From 7ceb325dc3da890e5555faf9c8ea32a1a040c6c6 Mon Sep 17 00:00:00 2001 From: Xiaofan <83447078+xiaofan-luan@users.noreply.github.com> Date: Fri, 17 Sep 2021 09:51:49 +0800 Subject: [PATCH] [skip ci]Add basic_types.md document for segcore (#8116) Signed-off-by: xiaofan-luan --- docs/design_docs/segcore/basic_types.md | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) create mode 100644 docs/design_docs/segcore/basic_types.md diff --git a/docs/design_docs/segcore/basic_types.md b/docs/design_docs/segcore/basic_types.md new file mode 100644 index 0000000000..1242fef365 --- /dev/null +++ b/docs/design_docs/segcore/basic_types.md @@ -0,0 +1,16 @@ +# Glossary + +- `Collection`: Data table, contains multiple Segments +- `Segment`: The memory structure for storing a piece of data supports concurrent insertion, deletion, query, index loading, monitoring and statistics +- `Schema`: Definition of collection data format, including + - `vector`, Order list of FieldMeta + - `isAutoId`. if set to True , default primary field is `RowId` and it is auto generated + - `primaryKey` (when `isAutoId = False`), specify primary key field +- `FieldMeta`: field properties, including + - `DataType` data type, including Int8...Int64, Float, Double, FloatVector, BinaryVector and String later + - `Dim` (when dataType is vector type): vector dimension + - `metric_type` (when dataType is vector type, optional): (when datatype is vector type, optional): the metrictype corresponding to this vector is related to the small batch index and can be empty + - `FieldName`: column name + - `FieldId`: unique number of the column + - (hidden) `FieldOffset`, which is the subscript of `vector` in the schema. The internal calculation of segcore is basically based on fieldoffset +- `Span` is similar to STD:: span， It supports vector type data and can be implicitly converted to `SpanBase` for interface overwrite