timelake 0.5.0

The open time-series lakehouse: fast queries on your own object storage, with no database and no maintenance jobs
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161

//! Mapping Arrow data onto a table's schema: on create (types Iceberg can store) and on write
//! (incoming batches cast and reordered to the table schema).

use arrow::array::{ArrayRef, RecordBatch, new_null_array};
use arrow::compute::cast;
use arrow::datatypes::{DataType, Field, Schema, SchemaRef, TimeUnit};

use crate::error::{Error, Result};

/// The time zone timestamps with any zone are stored in.
pub(crate) const UTC: &str = "+00:00";

/// Maps Arrow types to ones with a lossless Iceberg equivalent: any timestamp → microseconds,
/// small ints → int, dictionaries → their value type, large/view strings → strings. The time
/// column is made NOT NULL, and a text time column (CSV, JSON) becomes a UTC timestamp.
pub(crate) fn normalize_schema(schema: &Schema, time_column: Option<&str>) -> Result<Schema> {
    let fields = schema
        .fields()
        .iter()
        .map(|f| {
            let is_time = Some(f.name().as_str()) == time_column;
            let data_type = match normalize_type(f.data_type())? {
                DataType::Utf8 if is_time => DataType::Timestamp(TimeUnit::Microsecond, Some(UTC.into())),
                dt => dt,
            };
            Ok(Field::new(f.name(), data_type, f.is_nullable() && !is_time))
        })
        .collect::<Result<Vec<_>>>()?;
    Ok(Schema::new(fields))
}

fn normalize_type(dt: &DataType) -> Result<DataType> {
    Ok(match dt {
        DataType::Int8 | DataType::Int16 | DataType::UInt8 | DataType::UInt16 => DataType::Int32,
        DataType::UInt32 => DataType::Int64,
        DataType::Float16 => DataType::Float32,
        DataType::Timestamp(_, tz) => DataType::Timestamp(TimeUnit::Microsecond, tz.as_ref().map(|_| UTC.into())),
        DataType::Date64 => DataType::Date32,
        DataType::Dictionary(_, v) => normalize_type(v)?,
        DataType::LargeUtf8 | DataType::Utf8View => DataType::Utf8,
        DataType::LargeBinary | DataType::BinaryView => DataType::Binary,
        DataType::UInt64 => return Err(Error::invalid("UInt64 columns are not supported by Iceberg; cast to Int64")),
        other => other.clone(),
    })
}

/// Casts and reorders a batch to the table schema by column name. Missing nullable columns are
/// filled with nulls; extra columns are an error (add them to the schema first).
pub fn align_batch(batch: &RecordBatch, schema: &SchemaRef) -> Result<RecordBatch> {
    let input = batch.schema();
    if let Some(extra) = input.fields().iter().find(|f| schema.field_with_name(f.name()).is_err()) {
        return Err(Error::invalid(format!("column {:?} is not in the table schema", extra.name())));
    }
    let columns = schema
        .fields()
        .iter()
        .map(|f| match input.index_of(f.name()) {
            Ok(i) if batch.column(i).data_type() == f.data_type() => Ok(batch.column(i).clone()),
            Ok(i) => Ok(cast(batch.column(i), f.data_type())?),
            Err(_) if f.is_nullable() => Ok(new_null_array(f.data_type(), batch.num_rows())),
            Err(_) => Err(Error::invalid(format!("required column {:?} is missing", f.name()))),
        })
        .collect::<Result<Vec<ArrayRef>>>()?;
    RecordBatch::try_new(schema.clone(), columns)
        .map_err(|e| Error::invalid(format!("batch does not fit the table schema: {e}")))
}

#[cfg(test)]
mod tests {
    use std::sync::Arc;

    use arrow::array::{Array, Int8Array, Int32Array, StringArray};

    use super::*;

    #[test]
    fn normalizes_types_iceberg_cannot_store() {
        let schema = Schema::new(vec![
            Field::new("t", DataType::Timestamp(TimeUnit::Nanosecond, Some("Europe/Kyiv".into())), true),
            Field::new("local", DataType::Timestamp(TimeUnit::Millisecond, None), true),
            Field::new("small", DataType::UInt8, false),
            Field::new("u32", DataType::UInt32, true),
            Field::new("tag", DataType::Dictionary(Box::new(DataType::Int8), Box::new(DataType::LargeUtf8)), true),
            Field::new("day", DataType::Date64, true),
        ]);
        let n = normalize_schema(&schema, Some("t")).unwrap();
        let types: Vec<_> = n.fields().iter().map(|f| f.data_type().clone()).collect();
        assert_eq!(
            types,
            [
                DataType::Timestamp(TimeUnit::Microsecond, Some(UTC.into())),
                DataType::Timestamp(TimeUnit::Microsecond, None),
                DataType::Int32,
                DataType::Int64,
                DataType::Utf8,
                DataType::Date32,
            ]
        );
        // The time column is required; other columns keep their nullability.
        assert!(!n.field(0).is_nullable() && !n.field(2).is_nullable() && n.field(3).is_nullable());
    }

    #[test]
    fn text_time_column_becomes_a_timestamp() {
        let schema = Schema::new(vec![Field::new("time", DataType::Utf8, true), Field::new("s", DataType::Utf8, true)]);
        let n = normalize_schema(&schema, Some("time")).unwrap();
        assert_eq!(n.field(0).data_type(), &DataType::Timestamp(TimeUnit::Microsecond, Some(UTC.into())));
        assert_eq!(n.field(1).data_type(), &DataType::Utf8);
    }

    #[test]
    fn rejects_uint64() {
        let schema = Schema::new(vec![Field::new("n", DataType::UInt64, true)]);
        assert!(normalize_schema(&schema, None).is_err());
    }

    fn table_schema() -> SchemaRef {
        Arc::new(Schema::new(vec![Field::new("id", DataType::Int32, false), Field::new("name", DataType::Utf8, true)]))
    }

    #[test]
    fn aligns_by_name_and_casts() {
        let input = Arc::new(Schema::new(vec![Field::new("id", DataType::Int8, false)]));
        let batch = RecordBatch::try_new(input, vec![Arc::new(Int8Array::from(vec![1, 2]))]).unwrap();
        let aligned = align_batch(&batch, &table_schema()).unwrap();
        assert_eq!(aligned.schema(), table_schema());
        assert_eq!(aligned.column(0).as_any().downcast_ref::<Int32Array>().unwrap().values(), &[1, 2]);
        assert_eq!(aligned.column(1).null_count(), 2);

        let reordered = RecordBatch::try_new(
            Arc::new(Schema::new(vec![
                Field::new("name", DataType::Utf8, true),
                Field::new("id", DataType::Int32, false),
            ])),
            vec![Arc::new(StringArray::from(vec!["a"])), Arc::new(Int32Array::from(vec![7]))],
        )
        .unwrap();
        let aligned = align_batch(&reordered, &table_schema()).unwrap();
        assert_eq!(aligned.column(0).as_any().downcast_ref::<Int32Array>().unwrap().value(0), 7);
    }

    #[test]
    fn rejects_extra_and_missing_required_columns() {
        let extra = RecordBatch::try_new(
            Arc::new(Schema::new(vec![
                Field::new("id", DataType::Int32, false),
                Field::new("x", DataType::Int32, true),
            ])),
            vec![Arc::new(Int32Array::from(vec![1])), Arc::new(Int32Array::from(vec![1]))],
        )
        .unwrap();
        assert!(align_batch(&extra, &table_schema()).is_err());

        let missing = RecordBatch::try_new(
            Arc::new(Schema::new(vec![Field::new("name", DataType::Utf8, true)])),
            vec![Arc::new(StringArray::from(vec!["a"]))],
        )
        .unwrap();
        assert!(align_batch(&missing, &table_schema()).is_err());
    }
}