package arcticc.pb2.descriptors_pb2

Get desktop application:
View/edit binary Protocol Buffers messages

Used in: NormalizationMetadata.ExperimentalArrow.ColumnMeta

• CATEGORICAL = 0

• LARGE_STRING = 1

• SMALL_STRING = 2

Used in: FrameMetadata, TimeSeriesDescriptor

• oneof id

  • string string_id = 1

  • int64 numeric_id = 2

• uint64 version_id = 3

• uint64 creation_ts = 4

• uint64 content_hash = 5

• oneof index_start

  • string string_start = 6

  • int64 numeric_start = 7

• oneof index_end

  • string string_end = 8

  • int64 numeric_end = 9

• KeyType key_type = 10

Used in: TimeSeriesDescriptor

• repeated uint32 group_ends = 1

  exclusive boundary of the group, ie for 2 elements: [0,group_ends[0][, [group_ends[0], group_ends[1][

  not used

• uint32 num_buckets = 2

  not used

• bool enabled = 3

• optional NormalizationMetadata normalization = 1

• optional UserDefinedMetadata user_meta = 2

• optional AtomKey next_key = 3

• optional UserDefinedMetadata multi_key_meta = 4

Used in: StreamDescriptor, mapped_file_storage_pb2.Config

• uint32 field_count = 1

• IndexDescriptor.Type kind = 2

Used in: IndexDescriptor

• UNKNOWN = 0

• EMPTY = 69

  Used then the dataframe has 0 rows. Convertible to any other type

  'E'

• ROWCOUNT = 82

  'R'

• STRING = 83

  'S'

• TIMESTAMP = 84

  'T'

• TIME_SYMBOL = 85

  'U'

Used in: AtomKey

• STREAM_GROUP = 0

• GENERATION = 1

• TABLE_DATA = 2

• TABLE_INDEX = 3

• VERSION = 4

• VERSION_JOURNAL = 5

• METRICS = 6

• SNAPSHOT = 7

• SYMBOL_LIST = 8

• VERSION_REF = 9

• STORAGE_INFO = 10

• APPEND_REF = 11

• MULTI_KEY = 12

• LOCK = 13

• SNAPSHOT_REF = 14

• TOMBSTONE = 15

• APPEND_DATA = 16

• LOG = 17

• PARTITION = 18

• UNDEFINED = 19

(message has no fields)

• UNSUPPORTED = 0

• PD_TIMESTAMP = 32

• PY_DATETIME = 33

• PY_TIMEDELTA = 34

• PY_PICKLE_2 = 102

• PY_PICKLE_3 = 103

* This type represents all the additional info required to reconstruct the object that has been written. The modelling reflects the extra info necessary to cater for the input specific data. For now focused on pandas, but support for any other type without a 1:1 mapping to the first class StreamDescriptor model should be handled through normalization and appropriate metadata additions here.

Used in: FrameMetadata, TimeSeriesDescriptor

• oneof input_type

  • NormalizationMetadata.PandasDataFrame df = 1

  • NormalizationMetadata.PandasDataFrame series = 2

  • NormalizationMetadata.NormalisedTimeSeries ts = 3

  • NormalizationMetadata.MsgPackFrame msg_pack_frame = 4

  • NormalizationMetadata.NdArray np = 5

  • NormalizationMetadata.ExperimentalArrow experimental_arrow = 6

• optional NormalizationMetadata.CustomNormalizerMeta custom = 64

Used in: NormalizationMetadata

• string class_name = 1

• bytes meta = 2

• bool py_datetime_marker = 3

• bool daterange_has_datetime_marker = 4

Used in: NormalizationMetadata

• bool has_index = 1

• map<string, ExperimentalArrow.ColumnMeta> columns = 2

Used in: ExperimentalArrow

• optional string timezone = 1

  Could make these a oneof, as they are mutually exclusive. But there may be other column-specific metadata in future that is not mutually exclusive, so this is more extensible.

• optional ArrowStringFormat string_format = 2

Used in: NormalizationMetadata

• uint32 version = 1

  To set at least one field

• uint64 size_bytes = 2

Used in: NormalizationMetadata

• repeated uint64 shape = 1

Metadata for an arcticdb TimeFrame; mirrors PandasDataFrame.

Used in: NormalizationMetadata

• bool mark = 1

• optional Pandas common = 2

• optional PandasMultiColumn multi_columns = 3

  Unused (MultiIndex columns are unimplemented).

• bool has_synthetic_columns = 4

  Columns were an unnamed RangeIndex (0..n).

* This represents the common metadata for pandas types which are shared between series and dataframes.

Used in: NormalisedTimeSeries, PandasDataFrame

• string name = 1

  Series name only (empty for DataFrames).

• bool has_name = 9

  Distinguishes a None series name from an empty-string name

• oneof index_type

  • PandasIndex index = 2

  • PandasMultiIndex multi_index = 3

• bool mark = 4

  Sentinel: set so the message is non-empty even when nothing else needs setting.

• map<string, Pandas.Categories> categories = 5

  Category values for string categorical columns, by column name.

• map<string, Pandas.IntCategories> int_categories = 6

  Category values for integer categorical columns, by column name.

• map<string, Pandas.ColumnName> col_names = 7

  On-disk column name -> original label. Populated only for columns that were sanitized.

• optional PandasIndex columns = 8

  Columns-axis metadata; only name and fake_name are used.

Used in: Pandas

• repeated string category = 1

Records the original label of a column whose on-disk name had to be sanitized.

Used in: Pandas

• bool is_none = 1

  Original column name was None.

• bool is_empty = 2

  Original column name was an empty string.

• string original_name = 3

  Original column label as a string.

• bool is_int = 4

  Original column name was an integer; cast back on read.

Used in: Pandas

• repeated uint64 category = 1

Used in: NormalizationMetadata

• optional Pandas common = 1

• optional PandasMultiColumn multi_columns = 2

  Unused (MultiIndex columns are unimplemented).

• bool has_synthetic_columns = 3

  Columns were an unnamed RangeIndex (0..n).

Metadata for a single-level pandas Index.

Used in: Pandas

• string name = 1

  Index name; empty when unnamed, see fake_name.

• string tz = 2

  Index timezone; empty if tz-naive.

• bool fake_name = 3

  The index had no name; stored under a synthetic name, restored to None on read.

• bool is_physically_stored = 4

  True if the index is stored on disk as a column. False if it can be reconstructed from the metadata at read time (a RangeIndex, rebuilt from start/step).

• int64 start = 5

  RangeIndex start (used when not physically stored).

• int64 step = 6

  RangeIndex step (used when not physically stored).

• bool is_int = 7

  The index name was an integer; cast back from its string form on read.

MultiIndex columns. Not implemented

Used in: NormalisedTimeSeries, PandasDataFrame

• repeated string levels = 1

• repeated PandasMultiColumn.Labels labels = 2

• map<uint32, string> timezone = 3

Used in: PandasMultiColumn

• repeated uint64 val = 1

Metadata for a pandas MultiIndex. Index levels are physically stored columns. If the first level is a timeseries it is used as a primary index (allowing date_range filtering etc.)

Used in: Pandas

• uint32 version = 1

  Unused

• map<uint32, string> timezone = 2

  Per-level timezone for levels >= 1 (level 0 uses `tz`).

• uint32 field_count = 3

  Number of index levels beyond the first.

• string name = 4

  First level name; empty when unnamed, see fake_field_pos.

• string tz = 5

  First level (level 0) timezone.

• repeated uint32 fake_field_pos = 6

  Positions of unnamed levels (0 == first level).

• bool is_int = 7

  The first level index name was an integer (as of July 2026 - never populated)

Used in: StreamDescriptor

• UNKNOWN = 0

• UNSORTED = 1

• ASCENDING = 2

• DESCENDING = 3

Used in: TimeSeriesDescriptor, encoding_pb2.SegmentHeader

• oneof id

  Channel can output a single type of data possibly containing multiple fields

  • uint64 num_id = 1

  • string str_id = 2

• optional IndexDescriptor index = 3

• repeated StreamDescriptor.FieldDescriptor fields = 4

• uint64 type_hash = 5

• string timezone = 6

  unused but reserved in case first class support is needed

• SortedValue sorted = 7

• uint64 in_bytes = 8

• uint64 out_bytes = 9

• uint64 row_count = 10

Used in: StreamDescriptor

• optional TypeDescriptor type_desc = 1

• string name = 2

• bool enabled = 1

• uint64 total_rows = 1

• optional StreamDescriptor stream_descriptor = 2

• optional NormalizationMetadata normalization = 3

• optional ColumnGroups column_groups = 4

• optional UserDefinedMetadata user_meta = 5

• optional AtomKey next_key = 6

• optional UserDefinedMetadata multi_key_meta = 7

Used in: StreamDescriptor.FieldDescriptor

• TypeDescriptor.ValueType value_type = 1

• TypeDescriptor.SizeBits size_bits = 2

  useful to specify 8/16/32/64 for [u]int, and float{32,64}. Defaults to 0 for unused

• uint32 dimension = 3

  Dimension of a single tick stored 0: scalar 1: array of values 2: matrix ... This maps numpy ndarray concept

Used in: TypeDescriptor

• UNKNOWN_SIZE_BITS = 0

• S8 = 1

• S16 = 2

• S32 = 3

• S64 = 4

Used in: TypeDescriptor

• UNKNOWN_VALUE_TYPE = 0

• UINT = 1

• INT = 2

• FLOAT = 3

• BOOL = 4

• NANOSECONDS_UTC = 5

• SYMBOL = 6

  categorical string of low cardinality suitable for dictionary encoding

• ASCII_STRING = 7

  fixed size string when dim > 1, inputs of type uint8_t, no encoding

• UTF8_STRING = 8

  fixed size string when dim > 1, inputs of type uint8_t, utf8 encoding

• BYTES = 9

  implies fixed size bytes array when dim > 1, opaque

• PICKLE = 10

  BYTES + pickle specific encoding

• DYNAMIC_STRING = 11

  offset reference to deduplicated string pool

• DYNAMIC_UTF8 = 12

• EMPTY = 13

• BOOL_OBJECT = 14

  Nullable boolean

Used in: FrameMetadata, TimeSeriesDescriptor

• UserDefinedMetadata.SerializationType type = 1

• uint32 version = 2

  in case logic needs to be versioned

• oneof storage_type

  • bytes inline_payload = 3

    capped to fixed size in code

Used in: UserDefinedMetadata

• MSG_PACK = 0

  defaults to msg pack