Support parallel decompression of tar.zst archives during restore

[hedbruno]

When restoring backups compressed with zstd, the current restore process decompresses archives sequentially, which becomes very slow for large backups with multiple tablespaces and terabytes of data. I propose adding support for parallel decompression of tar.zst files during restore to significantly speed up recovery.

[martinmarques]

How do you expect to decompress in parallel a zstd tarball? There is no multi-thread support in zstd for decompression, only for compression.
We could use pzstd. But that would require some research to see how much gain we have adding support for another tool.
We are, in any case, looking into improving how this works, although the current work is focused on restoring with barman-cloud. It may or may not help in your case, if you are doing barman restores.

[hedbruno]

In cases where there is more than one tablespace, multiple tar.zst files are generated.

example:

/barman/serverdb/base/20260714T151236/data/
├── 16429.tar.zst
├── 16435.tar.zst
├── 16436.tar.zst
├── 16450.tar.zst
├── 26353097.tar.zst
├── backup_manifest
└── base.tar.zst

My proposal is that, in these cases, we should be able to decompress multiple .tar.zst files at the same time, two or more files in a single operation.

Thank you for your attention.

[martinmarques]

Ok, that is more sensible. We'll have a discussion about it