promql/functions/
predict_linear.rs

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
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
// Copyright 2023 Greptime Team
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

//! Implementation of [`predict_linear`](https://prometheus.io/docs/prometheus/latest/querying/functions/#predict_linear) in PromQL. Refer to the [original
//! implementation](https://github.com/prometheus/prometheus/blob/90b2f7a540b8a70d8d81372e6692dcbb67ccbaaa/promql/functions.go#L859-L872).

use std::sync::Arc;

use datafusion::arrow::array::{Float64Array, TimestampMillisecondArray};
use datafusion::arrow::datatypes::TimeUnit;
use datafusion::common::DataFusionError;
use datafusion::logical_expr::{ScalarUDF, Volatility};
use datafusion::physical_plan::ColumnarValue;
use datafusion_expr::create_udf;
use datatypes::arrow::array::Array;
use datatypes::arrow::datatypes::DataType;

use crate::error;
use crate::functions::{extract_array, linear_regression};
use crate::range_array::RangeArray;

pub struct PredictLinear {
    /// Duration. The second param of (`predict_linear(v range-vector, t scalar)`).
    t: i64,
}

impl PredictLinear {
    fn new(t: i64) -> Self {
        Self { t }
    }

    pub const fn name() -> &'static str {
        "prom_predict_linear"
    }

    pub fn scalar_udf(t: i64) -> ScalarUDF {
        let input_types = vec![
            // time index column
            RangeArray::convert_data_type(DataType::Timestamp(TimeUnit::Millisecond, None)),
            // value column
            RangeArray::convert_data_type(DataType::Float64),
        ];
        create_udf(
            Self::name(),
            input_types,
            DataType::Float64,
            Volatility::Immutable,
            Arc::new(move |input: &_| Self::new(t).predict_linear(input)) as _,
        )
    }

    fn predict_linear(&self, input: &[ColumnarValue]) -> Result<ColumnarValue, DataFusionError> {
        // construct matrix from input.
        assert_eq!(input.len(), 2);
        let ts_array = extract_array(&input[0])?;
        let value_array = extract_array(&input[1])?;

        let ts_range: RangeArray = RangeArray::try_new(ts_array.to_data().into())?;
        let value_range: RangeArray = RangeArray::try_new(value_array.to_data().into())?;
        error::ensure(
            ts_range.len() == value_range.len(),
            DataFusionError::Execution(format!(
                "{}: input arrays should have the same length, found {} and {}",
                Self::name(),
                ts_range.len(),
                value_range.len()
            )),
        )?;
        error::ensure(
            ts_range.value_type() == DataType::Timestamp(TimeUnit::Millisecond, None),
            DataFusionError::Execution(format!(
                "{}: expect TimestampMillisecond as time index array's type, found {}",
                Self::name(),
                ts_range.value_type()
            )),
        )?;
        error::ensure(
            value_range.value_type() == DataType::Float64,
            DataFusionError::Execution(format!(
                "{}: expect Float64 as value array's type, found {}",
                Self::name(),
                value_range.value_type()
            )),
        )?;

        // calculation
        let mut result_array = Vec::with_capacity(ts_range.len());

        for index in 0..ts_range.len() {
            let timestamps = ts_range
                .get(index)
                .unwrap()
                .as_any()
                .downcast_ref::<TimestampMillisecondArray>()
                .unwrap()
                .clone();
            let values = value_range
                .get(index)
                .unwrap()
                .as_any()
                .downcast_ref::<Float64Array>()
                .unwrap()
                .clone();
            error::ensure(
                timestamps.len() == values.len(),
                DataFusionError::Execution(format!(
                    "{}: input arrays should have the same length, found {} and {}",
                    Self::name(),
                    timestamps.len(),
                    values.len()
                )),
            )?;

            let ret = predict_linear_impl(&timestamps, &values, self.t);

            result_array.push(ret);
        }

        let result = ColumnarValue::Array(Arc::new(Float64Array::from_iter(result_array)));
        Ok(result)
    }
}

fn predict_linear_impl(
    timestamps: &TimestampMillisecondArray,
    values: &Float64Array,
    t: i64,
) -> Option<f64> {
    if timestamps.len() < 2 {
        return None;
    }

    // last timestamp is evaluation timestamp
    let evaluate_ts = timestamps.value(timestamps.len() - 1);
    let (slope, intercept) = linear_regression(timestamps, values, evaluate_ts);

    if slope.is_none() || intercept.is_none() {
        return None;
    }

    Some(slope.unwrap() * t as f64 + intercept.unwrap())
}

#[cfg(test)]
mod test {
    use std::vec;

    use super::*;
    use crate::functions::test_util::simple_range_udf_runner;

    // build timestamp range and value range arrays for test
    fn build_test_range_arrays() -> (RangeArray, RangeArray) {
        let ts_array = Arc::new(TimestampMillisecondArray::from_iter(
            [
                0i64, 300, 600, 900, 1200, 1500, 1800, 2100, 2400, 2700, 3000,
            ]
            .into_iter()
            .map(Some),
        ));
        let ranges = [(0, 11)];

        let values_array = Arc::new(Float64Array::from_iter([
            0.0, 10.0, 20.0, 30.0, 40.0, 0.0, 10.0, 20.0, 30.0, 40.0, 50.0,
        ]));

        let ts_range_array = RangeArray::from_ranges(ts_array, ranges).unwrap();
        let value_range_array = RangeArray::from_ranges(values_array, ranges).unwrap();

        (ts_range_array, value_range_array)
    }

    #[test]
    fn calculate_predict_linear_none() {
        let ts_array = Arc::new(TimestampMillisecondArray::from_iter(
            [0i64].into_iter().map(Some),
        ));
        let ranges = [(0, 0), (0, 1)];
        let values_array = Arc::new(Float64Array::from_iter([0.0]));
        let ts_array = RangeArray::from_ranges(ts_array, ranges).unwrap();
        let value_array = RangeArray::from_ranges(values_array, ranges).unwrap();
        simple_range_udf_runner(
            PredictLinear::scalar_udf(0),
            ts_array,
            value_array,
            vec![None, None],
        );
    }

    #[test]
    fn calculate_predict_linear_test1() {
        let (ts_array, value_array) = build_test_range_arrays();
        simple_range_udf_runner(
            PredictLinear::scalar_udf(0),
            ts_array,
            value_array,
            // value at t = 0
            vec![Some(38.63636363636364)],
        );
    }

    #[test]
    fn calculate_predict_linear_test2() {
        let (ts_array, value_array) = build_test_range_arrays();
        simple_range_udf_runner(
            PredictLinear::scalar_udf(3000),
            ts_array,
            value_array,
            // value at t = 3000
            vec![Some(31856.818181818187)],
        );
    }

    #[test]
    fn calculate_predict_linear_test3() {
        let (ts_array, value_array) = build_test_range_arrays();
        simple_range_udf_runner(
            PredictLinear::scalar_udf(4200),
            ts_array,
            value_array,
            // value at t = 4200
            vec![Some(44584.09090909091)],
        );
    }

    #[test]
    fn calculate_predict_linear_test4() {
        let (ts_array, value_array) = build_test_range_arrays();
        simple_range_udf_runner(
            PredictLinear::scalar_udf(6600),
            ts_array,
            value_array,
            // value at t = 6600
            vec![Some(70038.63636363638)],
        );
    }

    #[test]
    fn calculate_predict_linear_test5() {
        let (ts_array, value_array) = build_test_range_arrays();
        simple_range_udf_runner(
            PredictLinear::scalar_udf(7800),
            ts_array,
            value_array,
            // value at t = 7800
            vec![Some(82765.9090909091)],
        );
    }
}