From 0f73f5bc5ed33c0c71a1cc2761247f265bb5f6e2 Mon Sep 17 00:00:00 2001
From: Raphael Michel <michel@rami.io>
Date: Tue, 11 May 2021 16:50:53 +0200
Subject: [PATCH] Optimize performance of voucher code generation on PostgreSQL

---
 src/pretix/control/views/vouchers.py | 44 ++++++++++++++++++++++++++--
 1 file changed, 41 insertions(+), 3 deletions(-)

diff --git a/src/pretix/control/views/vouchers.py b/src/pretix/control/views/vouchers.py
index 9ee1a81f5a..9605c1b919 100644
--- a/src/pretix/control/views/vouchers.py
+++ b/src/pretix/control/views/vouchers.py
@@ -55,7 +55,7 @@ from django.views.generic import (
     CreateView, DeleteView, ListView, TemplateView, UpdateView, View,
 )
 
-from pretix.base.models import CartPosition, LogEntry, OrderPosition, Voucher
+from pretix.base.models import CartPosition, LogEntry, OrderPosition, Voucher, Event
 from pretix.base.models.vouchers import _generate_random_code
 from pretix.base.services.locking import NoLockManager
 from pretix.base.services.vouchers import vouchers_send
@@ -463,11 +463,49 @@ class VoucherRNG(EventPermissionRequiredMixin, View):
             return HttpResponseBadRequest()
 
         prefix = request.GET.get('prefix')
+        batch_size = 500
+        if 'postgres' in settings.DATABASES['default']['ENGINE']:
+            batch_size = 5_000
+
+        """
+        We're trying to check if any of the requested codes already exists in the database. Generally, this is a
+
+            SELECT code FROM voucher WHERE code IN (…)
+
+        query. However, it turns out that this query get's rather slow if an organizer has lots of vouchers, even
+        with a organizer with just over 50_000 vouchers, we've seen that creating 20_000 new voucher codes took
+        just over 30 seconds. There's another way of doing this query on PostgreSQL, which is joining with a
+        temporary table
+
+            SELECT code FROM voucher INNER JOIN (VALUES …) vals(v) ON (code = v)
+
+        This is significantly faster, inserting 20_000 vouchers now takes 2-3s instead of 31s on the same dataset.
+        It's still slow, and removing the JOIN to the event table doesn't significantly speed it up. We might need
+        an entirely different approach at some point.
+        """
+
         while len(codes) < num:
             new_codes = set()
-            for i in range(min(num - len(codes), 500)):  # Work around SQLite's SQLITE_MAX_VARIABLE_NUMBER
+            for i in range(min(num - len(codes), batch_size)):  # Work around SQLite's SQLITE_MAX_VARIABLE_NUMBER
                 new_codes.add(_generate_random_code(prefix=prefix))
-            new_codes -= set([v['code'] for v in Voucher.objects.filter(code__in=new_codes).values('code')])
+
+            if 'postgres' in settings.DATABASES['default']['ENGINE']:
+                with connection.cursor() as cursor:
+                    args = list(new_codes) + [request.organizer.pk]
+                    tmptable = "VALUES " + (", ".join(['(%s)'] * len(new_codes)))
+                    cursor.execute(
+                        f'SELECT code '
+                        f'FROM "{Voucher._meta.db_table}" '
+                        f'INNER JOIN ({tmptable}) vals(v) ON ("{Voucher._meta.db_table}"."code" = "v")'
+                        f'INNER JOIN "{Event._meta.db_table}" ON ("{Voucher._meta.db_table}"."event_id" = "{Event._meta.db_table}"."id") '
+                        f'WHERE "{Event._meta.db_table}"."organizer_id" = %s',
+                        args
+                    )
+                    for row in cursor.fetchall():
+                        new_codes.remove(row[0])
+            else:
+                new_codes -= set([v['code'] for v in Voucher.objects.filter(code__in=new_codes).values('code')])
+
             codes |= new_codes
 
         return JsonResponse({