Request: undo_parent

[Faithfinder]

We're still experimenting with various partition configurations, and sometimes I decide that the current version is not something I want. In that case I just want to drop it, but I don't want garbage to stay in partman config. Can we have an undo_parent that will just remove everything about the table from partman configuration? Template and config entreis, and whatever

[keithf4]

Interesting idea. Maybe something of a more general config clean up function that by default removes the config but with parameters to do other things optionally (remove the template table, cleanup sub-partition config, etc). I'll see what I can come up with or you're free to work on a PR as well.

[keithf4]

Also, what does the undo_partition_proc() do or not do that you're trying to do differently here?

[Faithfinder]

Also, what does the undo_partition_proc() do or not do that you're trying to do differently here?

Does undo_partition, proc or not, do anything to the config? As I've said, the goal is to just clean up intermediate experimental versions of tables. I can do it as easily as DROP TABLE parent from the database's perspective, but this will obviously leave all the pg_partman config objects.

Cleaning up pg_partman manually is also quite possible, but that would require me to understand precisely what gets created when create_parent is run, which is not something I want to keep in my brain

[keithf4]

Yes, the undo_partition functions/procedure will clean up the configs and template tables as well. undo_partition_proc() just calls undo_partition in commit loops to avoid long transactions. All that's left is the target table with all the data.

I guess another option could be something to undo_partition that does not keep the data. So it just leaves an empty parent and cleans up all the config data and objects.

[Faithfinder]

Yeah, docs don't make it sound like that's what undo_partition does. In hindsight this line says it:

Once all child tables have been uninherited/dropped, the configuration data is removed from pg_partman automatically.

but I thought it just means that it reverts

When this function is run, the undo_in_progress column in the configuration table is set to true. This causes all partition creation and retention management to stop.

But yeah, in the case I describe there's no need to keep anything

[Faithfinder]

Also, naming: undo_partition just doesn't sounds like it's a reverse to create_parent

[keithf4]

The undo_in_progress flag is there to avoid issues of maintenance running while an undo session is running, which could be highly likely when undoing a very large partition set.

Good point on the name. It's been like that for a long time now, tho. Maybe I'll reconsider a rename during the next major version update.

[nusratulhassan]

I was performing some tests in my local and wanted to try partman.create_parent with new parameters but couldn't find something like drop_parent. I had to delete it manually using
Delete from partman.part_config where parent_table= 'public.my_parent_table_name';