat(str(osm_files)).st_size
+        options['osm2pgsql_cache'] = int(min((mem.available + mem.cached) * 0.75,
+                                             fsize * 2) / 1024 / 1024) + 1

 
 

+    run_osm2pgsql(options)
+
+    with connect(options['dsn']) as conn:
+        if not ignore_errors:
+            with conn.cursor() as cur:
+                cur.execute('SELECT * FROM place LIMIT 1')
+                if cur.rowcount == 0:
+                    raise UsageError('No data imported by osm2pgsql.')
+
+        if drop:
+            conn.drop_table('planet_osm_nodes')

 
 

-def install_module(src_dir, project_dir, module_dir):
-    """ Copy the normalization module from src_dir into the project
-        directory under the '/module' directory. If 'module_dir' is set, then
-        use the module from there instead and check that it is accessible
-        for Postgresql.
+    if drop and options['flatnode_file']:
+        Path(options['flatnode_file']).unlink()

 
 

-        The function detects when the installation is run from the
-        build directory. It doesn't touch the module in that case.
+
+def create_tables(conn, config, reverse_only=False):
+    """ Create the set of basic tables.
+        When `reverse_only` is True, then the main table for searching will
+        be skipped and only reverse search is possible.

     """
     """

-    if not module_dir:
-        module_dir = project_dir / 'module'
+    sql = SQLPreprocessor(conn, config)
+    sql.env.globals['db']['reverse_only'] = reverse_only

 
 

-        if not module_dir.exists() or not src_dir.samefile(module_dir):
+    sql.run_sql_file(conn, 'tables.sql')

 
 

-            if not module_dir.exists():
-                module_dir.mkdir()

 
 

-            destfile = module_dir / 'nominatim.so'
-            shutil.copy(str(src_dir / 'nominatim.so'), str(destfile))
-            destfile.chmod(0o755)
+def create_table_triggers(conn, config):
+    """ Create the triggers for the tables. The trigger functions must already
+        have been imported with refresh.create_functions().
+    """
+    sql = SQLPreprocessor(conn, config)
+    sql.run_sql_file(conn, 'table-triggers.sql')

 
 

-            LOG.info('Database module installed at %s', str(destfile))
-        else:
-            LOG.info('Running from build directory. Leaving database module as is.')
-    else:
-        LOG.info("Using custom path for database module at '%s'", module_dir)

 
 

-    return module_dir
+def create_partition_tables(conn, config):
+    """ Create tables that have explicit partitioning.
+    """
+    sql = SQLPreprocessor(conn, config)
+    sql.run_sql_file(conn, 'partition-tables.src.sql')

 
 
 
 

-def check_module_dir_path(conn, path):
-    """ Check that the normalisation module can be found and executed
-        from the given path.
+def truncate_data_tables(conn):
+    """ Truncate all data tables to prepare for a fresh load.

     """
     with conn.cursor() as cur:
     """
     with conn.cursor() as cur:

-        cur.execute("""CREATE FUNCTION nominatim_test_import_func(text)
-                       RETURNS text AS '{}/nominatim.so', 'transliteration'
-                       LANGUAGE c IMMUTABLE STRICT;
-                       DROP FUNCTION nominatim_test_import_func(text)
-                    """.format(path))
+        cur.execute('TRUNCATE placex')
+        cur.execute('TRUNCATE place_addressline')
+        cur.execute('TRUNCATE location_area')
+        cur.execute('TRUNCATE location_area_country')
+        cur.execute('TRUNCATE location_property_tiger')
+        cur.execute('TRUNCATE location_property_osmline')
+        cur.execute('TRUNCATE location_postcode')
+        if conn.table_exists('search_name'):
+            cur.execute('TRUNCATE search_name')
+        cur.execute('DROP SEQUENCE IF EXISTS seq_place')
+        cur.execute('CREATE SEQUENCE seq_place start 100000')
+
+        cur.execute("""SELECT tablename FROM pg_tables
+                       WHERE tablename LIKE 'location_road_%'""")
+
+        for table in [r[0] for r in list(cur)]:
+            cur.execute('TRUNCATE ' + table)

 
 

+    conn.commit()

 
 

-def import_base_data(dsn, sql_dir, ignore_partitions=False):
-    """ Create and populate the tables with basic static data that provides
-        the background for geocoding. Data is assumed to not yet exist.
-    """
-    db_utils.execute_file(dsn, sql_dir / 'country_name.sql')
-    db_utils.execute_file(dsn, sql_dir / 'country_osm_grid.sql.gz')

 
 

-    if ignore_partitions:
-        with connect(dsn) as conn:
-            with conn.cursor() as cur:
-                cur.execute('UPDATE country_name SET partition = 0')
-            conn.commit()
+_COPY_COLUMNS = pysql.SQL(',').join(map(pysql.Identifier,
+                                        ('osm_type', 'osm_id', 'class', 'type',
+                                         'name', 'admin_level', 'address',
+                                         'extratags', 'geometry')))

 
 
 
 

-def import_osm_data(osm_file, options, drop=False):
-    """ Import the given OSM file. 'options' contains the list of
-        default settings for osm2pgsql.
+def load_data(dsn, threads):
+    """ Copy data into the word and placex table.

     """
     """

-    options['import_file'] = osm_file
-    options['append'] = False
-    options['threads'] = 1
+    sel = selectors.DefaultSelector()
+    # Then copy data from place to placex in <threads - 1> chunks.
+    place_threads = max(1, threads - 1)
+    for imod in range(place_threads):
+        conn = DBConnection(dsn)
+        conn.connect()
+        conn.perform(
+            pysql.SQL("""INSERT INTO placex ({columns})
+                           SELECT {columns} FROM place
+                           WHERE osm_id % {total} = {mod}
+                             AND NOT (class='place' and (type='houses' or type='postcode'))
+                             AND ST_IsValid(geometry)
+                      """).format(columns=_COPY_COLUMNS,
+                                  total=pysql.Literal(place_threads),
+                                  mod=pysql.Literal(imod)))
+        sel.register(conn, selectors.EVENT_READ, conn)
+
+    # Address interpolations go into another table.
+    conn = DBConnection(dsn)
+    conn.connect()
+    conn.perform("""INSERT INTO location_property_osmline (osm_id, address, linegeo)
+                      SELECT osm_id, address, geometry FROM place
+                      WHERE class='place' and type='houses' and osm_type='W'
+                            and ST_GeometryType(geometry) = 'ST_LineString'
+                 """)
+    sel.register(conn, selectors.EVENT_READ, conn)
+
+    # Now wait for all of them to finish.
+    todo = place_threads + 1
+    while todo > 0:
+        for key, _ in sel.select(1):
+            conn = key.data
+            sel.unregister(conn)
+            conn.wait()
+            conn.close()
+            todo -= 1
+        print('.', end='', flush=True)
+    print('\n')

 
 

-    if not options['flatnode_file'] and options['osm2pgsql_cache'] == 0:
-        # Make some educated guesses about cache size based on the size
-        # of the import file and the available memory.
-        mem = psutil.virtual_memory()
-        fsize = os.stat(str(osm_file)).st_size
-        options['osm2pgsql_cache'] = int(min((mem.available + mem.cached) * 0.75,
-                                             fsize * 2) / 1024 / 1024) + 1
+    with connect(dsn) as conn:
+        with conn.cursor() as cur:
+            cur.execute('ANALYSE')

 
 

-    run_osm2pgsql(options)

 
 

-    with connect(options['dsn']) as conn:
-        with conn.cursor() as cur:
-            cur.execute('SELECT * FROM place LIMIT 1')
-            if cur.rowcount == 0:
-                raise UsageError('No data imported by osm2pgsql.')
+def create_search_indices(conn, config, drop=False):
+    """ Create tables that have explicit partitioning.
+    """

 
 

-        if drop:
-            conn.drop_table('planet_osm_nodes')
+    # If index creation failed and left an index invalid, they need to be
+    # cleaned out first, so that the script recreates them.
+    with conn.cursor() as cur:
+        cur.execute("""SELECT relname FROM pg_class, pg_index
+                       WHERE pg_index.indisvalid = false
+                             AND pg_index.indexrelid = pg_class.oid""")
+        bad_indices = [row[0] for row in list(cur)]
+        for idx in bad_indices:
+            LOG.info("Drop invalid index %s.", idx)
+            cur.execute(pysql.SQL('DROP INDEX {}').format(pysql.Identifier(idx)))
+    conn.commit()
+
+    sql = SQLPreprocessor(conn, config)

 
 

-    if drop:
-        if options['flatnode_file']:
-            Path(options['flatnode_file']).unlink()
+    sql.run_sql_file(conn, 'indices.sql', drop=drop)