Opened 14 years ago

Last modified 9 years ago

#13528 new Bug

db_table truncation is applied based on the properties of the default database

Reported by: Russell Keith-Magee Owned by:
Component: Database layer (models, ORM) Version: dev
Severity: Normal Keywords:
Cc: loic84 Triage Stage: Accepted
Has patch: yes Needs documentation: no
Needs tests: no Patch needs improvement: yes
Easy pickings: no UI/UX: no


Reported by Peter Long in django-dev:

In django/db/models/ (line 54), the contribute_to_class method uses the default database connection to get an ops module that is used to truncate the db_table name. The options class isn't in a position to do this truncation, and there's no guarantee that the default database has an appropriate truncation module.

Attachments (1)

13528-1.diff (987 bytes ) - added by Claude Paroz 11 years ago.
Possible fix

Download all attachments as: .zip

Change History (23)

comment:1 by Russell Keith-Magee, 14 years ago

Triage Stage: UnreviewedAccepted

comment:2 by Russell Keith-Magee, 14 years ago

milestone: 1.21.3

This isn't a trivial thing to fix, and it only affects users that have:

  1. multiple databases
  2. from different vendors
  3. where the default database has a longer maximum allowed length than the other vendors database.
  4. and there are tables whose table names exceed the shorter max_length
  5. and the db_table can't be renamed.

For example, if you have a PostgreSQL default database (max_length=64), and an Oracle secondary database (max_length=30), and you can't keep your table names to under 30 characters, you will be affected by this bug. This is a sufficiently edge case problem that I'm going to bump from 1.2.

comment:3 by Russell Keith-Magee, 14 years ago

Anssi Kaariainen reports that the same problem exists with the validate command.

comment:4 by Alex Gaynor, 14 years ago

FWIW solving this for the validate command will be a lot easier. Just loop through each DB for each model, check if the model should be synced (with the router) and go from there.

comment:5 by Andrew Godwin, 14 years ago

Owner: changed from nobody to Andrew Godwin
Status: newassigned

I'm working on this - my proposed solution is thus:

  • Deprecate Options.db_table (PendingDeprecationWarning for 1.3)
  • Add Options.get_db_table(using=db_alias)
  • Change all the places Django uses db_table to use get_db_table

This should both solve the problem, and allow third-party apps to continue using db_table (as it's still valid if you've only got one database).

comment:6 by Russell Keith-Magee, 14 years ago

See #13711 for an analogous problem with field.colname.

comment:7 by Andrew Godwin, 14 years ago

Update on this ticket: There's several places where the SQL is (imho incorrectly) part of the Query and other classes, so I'm waiting for the second queryset refactor work to get along to the point where the patch can be on top of that, otherwise there's going to be massive conflicts.

comment:8 by Julien Phalip, 13 years ago

Severity: Normal
Type: Bug

comment:9 by Jacob, 13 years ago

milestone: 1.3

Milestone 1.3 deleted

comment:11 by Aymeric Augustin, 12 years ago

UI/UX: unset

Change UI/UX from NULL to False.

comment:12 by Aymeric Augustin, 12 years ago

Easy pickings: unset

Change Easy pickings from NULL to False.

comment:13 by Anssi Kääriäinen, 12 years ago

Ive been thinking that model._meta should contain backend specific sections. This should contain at least quoted db_table name and column names. In addition to solving this problem such an arrangement should speed up SQL generation. Quoting the table and column names can consume a significant amount of time.

So, instead of doing model._meta.db_table you would do something like model._meta[conn].db_table. However, this requires somewhat extensive changes in the sql/ and sql/, so I guess this has to wait...

by Claude Paroz, 11 years ago

Attachment: 13528-1.diff added

Possible fix

comment:14 by Claude Paroz, 11 years ago

Has patch: set
Needs tests: set

Using router might be a solution, however I don't know if this does negatively affect performance or not.

comment:15 by ANUBHAV JOSHI, 10 years ago

claudep: In the diff you added:

db = router.db_for_read(self.model)
Option object doesn't have an attribute model

comment:16 by ANUBHAV JOSHI, 10 years ago

claudep: Making self.model to self.concrete_model works fine.

Last edited 10 years ago by ANUBHAV JOSHI (previous) (diff)

comment:17 by Andrew Godwin, 10 years ago

Owner: Andrew Godwin removed
Status: assignednew

comment:18 by ANUBHAV JOSHI, 10 years ago

Owner: set to ANUBHAV JOSHI
Status: newassigned

comment:19 by ANUBHAV JOSHI, 10 years ago

Version: 1.2-betamaster

As discussed with Florian on IRC, we should check for minimum length that all available dbs allow and truncate accordingly.

Last edited 10 years ago by ANUBHAV JOSHI (previous) (diff)

comment:20 by Tim Graham, 10 years ago

Needs tests: unset
Patch needs improvement: set

I don't think the approach of the latest PR is good. Let's try implementing what Andrew suggested in comment 5 of this ticket.

comment:21 by loic84, 10 years ago

Cc: loic84 added

in reply to:  20 comment:22 by ANUBHAV JOSHI, 10 years ago

Replying to timo:

I don't think the approach of the latest PR is good. Let's try implementing what Andrew suggested in comment 5 of this ticket.

I tried to implement the above, I think that it is infeasible, atleast at the moment for the following reason:

  • There are many places for eg. as in query creation where db_table is used but there is no knowledge of alias or connection in order to call get_db_table. Even if try to increase the function parameters it would be very hacky to continue all the way up the function calls until an alias is found and pass it down all the way where it is needed.
  • In the backends there is no alias used so we don't know using the connection to call this function will be correct or not.
  • There are many places where still connection is used instead of connections[alias] so that might continue to create similar multi-db problems as the above.

I discussed this issue a little with Loic and this is what he suggested that we can note down where and why it is possible to change the suggested behaviour and where all the alias is not used and should be so that we can move on from this and try to fix that bigger problem instead of this one and that might take considerable amount of time.

comment:23 by Tim Graham, 9 years ago

Owner: ANUBHAV JOSHI removed
Status: assignednew
Note: See TracTickets for help on using tickets.
Back to Top