pg_partman
pg_partman copied to clipboard
pgpartman
Support for uuid7 or ulid
In addition to time and id range it would be nice to have support for uuid7 or ulid (Universally Unique Lexicographically Sortable Identifiers).
Thanks!
I'm assuming something like this would generally not need ongoing partition generation? As in, you'd define your child tables up front for each block of values and then be done? That's pretty well supported without needing something like pg_partman which primarily provides ongoing maintenance features.
Or is there some scenario where you see that being needed with this data type?
Also, is this data type supported in PG? Don't see it specifically listed here in the uuid data types.
https://www.postgresql.org/docs/15/uuid-ossp.htm
Both UUIDv7 and ULID encode a unix timestamp which allow partitioning by time (hours, day, etc...) so the idea is to be able to use time partitioning using a unique identifier. Also, both UUIDv7 and ULID use UUID format and therefore are compatible with the PostgreSQL uuid type (but there is no built-in function to generate a UUIDv7 or a ULID yet).
Is it possible to decode the time value out of that? Otherwise, how do you see defining the partitioning interval on this? Not quite that familiar with UUID usage myself.
Absolutely, see for the format: https://www.ietf.org/archive/id/draft-peabody-dispatch-new-uuid-format-01.html#section-4.4
In a nutshell because the first 36-bit of the 128-bit UUID is a big-endian unsigned Unix Timestamp value, I suppose we can easily define the partitions range by time.
Also found https://github.com/fboulnois/pg_uuidv7
Ok, I see how it might work now, but it's a pretty specialized request. I have a lot of other stuff I'd like to get done first (see open issues) before I'd really have time to consider looking into adding this myself.
However, after 5.x is out, if you'd like to work on a PR to support this before I would have time to look into this, I'd be happy to review it.
I came here to ask for a similar feature, but from a different angle.
UUID7/ULID is a good identifier that is time-based and hence grows monotonically. It’s very natural to use it not just as Primary Key, but also for partitioning based on timestamp derived from it.
One way to solve this would have been maintaining a generated/unstored column that would extract timestamp from the UUID, but it would bring in extra effort for the application to account for the generated column on the read path, not to mention changing the PK to include it.
I thought a generic solution to this would be a way for pgpartman to support another way of declaring partitions: not by a date-type column, but by a function, so that the function would take the PK(s) and return the applicable time range for the partitioning.
@rauanmayemir declaring functions by a function seems like a brilliant idea.
@linglom1 you may want to consider timescaledb that supports the described scheme out of the box.
I have a similar requirement where we use time ordered uids similar to twitter's snowflake ids. This practice is becoming pretty common in many places. I was also thinking along the lines of @rauanmayemir 's suggestion to externalize the logic by a function, except something lightweight in the form of "encoder" function that encodes a timestamp as per one's id specification. This way the time range generation logic itself can continue to be leveraged from the extension. @keithf4 I'm happy to work on a PR for this, if you can help review.
@keithf4 bumping this up, let me know if you're open for a PR along the above lines ^
@keithf4 have raised this - https://github.com/pgpartman/pg_partman/pull/683 I understand this is a big change, open to some scrutiny on this