X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/32683f73c787464e16f2a146d4c08c4041087dd5..d9639a98671786f74b2be7588f4af2405ea08a68:/nominatim/tools/database_import.py?ds=sidebyside diff --git a/nominatim/tools/database_import.py b/nominatim/tools/database_import.py index 00ec95c0..cb620d41 100644 --- a/nominatim/tools/database_import.py +++ b/nominatim/tools/database_import.py @@ -1,25 +1,50 @@ +# 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. """ +from typing import Tuple, Optional, Union, Sequence, MutableMapping, Any import logging import os +import selectors import subprocess -import shutil 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.config import Configuration +from nominatim.db.connection import connect, get_pg_env, Connection +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() -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: str, actual: Tuple[int, int], expected: Tuple[int, int]) -> None: + """ 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: str, rouser: Optional[str] = None) -> 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 @@ -33,126 +58,196 @@ def create_db(dsn, rouser=None): 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: cnt = cur.scalar('SELECT count(*) FROM pg_user where usename = %s', (rouser, )) if cnt == 0: - LOG.fatal("Web user '%s' does not exists. Create it with:\n" + LOG.fatal("Web user '%s' does not exist. Create it with:\n" "\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') + postgis_version = conn.postgis_version_tuple() + if postgis_version[0] >= 3: + cur.execute('CREATE EXTENSION IF NOT EXISTS postgis_raster') -def setup_extensions(conn): - """ Set up all extensions needed for Nominatim. Also checks that the - versions of the extensions are sufficient. - """ - with conn.cursor() as cur: - cur.execute('CREATE EXTENSION IF NOT EXISTS hstore') - cur.execute('CREATE EXTENSION IF NOT EXISTS postgis') - conn.commit() - - 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.') + conn.commit() + _require_version('PostGIS', + conn.postgis_version_tuple(), + POSTGIS_REQUIRED_VERSION) -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. - The function detects when the installation is run from the - build directory. It doesn't touch the module in that case. +def import_osm_data(osm_files: Union[Path, Sequence[Path]], + options: MutableMapping[str, Any], + drop: bool = False, ignore_errors: bool = False) -> None: + """ Import the given OSM files. 'options' contains the list of + default settings for osm2pgsql. """ - if not module_dir: - module_dir = project_dir / 'module' + options['import_file'] = osm_files + options['append'] = False + options['threads'] = 1 - if not module_dir.exists() or not src_dir.samefile(module_dir): + 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 - if not module_dir.exists(): - module_dir.mkdir() + run_osm2pgsql(options) - destfile = module_dir / 'nominatim.so' - shutil.copy(str(src_dir / 'nominatim.so'), str(destfile)) - destfile.chmod(0o755) + 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.') - 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) + if drop: + conn.drop_table('planet_osm_nodes') - return module_dir + if drop and options['flatnode_file']: + Path(options['flatnode_file']).unlink() -def check_module_dir_path(conn, path): - """ Check that the normalisation module can be found and executed - from the given path. +def create_tables(conn: Connection, config: Configuration, reverse_only: bool = False) -> None: + """ 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. """ - 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)) + sql = SQLPreprocessor(conn, config) + sql.env.globals['db']['reverse_only'] = reverse_only + + sql.run_sql_file(conn, 'tables.sql') -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. +def create_table_triggers(conn: Connection, config: Configuration) -> None: + """ Create the triggers for the tables. The trigger functions must already + have been imported with refresh.create_functions(). """ - db_utils.execute_file(dsn, sql_dir / 'country_name.sql') - db_utils.execute_file(dsn, sql_dir / 'country_osm_grid.sql.gz') + sql = SQLPreprocessor(conn, config) + sql.run_sql_file(conn, 'table-triggers.sql') - if ignore_partitions: - with connect(dsn) as conn: - with conn.cursor() as cur: - cur.execute('UPDATE country_name SET partition = 0') - conn.commit() +def create_partition_tables(conn: Connection, config: Configuration) -> None: + """ Create tables that have explicit partitioning. + """ + sql = SQLPreprocessor(conn, config) + sql.run_sql_file(conn, 'partition-tables.src.sql') -def import_osm_data(osm_file, options, drop=False): - """ Import the given OSM file. 'options' contains the list of - default settings for osm2pgsql. + +def truncate_data_tables(conn: Connection) -> None: + """ Truncate all data tables to prepare for a fresh load. """ - options['import_file'] = osm_file - options['append'] = False - options['threads'] = 1 + with conn.cursor() as cur: + 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) - 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 + conn.commit() - 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.') +_COPY_COLUMNS = pysql.SQL(',').join(map(pysql.Identifier, + ('osm_type', 'osm_id', 'class', 'type', + 'name', 'admin_level', 'address', + 'extratags', 'geometry'))) - if drop: - conn.drop_table('planet_osm_nodes') - if drop: - if options['flatnode_file']: - Path(options['flatnode_file']).unlink() +def load_data(dsn: str, threads: int) -> None: + """ Copy data into the word and placex table. + """ + sel = selectors.DefaultSelector() + # Then copy data from place to placex in 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') + + with connect(dsn) as syn_conn: + with syn_conn.cursor() as cur: + cur.execute('ANALYSE') + + +def create_search_indices(conn: Connection, config: Configuration, + drop: bool = False, threads: int = 1) -> None: + """ Create tables that have explicit partitioning. + """ + + # 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) + + sql.run_parallel_sql_file(config.get_libpq_dsn(), + 'indices.sql', min(8, threads), drop=drop)