migrations: Batch fix_has_link_attribute migration.

This avoids risk of OOM issues on servers with relatively limited RAM
and millions of messages of history; apparently, fetching all messages
ordered by ID could be quite memory-intensive even with an iterator
usage model.

Fortunately, we have other migrations that already follow this pattern
of iterating over messages, so it's easy to borrow existing code to
make this migration run reasonably.

zerver/migrations/0257_fix_has_link_attribute.py

@@ -7,13 +7,12 @@ from django.db.backends.postgresql_psycopg2.schema import DatabaseSchemaEditor
 from django.db.migrations.state import StateApps
 
 import lxml
+import time
+BATCH_SIZE = 1000
 
-def fix_has_link(apps: StateApps, schema_editor: DatabaseSchemaEditor) -> None:
+def process_batch(apps: StateApps, id_start: int, id_end: int, last_id: int) -> None:
     Message = apps.get_model('zerver', 'Message')
-    total_messages = Message.objects.count()
-    print(" Processing messages...")
-
-    for message in Message.objects.order_by("id"):
+    for message in Message.objects.filter(id__gte=id_start, id__lte=id_end).order_by("id"):
         if message.rendered_content == "":
             # There have been bugs in the past that made it possible
             # for a message to have "" as its rendered_content; we
@@ -24,7 +23,7 @@ def fix_has_link(apps: StateApps, schema_editor: DatabaseSchemaEditor) -> None:
             continue
 
         if message.id % 1000 == 0:
-            print("Processed %s / %s" % (message.id, total_messages))
+            print("Processed %s / %s" % (message.id, last_id))
 
         # Because we maintain the Attachment table, this should be as
         # simple as just just checking if there's any Attachment
@@ -56,7 +55,33 @@ def fix_has_link(apps: StateApps, schema_editor: DatabaseSchemaEditor) -> None:
         message.has_attachment = has_attachment
         message.save(update_fields=['has_link', 'has_attachment', 'has_image'])
 
+def fix_has_link(apps: StateApps, schema_editor: DatabaseSchemaEditor) -> None:
+    Message = apps.get_model('zerver', 'Message')
+    if not Message.objects.exists():
+        # Nothing to do, and Message.objects.latest() will crash.
+        return
+
+    # This migration logic assumes that either the server is not
+    # running, or that it's being run after the logic to correct how
+    # `has_link` and friends are set for new messages have been
+    # deployed.
+    last_id = Message.objects.latest("id").id
+
+    id_range_lower_bound = 0
+    id_range_upper_bound = 0 + BATCH_SIZE
+    while id_range_upper_bound <= last_id:
+        process_batch(apps, id_range_lower_bound, id_range_upper_bound, last_id)
+
+        id_range_lower_bound = id_range_upper_bound + 1
+        id_range_upper_bound = id_range_lower_bound + BATCH_SIZE
+        time.sleep(0.1)
+
+    if last_id > id_range_lower_bound:
+        # Copy for the last batch.
+        process_batch(apps, id_range_lower_bound, last_id, last_id)
+
 class Migration(migrations.Migration):
+    atomic = False
 
     dependencies = [
         ('zerver', '0256_userprofile_stream_set_recipient_column_values'),