Hi all, This is a call for comments on what I hope will be the last two modifications to multi-db support for 1.2.
Barring negative feedback, I aim to commit this at the start of next week, just in time for the beta-1 feature freeze. Change 1: Identifying which models are on which database ------------------------------ To date, we have avoided any housekeeping over which models are available on which database. We have made the assumption that all models are available everywhere, and left it up to the user to manually configure their database if this isn't the case. This has a couple of consequences: * If you want to restrict some models to certain databases, you can't use syncdb - you need to take the manual route. This isn't a show-stopper, but it is a lot more complication than Django usually requires. * It is impossible to do completely generic metaprogramming - you can't, for example, build an admin interface that only displays the right models for the right databases, because there isn't an easy way to determine if a model is available on a given database. Again, manual configuration is possible here, but similarly complex. * The test system must assume that every model is on every database (since there is no opportunity for manual synchronization). This means you can't explicitly test behavior that relies on non-availability of a model on a particular database. I've opened #12672 to track the issue, and there is a candidate patch ready for comment. This patch follows an idea originally from Johannes Dollinger, and refined by Yuri Baburov - that we add a method to the database routers that can answer whether a specific model is available on a specific database. This method (allow_syncdb()) follows the same pattern as allow_relation(); return True if a model should be on a database, False if it shouldn't, and None if the router has no opinion. The bucket behavior is to allow all models on all databases, so there is no change to the current behavior if you don't have routers installed. The other benefit of this approach is that we can drop the --exclude arguments from loaddata and syncdb. Personally, I was never happy with the fact that we needed to introduce these arguments as part of the first multi-db patch, but they were a necessary evil caused by the lack of info about model database assignments. Since the need for these flags is going away, I've taken the opportunity to remove them. Code, tests and documentation is in the most recent (v2) patch on ticket #12672. Comments? Change 2: Test mirrors ------------------------------ The second change involves a change to the test system to handle multiple database setups. I have introduced a new test setting -- 'TEST_MIRROR' -- that redirects a particular database alias to point at a different connection while in the test environment. This is required for testing of master/slave setups. The short explanation is that the temporary databases that the test environment creates won't be created with any data replication strategy in place, which will break any tests that expect replication to occur. A full explanation, including an example is in the documentation portion of the patch, attached to ticket #12542. Comments? Yours Russ Magee %-) -- You received this message because you are subscribed to the Google Groups "Django developers" group. To post to this group, send email to django-develop...@googlegroups.com. To unsubscribe from this group, send email to django-developers+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/django-developers?hl=en.
