X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/602728895e7c35ed1f434acb8af6acfaa66033f3..96e6bbe3a1f2e2d41064e4ca19c0e22de42896c1:/nominatim/indexer/indexer.py

diff --git a/nominatim/indexer/indexer.py b/nominatim/indexer/indexer.py
index 41535af8..b7673aba 100644
--- a/nominatim/indexer/indexer.py
+++ b/nominatim/indexer/indexer.py
@@ -3,6 +3,7 @@ Main work horse for indexing (computing addresses) the database.
 """
 import logging
 import select
+import time
 
 import psycopg2.extras
 
@@ -13,6 +14,73 @@ from nominatim.db.connection import connect
 
 LOG = logging.getLogger()
 
+
+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 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.
+
+            Returns true if there is still data available.
+        """
+        ids = cur.fetchmany(100)
+
+        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
+
+        return True
+
+    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()
+
+        return self.current_ids
+
+    def __enter__(self):
+        return self
+
+
+    def __exit__(self, exc_type, exc_value, traceback):
+        self.conn.wait()
+        self.close()
+
 class WorkerPool:
     """ A pool of asynchronous database connections.
 
@@ -23,6 +91,7 @@ class WorkerPool:
     def __init__(self, dsn, pool_size):
         self.threads = [DBConnection(dsn) for _ in range(pool_size)]
         self.free_workers = self._yield_free_worker()
+        self.wait_time = 0
 
 
     def finish_all(self):
@@ -66,7 +135,9 @@ class WorkerPool:
                 ready = self.threads
                 command_stat = 0
             else:
+                tstart = time.time()
                 _, ready, _ = select.select([], self.threads, [])
+                self.wait_time += time.time() - tstart
 
 
     def __enter__(self):
@@ -74,6 +145,7 @@ class WorkerPool:
 
 
     def __exit__(self, exc_type, exc_value, traceback):
+        self.finish_all()
         self.close()
 
 
@@ -183,9 +255,6 @@ class Indexer:
                 total_tuples = cur.scalar(runner.sql_count_objects())
                 LOG.debug("Total number of rows: %i", total_tuples)
 
-                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)
@@ -194,49 +263,25 @@ class Indexer:
                 with conn.cursor(name='places') as cur:
                     cur.execute(runner.sql_get_objects())
 
-                    fetcher = DBConnection(self.dsn)
-                    psycopg2.extras.register_hstore(fetcher.conn,
-                                                    oid=hstore_oid,
-                                                    array_oid=hstore_array_oid)
+                    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()
 
-                    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:
-                                fetcher.wait()
-                                places = fetcher.cursor.fetchall()
+                                # asynchronously get the next batch
+                                has_more = fetcher.fetch_next_batch(cur, runner)
 
-                            # asynchronously get the next batch
-                            next_places = self._fetch_next_batch(cur, fetcher, runner)
+                                # 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))
 
-                            # And insert the curent batch
-                            for idx in range(0, len(places), batch):
-                                worker = pool.next_free_worker()
-                                part = places[idx:idx+batch]
-                                LOG.debug("Processing places: %s", str(part))
-                                runner.index_places(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()
-
-
-    def _fetch_next_batch(self, cur, fetcher, runner):
-        ids = cur.fetchmany(1000)
-
-        if not ids:
-            return None
-
-        if not hasattr(runner, 'get_place_details'):
-            return ids
-
-        runner.get_place_details(fetcher, ids)
-        return []