Struct AttrsBuilder 

pub struct AttrsBuilder {

    label: &'static str,
    parent_id: UInt32Builder,
    key: StringDictionaryBuilder<UInt16Type>,
    n_keys: usize,
    ty: UInt8Builder,
    str_: StringDictionaryBuilder<UInt32Type>,
    n_str: usize,
    str_bytes: usize,
    int: Int64Builder,
    double: Float64Builder,
    bool_: BooleanBuilder,
    bytes: BinaryBuilder,
    ser: BinaryBuilder,
}

Builder for one attribute table — log, span, event, link, data point, exemplar, resource or scope. They are all the same nine columns.

Fields

label: &'static str

Which table this is, for error messages. "log_attrs.key" beats "attrs.key" when three of them are open at once.

parent_id: UInt32Builder

key: StringDictionaryBuilder<UInt16Type>

n_keys: usize

Distinct entries in key. Tracked here because the dictionary builder does not expose it, and the seal decision has to be made before an append rather than after one has already failed.

ty: UInt8Builder

str_: StringDictionaryBuilder<UInt32Type>

See schema::ATTRS for why this one column is dictionary-encoded. The width is u32, so unlike key it has no cap and no seal-early check.

n_str: usize

Distinct values in str_, and the bytes they hold. Same reason as n_keys: the builder exposes neither, and with a dictionary the heap is what survives deduplication rather than what was appended — counting the appends would seal a block of one repeated 32 KB prompt hundreds of times too early.

str_bytes: usize

int: Int64Builder

double: Float64Builder

bool_: BooleanBuilder

bytes: BinaryBuilder

ser: BinaryBuilder

Implementations

impl AttrsBuilder

pub fn new(label: &'static str) -> Self

pub fn len(&self) -> usize

pub fn is_empty(&self) -> bool

pub fn has_headroom(&self, n: usize) -> bool

Whether n more attribute rows are guaranteed not to overflow the key dictionary. Assumes every one of them is a new key, because the cheap check has to be the conservative one.

pub fn heap_bytes(&self) -> usize

Bytes held in the variable-width value heaps. Fixed-width columns are estimated from the row count; these cannot be, because one row can be a 32 KB GenAI prompt.

pub fn append_all(&mut self, parent_id: u32, kvs: &[KeyValue]) -> Result<()>

Append every attribute of kvs as rows pointing at parent_id.

pub fn append( &mut self, parent_id: u32, key: &str, value: Option<&AnyValue>, ) -> Result<()>

pub fn finish(&self) -> Result<RecordBatch>

See DictColumn::finish for why this does not consume the builder.