X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/6ce6f62b8ef7cec8b5950293516845e319dd5f06..a0cd96e05ebad69c38f3406b40f5a6c615b40b4a:/nominatim/indexer/indexer.py?ds=inline diff --git a/nominatim/indexer/indexer.py b/nominatim/indexer/indexer.py index d685e83a..5425c8ff 100644 --- a/nominatim/indexer/indexer.py +++ b/nominatim/indexer/indexer.py @@ -1,80 +1,95 @@ +# 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. """ Main work horse for indexing (computing addresses) the database. """ +from typing import Optional, Any, cast import logging -import select import time import psycopg2.extras +from nominatim.tokenizer.base import AbstractTokenizer from nominatim.indexer.progress import ProgressLogger from nominatim.indexer import runners -from nominatim.db.async_connection import DBConnection -from nominatim.db.connection import connect +from nominatim.db.async_connection import DBConnection, WorkerPool +from nominatim.db.connection import connect, Connection, Cursor +from nominatim.typing import DictCursorResults LOG = logging.getLogger() -class WorkerPool: - """ A pool of asynchronous database connections. - The pool may be used as a context manager. +class PlaceFetcher: + """ Asynchronous connection that fetches place details for processing. """ - REOPEN_CONNECTIONS_AFTER = 100000 + def __init__(self, dsn: str, setup_conn: Connection) -> None: + self.wait_time = 0.0 + self.current_ids: Optional[DictCursorResults] = None + self.conn: Optional[DBConnection] = 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) -> None: + """ Close the underlying asynchronous connection. + """ + if self.conn: + self.conn.close() + self.conn = None - def __init__(self, dsn, pool_size): - self.threads = [DBConnection(dsn) for _ in range(pool_size)] - self.free_workers = self._yield_free_worker() + def fetch_next_batch(self, cur: Cursor, runner: runners.Runner) -> bool: + """ Send a request for the next batch of places. + If details for the places are required, they will be fetched + asynchronously. - def finish_all(self): - """ Wait for all connection to finish. + Returns true if there is still data available. """ - for thread in self.threads: - while not thread.is_done(): - thread.wait() + ids = cast(Optional[DictCursorResults], cur.fetchmany(100)) - self.free_workers = self._yield_free_worker() + if not ids: + self.current_ids = None + return False - def close(self): - """ Close all connections and clear the pool. - """ - for thread in self.threads: - thread.close() - self.threads = [] - self.free_workers = None + assert self.conn is not None + self.current_ids = runner.get_place_details(self.conn, ids) + return True - def next_free_worker(self): - """ Get the next free connection. + def get_batch(self) -> DictCursorResults: + """ Get the next batch of data, previously requested with + `fetch_next_batch`. """ - return next(self.free_workers) - - - def _yield_free_worker(self): - ready = self.threads - command_stat = 0 - while True: - for thread in ready: - if thread.is_done(): - command_stat += 1 - yield thread - - if command_stat > self.REOPEN_CONNECTIONS_AFTER: - for thread in self.threads: - while not thread.is_done(): - thread.wait() - thread.connect() - ready = self.threads - command_stat = 0 - else: - _, ready, _ = select.select([], self.threads, []) - - - def __enter__(self): + assert self.conn is not None + assert self.conn.cursor is not None + + 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 = cast(Optional[DictCursorResults], + self.conn.cursor.fetchall()) + + return self.current_ids if self.current_ids is not None else [] + + def __enter__(self) -> 'PlaceFetcher': return self - def __exit__(self, exc_type, exc_value, traceback): + def __exit__(self, exc_type: Any, exc_value: Any, traceback: Any) -> None: + assert self.conn is not None + self.conn.wait() self.close() @@ -82,14 +97,25 @@ class Indexer: """ Main indexing routine. """ - def __init__(self, dsn, tokenizer, num_threads): + def __init__(self, dsn: str, tokenizer: AbstractTokenizer, num_threads: int): self.dsn = dsn self.tokenizer = tokenizer self.num_threads = num_threads - def index_full(self, analyse=True): - """ Index the complete database. This will first index boudnaries + def has_pending(self) -> bool: + """ 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: bool = True) -> None: + """ 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. @@ -97,13 +123,10 @@ class Indexer: with connect(self.dsn) as conn: conn.autocommit = True - if analyse: - def _analyze(): + def _analyze() -> None: + if analyse: with conn.cursor() as cur: cur.execute('ANALYZE') - else: - def _analyze(): - pass self.index_by_rank(0, 4) _analyze() @@ -121,7 +144,7 @@ class Indexer: _analyze() - def index_boundaries(self, minrank, maxrank): + def index_boundaries(self, minrank: int, maxrank: int) -> None: """ Index only administrative boundaries within the given rank range. """ LOG.warning("Starting indexing boundaries using %s threads", @@ -131,7 +154,7 @@ class Indexer: for rank in range(max(minrank, 4), min(maxrank, 26)): self._index(runners.BoundaryRunner(rank, analyzer)) - def index_by_rank(self, minrank, maxrank): + def index_by_rank(self, minrank: int, maxrank: int) -> None: """ Index all entries of placex in the given rank range (inclusive) in order of their address rank. @@ -143,26 +166,23 @@ class Indexer: minrank, maxrank, self.num_threads) with self.tokenizer.name_analyzer() as analyzer: - for rank in range(max(1, minrank), maxrank): - self._index(runners.RankRunner(rank, analyzer)) + for rank in range(max(1, minrank), maxrank + 1): + self._index(runners.RankRunner(rank, analyzer), 20 if rank == 30 else 1) 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_postcodes(self): - """Index the entries ofthe location_postcode table. + def index_postcodes(self) -> None: + """Index the entries of the location_postcode table. """ LOG.warning("Starting indexing postcodes using %s threads", self.num_threads) self._index(runners.PostcodeRunner(), 20) - def update_status_table(self): + def update_status_table(self) -> None: """ Update the status in the status table to 'indexed'. """ with connect(self.dsn) as conn: @@ -171,7 +191,7 @@ class Indexer: conn.commit() - def _index(self, runner, batch=1): + def _index(self, runner: runners.Runner, batch: int = 1) -> None: """ 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 @@ -184,70 +204,33 @@ class Indexer: total_tuples = cur.scalar(runner.sql_count_objects()) LOG.debug("Total number of rows: %i", total_tuples) - # 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") - conn.commit() progress = ProgressLogger(runner.name(), total_tuples) - fetcher_wait = 0 - pool_wait = 0 - if total_tuples > 0: with conn.cursor(name='places') as cur: cur.execute(runner.sql_get_objects()) - fetcher = DBConnection(self.dsn, cursor_factory=psycopg2.extras.DictCursor) - psycopg2.extras.register_hstore(fetcher.conn, - oid=hstore_oid, - array_oid=hstore_array_oid) - - with WorkerPool(self.dsn, self.num_threads) as pool: - places = self._fetch_next_batch(cur, fetcher, runner) - while places is not None: - if not places: - t0 = time.time() - fetcher.wait() - fetcher_wait += time.time() - t0 - places = fetcher.cursor.fetchall() + 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() - # asynchronously get the next batch - next_places = self._fetch_next_batch(cur, fetcher, runner) + # asynchronously get the next batch + has_more = fetcher.fetch_next_batch(cur, runner) - # And insert the curent batch - for idx in range(0, len(places), batch): - t0 = time.time() - worker = pool.next_free_worker() - pool_wait += time.time() - t0 - part = places[idx:idx+batch] - LOG.debug("Processing places: %s", str(part)) - runner.index_places(worker, part) - progress.add(len(part)) + # And insert the current 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)) - places = next_places - - pool.finish_all() - - fetcher.wait() - fetcher.close() + LOG.info("Wait time: fetcher: %.2fs, pool: %.2fs", + fetcher.wait_time, pool.wait_time) conn.commit() progress.done() - LOG.warning("Wait time: fetcher: {}s, pool: {}s".format(fetcher_wait, pool_wait)) - - - def _fetch_next_batch(self, cur, fetcher, runner): - ids = cur.fetchmany(100) - - if not ids: - return None - - if not hasattr(runner, 'get_place_details'): - return ids - - runner.get_place_details(fetcher, ids) - return []