models: Add is_private flag to UserMessage and add index for it.

The is_private flag is intended to be set if recipient type is
'private'(1) or 'huddle'(3), otherwise i.e if it is 'stream'(2), it
should be unset.

This commit adds a database index for the is_private flag (which we'll
need to use it). That index is used to reset the flag if it was
already set. The already set flags were due to a previous removal of
is_me_message flag for which the values were not cleared out.

For now, the is_private flag is always 0 since the really hard part of
this migration is clearing the unspecified previous state; future
commits will fully implement it actually doing something.

History: Migration rewritten significantly by tabbott to ensure it
runs in only 3 minutes on chat.zulip.org.  A key detail in making that
work was to ensure that we use the new index for the queries to find
rows to update (which currently requires the `order_by` and `limit`
clauses).

docs/overview/changelog.md

@@ -15,6 +15,17 @@ in bursts.
   access full history, even before they joined the stream.
 - Added support for announcement-only streams.
 
+**Upgrade notes:**
+
+* Zulip 1.9 contains a significant database migration that can take
+  several minutes to run.  The upgrade process automatically minimizes
+  disruption by running this migration first, before beginning the
+  user-facing downtime.  However, if you'd like to watch the downtime
+  phase of the upgrade closely, we recommend
+  [running them first manually](../production/expensive-migrations.html)
+  and as well as the usual trick of
+  [doing an apt upgrade first](../production/maintain-secure-upgrade.html#applying-system-updates).
+
 **Full feature changelog:**
 - Added an organization setting for message deletion time limits.
 - Added an organization setting to control who can edit topics.

docs/production/expensive-migrations.md

@@ -4,8 +4,8 @@
 
 # Running expensive migrations early
 
-Zulip 1.7 contains some significant database migrations that can take
+Zulip 1.7 and 1.9 each contain some significant database migrations
-several minutes to run.
+that can take several minutes to run.
 
 The upgrade process automatically minimizes disruption by running
 these first, before beginning the user-facing downtime.  However, if
@@ -19,6 +19,14 @@ can run them manually before starting the upgrade:
   Postgres database.
 3. In the postgres shell, run the following commands:
 
+        CREATE INDEX CONCURRENTLY
+        zerver_usermessage_is_private_message_id
+        ON zerver_usermessage (user_profile_id, message_id)
+        WHERE (flags & 2048) != 0;
+
+(This first migration, `zerver_usermessage_is_private_message_id`, is
+the only one new in Zulip 1.9).
+
         CREATE INDEX CONCURRENTLY
         zerver_usermessage_mentioned_message_id
         ON zerver_usermessage (user_profile_id, message_id)
@@ -44,13 +52,13 @@ can run them manually before starting the upgrade:
         ON zerver_usermessage (user_profile_id, message_id)
         WHERE (flags & 1) = 0;
 
-4. These will take some time to run, during which the server will
+These will take some time to run, during which the server will
-  continue to serve user traffic as usual with no disruption.  Once
+continue to serve user traffic as usual with no disruption.  Once they
-  they finish, you can proceed with installing Zulip 1.7.
+finish, you can proceed with installing Zulip 1.7.
 
 To help you estimate how long these will take on your server: count
 the number of UserMessage rows, with `select COUNT(*) from zerver_usermessage;`
 at the `./manage.py dbshell` prompt.  At the time these migrations
-were run on chat.zulip.org, it had 75M UserMessage rows; the first 4
+were run on chat.zulip.org, it had 75M UserMessage rows; the first 5
 indexes took about 1 minute each to create, and the final,
 "unread_message" index took more like 10 minutes.

scripts/lib/upgrade-zulip-stage-2

@@ -98,6 +98,7 @@ usermessage_index_migrations = [
     "[ ] 0095_index_unread_user_messages",
     "[ ] 0098_index_has_alert_word_user_messages",
     "[ ] 0099_index_wildcard_mentioned_user_messages",
+    "[ ] 0177_user_message_add_and_index_is_private_flag",
 ]
 # Our next optimization is to check whether any migrations are needed
 # before we start the critical section of the restart.  This saves

zerver/management/commands/create_large_indexes.py

@@ -88,6 +88,14 @@ def create_indexes() -> None:
         where_clause='WHERE (flags & 8) != 0 OR (flags & 16) != 0',
     )
 
+    # copied from 0177
+    create_index_if_not_exist(
+        index_name='zerver_usermessage_is_private_message_id',
+        table_name='zerver_usermessage',
+        column_string='user_profile_id, message_id',
+        where_clause='WHERE (flags & 2048) != 0',
+    )
+
 class Command(ZulipBaseCommand):
     help = """Create concurrent indexes for large tables."""
 

zerver/migrations/0177_user_message_add_and_index_is_private_flag.py

@@ -0,0 +1,78 @@
+# -*- coding: utf-8 -*-
+# Generated by Django 1.11.13 on 2018-06-14 13:39
+from __future__ import unicode_literals
+
+import sys
+import bitfield.models
+from django.db import migrations, transaction
+from zerver.lib.migrate import create_index_if_not_exist  # nolint
+from django.db.backends.postgresql_psycopg2.schema import DatabaseSchemaEditor
+from django.db.migrations.state import StateApps
+from django.db.models import F
+
+def reset_is_private_flag(
+        apps: StateApps, schema_editor: DatabaseSchemaEditor) -> None:
+    UserMessage = apps.get_model("zerver", "UserMessage")
+    UserProfile = apps.get_model("zerver", "UserProfile")
+    user_profile_ids = UserProfile.objects.all().order_by("id").values_list("id", flat=True)
+    # We only need to do this because previous migration
+    # zerver/migrations/0100_usermessage_remove_is_me_message.py
+    # didn't clean the field after removing it.
+
+    i = 0
+    total = len(user_profile_ids)
+    print("Setting default values for the new flag...")
+    sys.stdout.flush()
+    for user_id in user_profile_ids:
+        while True:
+            # Ideally, we'd just do a single database query per user.
+            # Unfortunately, Django doesn't use the fancy new index on
+            # is_private that we just generated if we do that,
+            # resulting in a very slow migration that could take hours
+            # on a large server.  We address this issue by doing a bit
+            # of hackery to generate the SQL just right (with an
+            # `ORDER BY` clause that forces using the new index).
+            flag_set_objects = UserMessage.objects.filter(user_profile__id = user_id).extra(
+                where=["flags & 2048 != 0"]).order_by("message_id")[0:1000]
+            user_message_ids = flag_set_objects.values_list("id", flat=True)
+            count = UserMessage.objects.filter(id__in=user_message_ids).update(
+                flags=F('flags').bitand(~UserMessage.flags.is_private))
+            if count < 1000:
+                break
+
+        i += 1
+        if (i % 50 == 0 or i == total):
+            percent = round((i / total) * 100, 2)
+            print("Processed %s/%s %s%%" % (i, total, percent))
+            sys.stdout.flush()
+
+class Migration(migrations.Migration):
+    atomic = False
+
+    dependencies = [
+        ('zerver', '0176_remove_subscription_notifications'),
+    ]
+
+    operations = [
+        migrations.AlterField(
+            model_name='archivedusermessage',
+            name='flags',
+            field=bitfield.models.BitField(['read', 'starred', 'collapsed', 'mentioned', 'wildcard_mentioned', 'summarize_in_home', 'summarize_in_stream', 'force_expand', 'force_collapse', 'has_alert_word', 'historical', 'is_private'], default=0),
+        ),
+        migrations.AlterField(
+            model_name='usermessage',
+            name='flags',
+            field=bitfield.models.BitField(['read', 'starred', 'collapsed', 'mentioned', 'wildcard_mentioned', 'summarize_in_home', 'summarize_in_stream', 'force_expand', 'force_collapse', 'has_alert_word', 'historical', 'is_private'], default=0),
+        ),
+        migrations.RunSQL(
+            create_index_if_not_exist(
+                index_name='zerver_usermessage_is_private_message_id',
+                table_name='zerver_usermessage',
+                column_string='user_profile_id, message_id',
+                where_clause='WHERE (flags & 2048) != 0',
+            ),
+            reverse_sql='DROP INDEX zerver_usermessage_is_private_message_id;'
+        ),
+        migrations.RunPython(reset_is_private_flag,
+                             reverse_code=migrations.RunPython.noop),
+    ]

zerver/models.py

@@ -1523,13 +1523,9 @@ class Reaction(models.Model):
 # though each row is only 4 integers.
 class AbstractUserMessage(models.Model):
     user_profile = models.ForeignKey(UserProfile, on_delete=CASCADE)  # type: UserProfile
-    # WARNING: We removed the previously-final flag,
-    # is_me_message, without clearing any values it might have had in
-    # the database.  So when we next add a flag, you need to do a
-    # migration to set it to 0 first
     ALL_FLAGS = ['read', 'starred', 'collapsed', 'mentioned', 'wildcard_mentioned',
                  'summarize_in_home', 'summarize_in_stream', 'force_expand', 'force_collapse',
-                 'has_alert_word', "historical"]
+                 'has_alert_word', "historical", "is_private"]
     flags = BitField(flags=ALL_FLAGS, default=0)  # type: BitHandler
 
     class Meta: