Django backup strategy with dumpdata and migrations

问题

As in this question, I set up a dumpdata-based backup system for my database. The setup is akin to running a cron script that calls dumpdata and moves the backup to a remote server, with the aim of simply using loaddata to recover the database. However, I'm not sure this plays well with migrations. loaddata now has an ignorenonexistent switch to deal with deleted models/fields, but it is not able to resolve cases where columns were added with one-off defaults or apply RunPython code.

The way I see it, there are two sub-problems to address:

Tag each dumpdata output file with the current version of each app
Splice the fixtures into the migration path

I'm stumped about how to tackle the first problem without introducing a ton of overhead. Would it be enough to save an extra file per backup that contained an {app_name: migration_number} mapping?

The second problem I think is easier once the first one is solved, since the process is roughly:

Create a new database
Run migrations forward to the appropriate point for each app
Call loaddata with the given fixture file
Run the rest of the migrations

There's some code in this question (linked from the bug report) that I think could be adapted for this purpose.

Since these are fairly regular/large snapshots of the database, I don't want to keep them as data migrations cluttering up the migrations directory.

回答1:

I am taking the following steps to backup, restore or transfer my postgresql database between any instance of my project:

The idea is to keep the least possible migrations as if manage.py makemigrations was run for the first time on an empty database.

Let's assume that we have a working database to our development environment. This database is a current copy of the production database that should not be open to any changes. We have added models, altered attributes etc and those actions have generated additional migrations.

Now the database is ready to be migrated to production which -as stated before- is not open to public so it is not altered in any way. In order to achieve this:

I perform the normal procedure in the development environment.
I copy the project to the production environment.
I perform the normal procedure in the production environment

We make the changes in our development environment. No changes should happen in the production database because they will be overridden.

Normal Procedure

Before anything else, I have a backup of the project directory (which includes a requirements.txt file), a backup of the database and -of course- git is a friend of mine.

I take a dumpdata backup in case I need it. However, dumpdata has some serious limitations regarding content types, permissions or other cases where a natural foreignkey should be used:
```
./manage.py dumpdata --exclude auth.permission --exclude contenttypes  --exclude admin.LogEntry --exclude sessions --indent 2 > db.json
```

I take a pg_dump backup to use:

pg_dump -U $user -Fc $database --exclude-table=django_migrations > path/to/backup-dir/db.dump

Only if I want to merge existing migrations in one, I delete all migrations from every application.

In my case the migrations folder is a symlink, so I use the following script:
```
#!/bin/bash
for dir in $(find -L -name "migrations")
do
  rm -Rf $dir/*
done
```

I delete and recreate the database:

For example, a bash script can include the following commands:

su -l postgres -c "PGPASSWORD=$password psql -c 'drop database $database ;'"
su -l postgres -c "createdb --owner $username $database"
su -l postgres -c "PGPASSWORD=$password psql $database -U $username -c 'CREATE EXTENSION $extension ;'"

I restore the database from the dump:

pg_restore -Fc -U $username -d $database path/to/backup-dir/db.dump

If migrations were deleted in step 3, I recreate them in the following way:

./manage.py makemigrations <app1> <app2> ... <appn>

... by using the following script:

#!/bin/bash
apps=()
for app in $(find ./ -maxdepth 1 -type d ! -path "./<project-folder> ! -path "./.*" ! -path "./")
do
  apps+=(${app#??})
done
all_apps=$(printf "%s "  "${apps[@]}")

./manage.py makemigrations $all_apps

I migrate using a fake migration:
```
./manage.py migrate --fake
```

In case something has gone completely wrong and everything is ***, (this can happen, indeed), I can use the backup to revert everything to its previous working state. If I would like to use the db.json file from step one, it goes like this: