Refs #21906 -- Changed dumpdata --database default to router.db_for_read.

[hramezani]

Ticket 21906.

[felixxm]

@hramezani Thanks for this patch 👍 IMO we should change loaddata in the same commit, because changing only dumpdata will make this scenario even more error prone.

[felixxm]

Update help.

[felixxm]

I would change

index df9eecc0f8..953f261ef4 100644
--- a/django/core/management/commands/dumpdata.py
+++ b/django/core/management/commands/dumpdata.py
@@ -67,7 +67,7 @@ class Command(BaseCommand):
     def handle(self, *app_labels, **options):
         format = options['format']
         indent = options['indent']
-        using = options['database']
+        database = options['database']
         excludes = options['exclude']
         output = options['output']
         show_traceback = options['traceback']

to simplify this.

We should also skip models with different db for read, instead of checking router.allow_migrate_model(), e.g.

using = router.db_for_read(model)
if not model._meta.proxy and (not database or using == database):
    if use_base_manager:
        ...

[felixxm]

This should go to the Backwards incompatible changes.

[hramezani]

@felixxm Thanks for the review.

@hramezani Thanks for this patch IMO we should change loaddata in the same commit, because changing only dumpdata will make this scenario even more error prone.

I take a look at loaddata and I want to say it's hard to change it because:

• In dumpdata we loop over models and fetch objects, so we can specify the database to read model objects from. but in loaddata we load data file by file and it each file we can have objects from multiple models.
• In loaddata we create an atomic transaction on a database and load data to it. if we have a file contains object from several models, we should create multiple atomic transactions and it is hard to handle atomicity over multiple databases.
• The deserialize depend on one DB, so we need to change them as well to fetch DB for each model from router

In general, I would say the loaddata implementation highly depends on just one database.

I am eager to finish it, so If you have any idea it would be appreciated.

Thanks

[felixxm]

@hramezani IMO these changes cannot be implemented separately, because we will create an option to dump data from multiple databases but they will be useless for loaddata. It seems that we have two options:

• we can raise an error in dumpdata for multiple "read" databases, or
• we can try to handle this in loaddata with multiple transactions.

I will try to ask for opinions/ideas on the mailing list.

[felixxm]

@hramezani "I will try to ask ..." -> "I would try to ask ..." 😄