Blocking characteristics

[agersant]

Hello! Would it be possible to get some general idea of the expected performance characteristics of transaction operations?

More precisely, I would like to know:

• In what circumstances do transaction operations (get, insert, remove, commit, etc.) end up performing disk I/O on the thread they were called?
• Is the above answer any different if the amount of managed data is below the cache_size?
• How frequently is data persisted to disk, and what guarantees are there in case of unexpected interruptions (eg. power loss)?
• I see there is some tokio support in Cargo.toml but it's not clear to me what enabling it would do.

[vincent-herlemont]

Native DB entirely delegates the management of storage to the redb backend (database) and it's necessary to refer to this documentation for details redb/design. Native DB can be seen as an overlay to redb. However, here are some brief responses:

In what circumstances do transaction operations (get, insert, remove, commit, etc.) end up performing disk I/O on the thread they were called?

Transaction operations perform disk I/O when accessing data not in cache or when writing changes to disk during a commit.

Is the above answer any different if the amount of managed data is below the cache_size?

Yes, if all data fits within the cache, read operations may avoid disk I/O, but writes still require disk I/O during commits to persist changes.

How frequently is data persisted to disk, and what guarantees are there in case of unexpected interruptions (eg. power loss)?

Data is persisted to disk during commit operations, and redb ensures data integrity during unexpected interruptions by using checksums, atomic writes, and recovery mechanisms.

---

I see there is some tokio support in Cargo.toml but it's not clear to me what enabling it would do.

For the moment, tokio for Native DB has nothing to do with storage but is simply the support for channels used to listen to database events useful for real-time applications (see watch).

[agersant]

Thank you very much!