Query Planner And Fast Paths

DataVo has a general SQL path and a set of narrower fast paths. The rule is simple: a fast path must either produce the same result as the normal path or decline and fall back.

The general path is what you use first.

SELECT Id, Name, Score
FROM Users
WHERE Score > 90
ORDER BY Score DESC
LIMIT 10;

Source-generated queries remove repeated parse and mapping work for supported query patterns.

public sealed record PlayerRow(int Id, string Name, int Level);

public static partial class PlayerQueries
{
    [DataVoQuery("SELECT Id, Name, Level FROM Players WHERE Id = @id")]
    public static partial PlayerRow? GetPlayer(DataVoContext db, int id);
}

The fixed-width update fast path is intentionally narrow. It is for compiled single-row updates that can be found through an integer primary-key lane and patched safely.

public static partial class PlayerWrites
{
    [DataVoQuery("UPDATE Players SET Level = @level WHERE Id = @id")]
    public static partial int UpdateLevel(DataVoContext db, int id, int level);
}

Vector predicate fast paths can use ANN prefiltering when the planner estimates that an index will reduce work.

SELECT Id, Label, Emb <=> '[1,0,0]' AS distance
FROM Embeddings
WHERE Emb <=> '[1,0,0]' < 0.25
ORDER BY distance ASC
LIMIT 10;

The Volcano planner settings are present for limited experiments. Leave them off unless you are specifically testing planner behavior.

using var db = new DataVoContext(new DataVoConfig
{
    StorageMode = StorageMode.InMemory,
    EnableVolcanoExecution = false,
    EnableVolcanoSpillGuardrails = true
});

Planner And Fast Path Support Summary

Supported paths include the general parser/executor (the default for DataVoContext.Execute), source-generated SELECT plans for narrow equality shapes ([DataVoQuery]), typed row readers, single-column manifest index hints, fixed-width compiled updates for narrow single-row shapes, the arena-backed LSM MemTable write path, and the vector-predicate ANN prefilter fast path. Volcano execution is present but disabled by default (planned); there is no arbitrary-SQL fast path, and the planner is not a cost-based optimizer at PostgreSQL parity.