qql-plan 0.3.0

Typed QQL execution plans and Qdrant route lowering
Documentation

qql-plan

Transport-neutral lowering: AST → PlannedOperation.

Proposition

One plan for every backend. REST is a projection (to_rest_route); gRPC maps the same typed structs to protobuf. Never invent wire shapes in bindings — plan here once.

Stmt (qql-core)
    │ plan()
    ▼
PlannedOperation          ← source of truth
    ├── filter            → REST filter / gRPC Filter
    ├── shard_key         → REST shard_key / gRPC ShardKeySelector
    ├── to_rest_route()   → Route { method, path, query, body }
    └── gRPC match arm    → typed protobuf (runtime)

Filter never carries shard_key. Isolation and routing are siblings on the request (OpenAPI + proto both model it that way).

Usage

use qql_core::parser::Parser;
use qql_plan::plan::{plan, to_rest_route, try_route};

let stmt = Parser::parse(
    "QUERY TEXT 'hello' FROM docs USING dense SHARD 'acme' LIMIT 5"
)?;
let op = plan(&stmt)?;
let route = to_rest_route(&op)?;   // fallible REST projection
// Prefer try_route(&stmt) for libraries; avoid panic-prone route()

PlannedOperation (selected)

Family Variants
Search Query, QueryGroups, GetPoints, Scroll, Count
Mutations Upsert, Delete, ClearPayload, UpdateVectors, DeleteVectors, UpdatePayload, …
DDL CreateCollection, UpdateCollection, indexes, CreateShardKey / DropShardKey / ListShardKeys
Quotas GetQuotas / SetQuotas → REST `GET
Client-side CrossRerank (no single Qdrant route)

Batch families

  • Query — contiguous same-collection queries → /points/query/batch
  • Mutation — contiguous mutations → /points/batch
  • Single — DDL, scroll, count, quotas, …

Semantic primitives

PlanPointId, PlanVectorValue (Dense / Sparse / MultiDense), PlanQueryInput, typed formula trees — stay typed until a transport boundary. MemoryPlacement / VectorDatatype re-exported from qql-core.

Qdrant 1.19 lowering notes

Surface Plan behavior
SET QUOTA (…) Full replace body (SetQuotaRequest). Omitted keys / key = null are unset in the PUT body — not a merge of the previous config. Invalid keys/ranges → QQL-PLAN-QUOTA. Optional WAIT → query ?wait=.
PARAMS (idf = …) 'global' or WHERE <filter>; empty lowered corpus → QQL-PLAN-IDF.
MATCH PREFIX / SLICE MatchValue::Prefix / SliceCondition on the filter IR.
memory / payload_memory / datatype / keyword prefix Forwarded on collection, vector, HNSW, quantization, and index REST bodies.
let op = plan(&Parser::parse(
    "SET QUOTA (enabled = true, max_resident_memory_percent = 80) WAIT true;"
)?)?;
let route = to_rest_route(&op)?; // PUT /quotas?wait=true

Modules

Module Role
plan plan, to_rest_route, try_route, compile_statement
query / mutation / ddl Lowering (including quotas / IDF / memory)
filter FilterExpression only (no routing fields)
types Request IR (SetQuotaRequest, IdfSearchParams, …)

Docs

Test

cargo test -p qql-plan