Struct SchemaRegistry 

pub struct SchemaRegistry { /* private fields */ }

Thread-safe, in-memory schema registry.

Tracks the schema for every measurement. Schemas are created automatically on first write and evolve additively (new tags/fields added, type conflicts rejected).

Schema Evolution

The registry follows a schema-on-write model:

Operation	Supported	Mechanism
Add tag/field	Yes	Automatic on write
Type conflict	Rejected	SchemaError
Remove column	No	N/A (append-only)
Rename column	No	N/A
Change column type	No	Write new field

Durability is the catalog manifest’s job: the database persists every action this registry returns before it writes the data that needs it.

Implementations

impl SchemaRegistry

pub fn new() -> Self

Create an empty schema registry.

pub fn register_batch( &self, points: &[&Point], ) -> Result<Vec<SchemaAction>, SchemaError>

Register every point of a batch, or none of them.

Validates the whole batch against the current schemas and against the columns earlier points of the same batch introduce, and only then applies the additions. A type conflict anywhere in the batch rejects it with the registry untouched — which is the property that lets the caller persist every returned action unconditionally. Registering point by point left the columns of the first points behind when a later point was refused, and nothing ever persisted them.

Returns the actions the batch caused, in application order. Empty when the batch fits the existing schemas, which is the common case and takes no lock.

Errors

Returns SchemaError::TypeConflict if any field’s type conflicts with the schema, or with the type another point of the batch gives the same field.

pub fn normalize_decimals( &self, points: &[&Point], ) -> Result<Option<Vec<Point>>, SchemaError>

Rescale every decimal field of points to its column’s scale.

A decimal column stores one number of fractional digits, and this is where a value that arrived with fewer is widened to it — 1.5 into a scale-4 column becomes 1.5000. Without it the memtable, the WAL and two segments of the same column could each hold a different scale, and a scan across them could not produce one Arrow schema.

Returns None when nothing needed changing, which is the common case and copies nothing: a batch with no decimal fields never allocates. Otherwise returns the whole batch, with the affected points rewritten.

Call after register_batch, so the columns the batch introduces already exist.

Errors

SchemaError::DecimalScaleConflict if a value carries more fractional digits than its column stores. register_batch refuses the same batch for the same reason, so reaching this is a sign the two were called out of order.

pub fn declare_field( &self, measurement: &str, field: &str, column_type: ColumnType, ) -> Result<Vec<SchemaAction>, SchemaError>

Declare a field column before anything is written to it.

The reason this exists is decimals: a decimal column’s scale is fixed by whatever creates the column, and letting the first meter reading decide how many fractional digits a settlement register keeps is a coin toss. Declaring Decimal { scale: 4 } up front makes it a decision.

Returns the actions to persist — empty when the column already existed with exactly this type.

Errors

SchemaError::TypeConflict if the column exists with a different type, a different decimal scale, or as a tag.

pub fn register_point( &self, point: &Point, ) -> Result<Vec<SchemaAction>, SchemaError>

Register a single point — register_batch for one point.

Errors

Returns SchemaError::TypeConflict if a field’s type doesn’t match the existing schema.

pub fn lookup(&self, measurement: &str) -> Option<Arc<MeasurementSchema>>

Look up the schema for a measurement.

Returns Arc<MeasurementSchema> for an O(1) clone.

pub fn register_schema(&self, name: &str, schema: MeasurementSchema)

Register a pre-built schema directly (e.g. during catalog replay on startup). If the measurement already has a schema, the existing one is kept unchanged.

pub fn register_measurement(&self, ms: MeasurementSchema)

Register a measurement from a MeasurementSchema (WAL replay).

Idempotent: if the measurement already exists the call is a no-op.

pub fn apply_add_column(&self, measurement: &str, column: ColumnDef)

Apply an AddColumn action from WAL replay.

Idempotent: if the column already exists it is not duplicated. Checks for role conflicts (e.g., tag vs field with same name).

pub fn measurement_count(&self) -> usize

Returns the number of registered measurements.

pub fn measurement_names(&self) -> Vec<String>

Returns all registered measurement names.

pub fn remove(&self, measurement: &str) -> Option<Arc<MeasurementSchema>>

Remove a measurement schema. Returns the removed schema if it existed.

Trait Implementations

impl Clone for SchemaRegistry

fn clone(&self) -> SchemaRegistry

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SchemaRegistry

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for SchemaRegistry

fn default() -> Self

Returns the “default value” for a type.