Faster `pokedex load` for PostgreSQL #526

Also added the -S (--safe) option, which disables the backend-specific
optimizations.

This gives over 3× speedup on my machine :)

diff --git a/pokedex/__init__.py b/pokedex/__init__.py
index 45b4d7a..e4397af 100644 (file)
--- a/pokedex/__init__.py
+++ b/pokedex/__init__.py
@@ -122,6 +122,8 @@ def command_load(*args):
     parser = get_parser(verbose=True)
     parser.add_option('-d', '--directory', dest='directory', default=None)
     parser.add_option('-D', '--drop-tables', dest='drop_tables', default=False, action='store_true')
+    parser.add_option('-S', '--safe', dest='safe', default=False, action='store_true',
+        help="Do not use backend-specific optimalizations.")
     options, tables = parser.parse_args(list(args))
 
     if not options.engine_uri:
@@ -138,7 +140,7 @@ def command_load(*args):
                                   drop_tables=options.drop_tables,
                                   tables=tables,
                                   verbose=options.verbose,
-                                  safe=False)
+                                  safe=options.safe)
 
 def command_reindex(*args):
     parser = get_parser(verbose=True)

diff --git a/pokedex/db/load.py b/pokedex/db/load.py
index e6b3287..4512bf6 100644 (file)
--- a/pokedex/db/load.py
+++ b/pokedex/db/load.py
@@ -168,6 +168,36 @@ def load(session, tables=[], directory=None, drop_tables=False, verbose=False, s
         reader = csv.reader(csvfile, lineterminator='\n')
         column_names = [unicode(column) for column in reader.next()]
 
+        if not safe and session.connection().dialect.name == 'postgresql':
+            """
+            Postgres' CSV dialect is nearly the same as ours, except that it
+            treats completely empty values as NULL, and empty quoted
+            strings ("") as an empty strings.
+            Pokedex dump does not quote empty strings. So, both empty strings
+            and NULLs are read in as NULL.
+            For an empty string in a NOT NULL column, the load will fail, and
+            load will fall back to the cross-backend row-by-row loading. And in
+            nullable columns, we already load empty stings as NULL.
+            """
+            session.commit()
+            not_null_cols = [c for c in column_names if not table_obj.c[c].nullable]
+            if not_null_cols:
+                force_not_null = 'FORCE NOT NULL ' + ','.join('"%s"' % c for c in not_null_cols)
+            else:
+                force_not_null = ''
+            command = "COPY {table_name} ({columns}) FROM '{csvpath}' CSV HEADER {force_not_null}"
+            session.connection().execute(
+                    command.format(
+                            table_name=table_name,
+                            csvpath=csvpath,
+                            columns=','.join('"%s"' % c for c in column_names),
+                            force_not_null=force_not_null,
+                        )
+                )
+            session.commit()
+            print_done()
+            continue
+
         # Self-referential tables may contain rows with foreign keys of other
         # rows in the same table that do not yet exist.  Pull these out and add
         # them to the session last