Make use of prepared commands

[roji]

ADO.NET allows users to prepare commands, which allows databases to cache query plans and parses. IIRC in modern version of SqlServer this doesn't have much effect (any?), but in other providers this is an important feature - repeated complicated queries in PostgreSQL definitely benefit from prepared commands.

I know there's already an SQL caching mechanism inside EFCore, and although I'm sure it would be non-trivial to cache commands as well it would be a great feature.

Definitely a lower-priority backlog feature request.

[bricelam]

If we do this, we should definitely do aspnet/Microsoft.Data.Sqlite#235 😉

[roji]

FYI, I'm working on some benchmarking and performance aspects of Npgsql, and prepared command use in EFCore would potentially be huge. I hope it gets prioritized at some point.

[roji]

Note that for the next release of Npgsql (3.2), I plan to introduce an automatic/implicit preparation mechanism at the driver level - any statement used more than X times would be automatically prepared up to a certain quota, etc. This should increase performance with EFCore in a very significant way, without any change on your end.

However, it seems like EFCore may have more knowledge as to which statements to prepare - in any case it seems better-placed to make such decisions than the ADO.NET driver. So I think this feature is still relevant at the EFCore level.

[bricelam]

Microsoft.Data.Sqlite will support prepared statements in version 2.1.0. PR: aspnet/Microsoft.Data.Sqlite#388

[bricelam]

The biggest impact this feature would have on SQLite is in the update pipeline. Ironically, not batching statements, but instead re-using single-statement commands yields the best performance.

[roji]

@bricelam is it not possible to both batch and use prepared statements at the same time? With Npgsql/PostgreSQL this yields the best performance.

[bricelam]

SQLite doesn't actually support batching. 😀 We fake it in the ADO.NET layer.

[smitpatel]

Fake it till you make it

[roji]

Oh right, for a moment I forgot we're talking about a local database where batching would be of limited utility anyway :)