[Enhancement]: Attach space dimension to tablespace

[glep207]

What type of enhancement is this?

Other

What subsystems and features will be improved?

Partitioning

What does the enhancement do?

A single hypertable receives different data with very different volumes and retention requirements. The application writing to Timescale does not allow spreading the data across several hypertables (which would be another solution). Disk space is limited.
I would like to define space dimensions with a different retention period, with the fast-growing data (with a shorter retention) being assigned to a specific tablespace, which will be located on tmpfs to relieve the main disk. The risk of data loss by placing the tablespace on tmpfs is accepted - only a short time span would be affected. The data would be aggregated with a continuous aggregate before the chunks are dropped.

The problem is that the current partitioning scheme distributes chunks to the available tablespaces randomly - chunks from a given partition are allocated preferentially to the tablespace, but one cannot specify that tablespace.

The proposal is to add an argument "tablespace" to add_dimension(), so that all chunks for that dimension will be placed on the specified tablespace.

The feature is probably relevant for single-node deployments only.

Related issues (links to follow):

• drop_chunks() on one space dimension only Drop chunks of specific a space partition #545
• add retention policy per space dimension Use space dimension as a retention setting. #563

Implementation challenges

No response