risingwave_connector/sink/encoder/
avro.rs

1// Copyright 2025 RisingWave Labs
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7//     http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15use std::collections::HashMap;
16use std::sync::Arc;
17
18use apache_avro::schema::{Name, RecordSchema, Schema as AvroSchema};
19use apache_avro::types::{Record, Value};
20use risingwave_common::array::VECTOR_ITEM_TYPE;
21use risingwave_common::catalog::Schema;
22use risingwave_common::row::Row;
23use risingwave_common::types::{DataType, DatumRef, ScalarRefImpl, StructType};
24use risingwave_common::util::iter_util::{ZipEqDebug, ZipEqFast};
25use risingwave_connector_codec::decoder::utils::rust_decimal_to_scaled_bigint;
26use thiserror_ext::AsReport;
27
28use super::{FieldEncodeError, Result as SinkResult, RowEncoder, SerTo};
29
30type Result<T> = std::result::Result<T, FieldEncodeError>;
31struct NamesRef(HashMap<Name, AvroSchema>);
32
33pub struct AvroEncoder {
34    schema: Schema,
35    col_indices: Option<Vec<usize>>,
36    avro_schema: Arc<AvroSchema>,
37    refs: NamesRef,
38    header: AvroHeader,
39}
40
41#[derive(Debug, Clone, Copy)]
42pub enum AvroHeader {
43    None,
44    /// <https://avro.apache.org/docs/1.11.1/specification/#single-object-encoding>
45    ///
46    /// * C3 01
47    /// * 8-byte little-endian CRC-64-AVRO fingerprint
48    SingleObject,
49    /// <https://avro.apache.org/docs/1.11.1/specification/#object-container-files>
50    ///
51    /// * 4F 62 6A 01
52    /// * schema
53    /// * 16-byte random sync marker
54    ContainerFile,
55    /// <https://docs.confluent.io/platform/7.5/schema-registry/fundamentals/serdes-develop/index.html#messages-wire-format>
56    ///
57    /// * 00
58    /// * 4-byte big-endian schema ID
59    ConfluentSchemaRegistry(i32),
60    /// <https://github.com/awslabs/aws-glue-schema-registry/blob/v1.1.20/common/src/main/java/com/amazonaws/services/schemaregistry/utils/AWSSchemaRegistryConstants.java#L59-L61>
61    ///
62    /// * 03
63    /// * 00
64    /// * 16-byte UUID identifying a specific schema version
65    GlueSchemaRegistry(uuid::Uuid),
66}
67
68impl AvroEncoder {
69    pub fn new(
70        schema: Schema,
71        col_indices: Option<Vec<usize>>,
72        avro_schema: Arc<AvroSchema>,
73        header: AvroHeader,
74    ) -> SinkResult<Self> {
75        let refs = NamesRef::new(&avro_schema)?;
76        match &col_indices {
77            Some(col_indices) => validate_fields(
78                col_indices.iter().map(|idx| {
79                    let f = &schema[*idx];
80                    (f.name.as_str(), &f.data_type)
81                }),
82                &avro_schema,
83                &refs,
84            )?,
85            None => validate_fields(
86                schema
87                    .fields
88                    .iter()
89                    .map(|f| (f.name.as_str(), &f.data_type)),
90                &avro_schema,
91                &refs,
92            )?,
93        };
94
95        Ok(Self {
96            schema,
97            col_indices,
98            avro_schema,
99            refs,
100            header,
101        })
102    }
103}
104
105impl NamesRef {
106    fn new(root: &AvroSchema) -> std::result::Result<Self, apache_avro::Error> {
107        let resolved = apache_avro::schema::ResolvedSchema::try_from(root)?;
108        let refs = resolved
109            .get_names()
110            .iter()
111            .map(|(k, v)| (k.to_owned(), (*v).to_owned()))
112            .collect();
113        Ok(Self(refs))
114    }
115
116    fn lookup<'a>(&'a self, avro: &'a AvroSchema) -> &'a AvroSchema {
117        match avro {
118            AvroSchema::Ref { name } => &self.0[name],
119            _ => avro,
120        }
121    }
122}
123
124pub struct AvroEncoded {
125    value: Value,
126    schema: Arc<AvroSchema>,
127    header: AvroHeader,
128}
129
130impl RowEncoder for AvroEncoder {
131    type Output = AvroEncoded;
132
133    fn schema(&self) -> &Schema {
134        &self.schema
135    }
136
137    fn col_indices(&self) -> Option<&[usize]> {
138        self.col_indices.as_deref()
139    }
140
141    fn encode_cols(
142        &self,
143        row: impl Row,
144        col_indices: impl Iterator<Item = usize>,
145    ) -> SinkResult<Self::Output> {
146        let record = encode_fields(
147            col_indices.map(|idx| {
148                let f = &self.schema[idx];
149                ((f.name.as_str(), &f.data_type), row.datum_at(idx))
150            }),
151            &self.avro_schema,
152            &self.refs,
153        )?;
154        Ok(AvroEncoded {
155            value: record.into(),
156            schema: self.avro_schema.clone(),
157            header: self.header,
158        })
159    }
160}
161
162impl SerTo<Vec<u8>> for AvroEncoded {
163    fn ser_to(self) -> SinkResult<Vec<u8>> {
164        use bytes::BufMut as _;
165
166        let header = match self.header {
167            AvroHeader::ConfluentSchemaRegistry(schema_id) => {
168                let mut buf = Vec::with_capacity(1 + 4);
169                buf.put_u8(0);
170                buf.put_i32(schema_id);
171                buf
172            }
173            AvroHeader::GlueSchemaRegistry(schema_version_id) => {
174                let mut buf = Vec::with_capacity(1 + 1 + 16);
175                buf.put_u8(3);
176                buf.put_u8(0);
177                buf.put_slice(schema_version_id.as_bytes());
178                buf
179            }
180            AvroHeader::None | AvroHeader::SingleObject | AvroHeader::ContainerFile => {
181                return Err(crate::sink::SinkError::Encode(format!(
182                    "{:?} unsupported yet",
183                    self.header
184                )));
185            }
186        };
187
188        let raw = apache_avro::to_avro_datum(&self.schema, self.value)
189            .map_err(|e| crate::sink::SinkError::Encode(e.to_report_string()))?;
190        let mut buf = Vec::with_capacity(header.len() + raw.len());
191        buf.put_slice(&header);
192        buf.put_slice(&raw);
193
194        Ok(buf)
195    }
196}
197
198enum OptIdx {
199    /// `T`
200    NotUnion,
201    /// `[T]`
202    Single,
203    /// `[null, T]`
204    NullLeft,
205    /// `[T, null]`
206    NullRight,
207}
208
209/// A trait that assists code reuse between `validate` and `encode`.
210/// * For `validate`, the inputs are (RisingWave type, ProtoBuf type).
211/// * For `encode`, the inputs are (RisingWave type, RisingWave data, ProtoBuf type).
212///
213/// Thus we impl [`MaybeData`] for both [`()`] and [`DatumRef`].
214trait MaybeData: std::fmt::Debug {
215    type Out;
216
217    fn on_base(self, f: impl FnOnce(ScalarRefImpl<'_>) -> Result<Value>) -> Result<Self::Out>;
218
219    /// Switch to `RecordSchema` after #12562
220    fn on_struct(self, st: &StructType, avro: &AvroSchema, refs: &NamesRef) -> Result<Self::Out>;
221
222    fn on_list(self, elem: &DataType, avro: &AvroSchema, refs: &NamesRef) -> Result<Self::Out>;
223
224    fn on_map(
225        self,
226        value_type: &DataType,
227        avro_value_schema: &AvroSchema,
228        refs: &NamesRef,
229    ) -> Result<Self::Out>;
230
231    fn handle_nullable_union(out: Self::Out, opt_idx: OptIdx) -> Result<Self::Out>;
232}
233
234impl MaybeData for () {
235    type Out = ();
236
237    fn on_base(self, _: impl FnOnce(ScalarRefImpl<'_>) -> Result<Value>) -> Result<Self::Out> {
238        Ok(self)
239    }
240
241    fn on_struct(self, st: &StructType, avro: &AvroSchema, refs: &NamesRef) -> Result<Self::Out> {
242        validate_fields(st.iter(), avro, refs)
243    }
244
245    fn on_list(self, elem: &DataType, avro: &AvroSchema, refs: &NamesRef) -> Result<Self::Out> {
246        on_field(elem, (), avro, refs)
247    }
248
249    fn on_map(self, elem: &DataType, avro: &AvroSchema, refs: &NamesRef) -> Result<Self::Out> {
250        on_field(elem, (), avro, refs)
251    }
252
253    fn handle_nullable_union(out: Self::Out, _: OptIdx) -> Result<Self::Out> {
254        Ok(out)
255    }
256}
257
258impl MaybeData for DatumRef<'_> {
259    type Out = Value;
260
261    fn on_base(self, f: impl FnOnce(ScalarRefImpl<'_>) -> Result<Value>) -> Result<Self::Out> {
262        match self {
263            Some(s) => f(s),
264            None => Ok(Value::Null),
265        }
266    }
267
268    fn on_struct(self, st: &StructType, avro: &AvroSchema, refs: &NamesRef) -> Result<Self::Out> {
269        let d = match self {
270            Some(s) => s.into_struct(),
271            None => return Ok(Value::Null),
272        };
273        let record = encode_fields(st.iter().zip_eq_debug(d.iter_fields_ref()), avro, refs)?;
274        Ok(record.into())
275    }
276
277    fn on_list(self, elem: &DataType, avro: &AvroSchema, refs: &NamesRef) -> Result<Self::Out> {
278        let d = match self {
279            Some(s) => s.into_list(),
280            None => return Ok(Value::Null),
281        };
282        let vs = d
283            .iter()
284            .map(|d| on_field(elem, d, avro, refs))
285            .try_collect()?;
286        Ok(Value::Array(vs))
287    }
288
289    fn on_map(self, elem: &DataType, avro: &AvroSchema, refs: &NamesRef) -> Result<Self::Out> {
290        let d = match self {
291            Some(s) => s.into_map(),
292            None => return Ok(Value::Null),
293        };
294        let vs = d
295            .iter()
296            .map(|(k, v)| {
297                let k = k.into_utf8().to_owned();
298                let v = on_field(elem, v, avro, refs)?;
299                Ok((k, v))
300            })
301            .try_collect()?;
302        Ok(Value::Map(vs))
303    }
304
305    fn handle_nullable_union(out: Self::Out, opt_idx: OptIdx) -> Result<Self::Out> {
306        use OptIdx::*;
307
308        match out == Value::Null {
309            true => {
310                let ni = match opt_idx {
311                    NotUnion | Single => {
312                        return Err(FieldEncodeError::new("found null but required"));
313                    }
314                    NullLeft => 0,
315                    NullRight => 1,
316                };
317                Ok(Value::Union(ni, out.into()))
318            }
319            false => {
320                let vi = match opt_idx {
321                    NotUnion => return Ok(out),
322                    NullLeft => 1,
323                    Single | NullRight => 0,
324                };
325                Ok(Value::Union(vi, out.into()))
326            }
327        }
328    }
329}
330
331fn validate_fields<'rw>(
332    rw_fields: impl Iterator<Item = (&'rw str, &'rw DataType)>,
333    avro: &AvroSchema,
334    refs: &NamesRef,
335) -> Result<()> {
336    let avro = refs.lookup(avro);
337    let AvroSchema::Record(RecordSchema { fields, lookup, .. }) = avro else {
338        return Err(FieldEncodeError::new(format!(
339            "expect avro record but got {}",
340            avro.canonical_form(),
341        )));
342    };
343    let mut present = vec![false; fields.len()];
344    for (name, t) in rw_fields {
345        let Some(&idx) = lookup.get(name) else {
346            return Err(FieldEncodeError::new("field not in avro").with_name(name));
347        };
348        present[idx] = true;
349        let avro_field = &fields[idx];
350        on_field(t, (), &avro_field.schema, refs).map_err(|e| e.with_name(name))?;
351    }
352    for (p, avro_field) in present.into_iter().zip_eq_fast(fields) {
353        if p {
354            continue;
355        }
356        if !avro_field.is_nullable() {
357            return Err(
358                FieldEncodeError::new("field not present but required").with_name(&avro_field.name)
359            );
360        }
361    }
362    Ok(())
363}
364
365fn encode_fields<'avro, 'rw>(
366    fields_with_datums: impl Iterator<Item = ((&'rw str, &'rw DataType), DatumRef<'rw>)>,
367    schema: &'avro AvroSchema,
368    refs: &'avro NamesRef,
369) -> Result<Record<'avro>> {
370    let schema = refs.lookup(schema);
371    let mut record = Record::new(schema).unwrap();
372    let AvroSchema::Record(RecordSchema { fields, lookup, .. }) = schema else {
373        unreachable!()
374    };
375    let mut present = vec![false; fields.len()];
376    for ((name, t), d) in fields_with_datums {
377        let idx = lookup[name];
378        present[idx] = true;
379        let avro_field = &fields[idx];
380        let value = on_field(t, d, &avro_field.schema, refs).map_err(|e| e.with_name(name))?;
381        record.put(name, value);
382    }
383    // Unfortunately, the upstream `apache_avro` does not handle missing fields as nullable correctly.
384    // The correct encoding is `Value::Union(null_index, Value::Null)` but it simply writes `Value::Null`.
385    // See [`tests::test_encode_avro_lib_bug`].
386    for (p, avro_field) in present.into_iter().zip_eq_fast(fields) {
387        if p {
388            continue;
389        }
390        let AvroSchema::Union(u) = &avro_field.schema else {
391            unreachable!()
392        };
393        // We could have saved null index of each field during [`validate_fields`] to avoid repeated lookup.
394        // But in most cases it is the 0th.
395        // Alternatively, we can simplify by enforcing the best practice of `null at 0th`.
396        let ni = u
397            .variants()
398            .iter()
399            .position(|a| a == &AvroSchema::Null)
400            .unwrap();
401        record.put(
402            &avro_field.name,
403            Value::Union(ni.try_into().unwrap(), Value::Null.into()),
404        );
405    }
406    Ok(record)
407}
408
409/// Handles both `validate` (without actual data) and `encode`.
410/// See [`MaybeData`] for more info.
411fn on_field<D: MaybeData>(
412    data_type: &DataType,
413    maybe: D,
414    expected: &AvroSchema,
415    refs: &NamesRef,
416) -> Result<D::Out> {
417    use risingwave_common::types::Interval;
418
419    let no_match_err = || {
420        Err(FieldEncodeError::new(format!(
421            "cannot encode {} column as {} field",
422            data_type,
423            expected.canonical_form()
424        )))
425    };
426
427    // For now, we only support optional single type, rather than general union.
428    // For example, how do we encode int16 into avro `["int", "long"]`?
429    let (inner, opt_idx) = match expected {
430        AvroSchema::Union(union) => match union.variants() {
431            [] => return no_match_err(),
432            [one] => (one, OptIdx::Single),
433            [AvroSchema::Null, r] => (r, OptIdx::NullLeft),
434            [l, AvroSchema::Null] => (l, OptIdx::NullRight),
435            _ => return no_match_err(),
436        },
437        _ => (expected, OptIdx::NotUnion),
438    };
439
440    let inner = refs.lookup(inner);
441
442    let value = match &data_type {
443        // Group A: perfect match between RisingWave types and Avro types
444        DataType::Boolean => match inner {
445            AvroSchema::Boolean => maybe.on_base(|s| Ok(Value::Boolean(s.into_bool())))?,
446            _ => return no_match_err(),
447        },
448        DataType::Varchar => match inner {
449            AvroSchema::String => maybe.on_base(|s| Ok(Value::String(s.into_utf8().into())))?,
450            _ => return no_match_err(),
451        },
452        DataType::Bytea => match inner {
453            AvroSchema::Bytes => maybe.on_base(|s| Ok(Value::Bytes(s.into_bytea().into())))?,
454            _ => return no_match_err(),
455        },
456        DataType::Float32 => match inner {
457            AvroSchema::Float => maybe.on_base(|s| Ok(Value::Float(s.into_float32().into())))?,
458            _ => return no_match_err(),
459        },
460        DataType::Float64 => match inner {
461            AvroSchema::Double => maybe.on_base(|s| Ok(Value::Double(s.into_float64().into())))?,
462            _ => return no_match_err(),
463        },
464        DataType::Int32 => match inner {
465            AvroSchema::Int => maybe.on_base(|s| Ok(Value::Int(s.into_int32())))?,
466            _ => return no_match_err(),
467        },
468        DataType::Int64 => match inner {
469            AvroSchema::Long => maybe.on_base(|s| Ok(Value::Long(s.into_int64())))?,
470            _ => return no_match_err(),
471        },
472        DataType::Serial => match inner {
473            AvroSchema::Long => maybe.on_base(|s| Ok(Value::Long(s.into_serial().into_inner())))?,
474            _ => return no_match_err(),
475        },
476        DataType::Struct(st) => match inner {
477            AvroSchema::Record { .. } => maybe.on_struct(st, inner, refs)?,
478            _ => return no_match_err(),
479        },
480        DataType::List(elem) => match inner {
481            AvroSchema::Array(avro_elem) => maybe.on_list(elem, avro_elem, refs)?,
482            _ => return no_match_err(),
483        },
484        DataType::Map(m) => {
485            if *m.key() != DataType::Varchar {
486                return no_match_err();
487            }
488            match inner {
489                AvroSchema::Map(avro_value_type) => {
490                    maybe.on_map(m.value(), avro_value_type, refs)?
491                }
492                _ => return no_match_err(),
493            }
494        }
495
496        // Group B: match between RisingWave types and Avro logical types
497        DataType::Timestamptz => match inner {
498            AvroSchema::TimestampMicros => maybe.on_base(|s| {
499                Ok(Value::TimestampMicros(
500                    s.into_timestamptz().timestamp_micros(),
501                ))
502            })?,
503            AvroSchema::TimestampMillis => maybe.on_base(|s| {
504                Ok(Value::TimestampMillis(
505                    s.into_timestamptz().timestamp_millis(),
506                ))
507            })?,
508            _ => return no_match_err(),
509        },
510        DataType::Timestamp => return no_match_err(),
511        DataType::Date => match inner {
512            AvroSchema::Date => {
513                maybe.on_base(|s| Ok(Value::Date(s.into_date().get_nums_days_unix_epoch())))?
514            }
515            _ => return no_match_err(),
516        },
517        DataType::Time => match inner {
518            AvroSchema::TimeMicros => {
519                maybe.on_base(|s| Ok(Value::TimeMicros(Interval::from(s.into_time()).usecs())))?
520            }
521            AvroSchema::TimeMillis => maybe.on_base(|s| {
522                Ok(Value::TimeMillis(
523                    (Interval::from(s.into_time()).usecs() / 1000)
524                        .try_into()
525                        .unwrap(),
526                ))
527            })?,
528            _ => return no_match_err(),
529        },
530        DataType::Interval => match inner {
531            AvroSchema::Duration => maybe.on_base(|s| {
532                use apache_avro::{Days, Duration, Millis, Months};
533                let iv = s.into_interval();
534
535                let overflow = |_| FieldEncodeError::new(format!("{iv} overflows avro duration"));
536
537                Ok(Value::Duration(Duration::new(
538                    Months::new(iv.months().try_into().map_err(overflow)?),
539                    Days::new(iv.days().try_into().map_err(overflow)?),
540                    Millis::new((iv.usecs() / 1000).try_into().map_err(overflow)?),
541                )))
542            })?,
543            _ => return no_match_err(),
544        },
545        // Group C: experimental
546        DataType::Int16 => match inner {
547            AvroSchema::Int => maybe.on_base(|s| Ok(Value::Int(s.into_int16() as i32)))?,
548            _ => return no_match_err(),
549        },
550        DataType::Decimal => match inner {
551            AvroSchema::Decimal(decimal_schema) => {
552                maybe.on_base(|s| {
553                    match s.into_decimal() {
554                        risingwave_common::types::Decimal::Normalized(decimal) => {
555                            // convert to bigint with scale
556                            // rescale the rust_decimal to the scale of the avro decimal
557                            //
558                            // From bigdecimal::BigDecimal::with_scale:
559                            // If the new_scale is lower than the current value (indicating a larger
560                            // power of 10), digits will be dropped (as precision is lower)
561                            let signed_bigint_bytes =
562                                rust_decimal_to_scaled_bigint(decimal, decimal_schema.scale)
563                                    .map_err(FieldEncodeError::new)?;
564                            Ok(Value::Decimal(apache_avro::Decimal::from(
565                                &signed_bigint_bytes,
566                            )))
567                        }
568                        d @ risingwave_common::types::Decimal::NaN
569                        | d @ risingwave_common::types::Decimal::NegativeInf
570                        | d @ risingwave_common::types::Decimal::PositiveInf => {
571                            Err(FieldEncodeError::new(format!(
572                                "Avro Decimal does not support NaN or Inf, but got {}",
573                                d
574                            )))
575                        }
576                    }
577                })?
578            }
579            _ => return no_match_err(),
580        },
581        DataType::Jsonb => match inner {
582            AvroSchema::String => {
583                maybe.on_base(|s| Ok(Value::String(s.into_jsonb().to_string())))?
584            }
585            _ => return no_match_err(),
586        },
587        DataType::Vector(_) => match inner {
588            AvroSchema::Array(avro_elem) => maybe.on_list(&VECTOR_ITEM_TYPE, avro_elem, refs)?,
589            _ => return no_match_err(),
590        },
591        // Group D: unsupported
592        DataType::Int256 => {
593            return no_match_err();
594        }
595    };
596
597    D::handle_nullable_union(value, opt_idx)
598}
599
600#[cfg(test)]
601mod tests {
602    use std::collections::HashMap;
603    use std::str::FromStr;
604
605    use expect_test::expect;
606    use itertools::Itertools;
607    use risingwave_common::array::{ArrayBuilder, MapArrayBuilder};
608    use risingwave_common::catalog::Field;
609    use risingwave_common::row::OwnedRow;
610    use risingwave_common::types::{
611        Date, Datum, Interval, JsonbVal, ListValue, MapType, MapValue, Scalar, ScalarImpl,
612        StructValue, Time, Timestamptz, ToDatumRef,
613    };
614
615    use super::*;
616
617    #[track_caller]
618    fn test_ok(rw_type: &DataType, rw_datum: Datum, avro_type: &str, expected: Value) {
619        let avro_schema = AvroSchema::parse_str(avro_type).unwrap();
620        let refs = NamesRef::new(&avro_schema).unwrap();
621        let actual = on_field(rw_type, rw_datum.to_datum_ref(), &avro_schema, &refs).unwrap();
622        assert_eq!(actual, expected);
623    }
624
625    #[track_caller]
626    fn test_err<D: MaybeData>(t: &DataType, d: D, avro: &str, expected: &str)
627    where
628        D::Out: std::fmt::Debug,
629    {
630        let avro_schema = AvroSchema::parse_str(avro).unwrap();
631        let refs = NamesRef::new(&avro_schema).unwrap();
632        let err = on_field(t, d, &avro_schema, &refs).unwrap_err();
633        assert_eq!(err.to_string(), expected);
634    }
635
636    #[track_caller]
637    fn test_v2(rw_type: &str, rw_scalar: &str, avro_type: &str, expected: expect_test::Expect) {
638        let avro_schema = AvroSchema::parse_str(avro_type).unwrap();
639        let refs = NamesRef::new(&avro_schema).unwrap();
640        let rw_type = DataType::from_str(rw_type).unwrap();
641        let rw_datum = ScalarImpl::from_text_for_test(rw_scalar, &rw_type).unwrap();
642
643        if let Err(validate_err) = on_field(&rw_type, (), &avro_schema, &refs) {
644            expected.assert_debug_eq(&validate_err);
645            return;
646        }
647        let actual = on_field(&rw_type, Some(rw_datum).to_datum_ref(), &avro_schema, &refs);
648        match actual {
649            Ok(v) => expected.assert_eq(&print_avro_value(&v)),
650            Err(e) => expected.assert_debug_eq(&e),
651        }
652    }
653
654    fn print_avro_value(v: &Value) -> String {
655        match v {
656            Value::Map(m) => {
657                let mut res = "Map({".to_owned();
658                for (k, v) in m.iter().sorted_by_key(|x| x.0) {
659                    res.push_str(&format!("{}: {}, ", k, print_avro_value(v)));
660                }
661                res.push_str("})");
662                res
663            }
664            _ => format!("{v:?}"),
665        }
666    }
667
668    #[test]
669    fn test_encode_v2() {
670        test_v2(
671            "boolean",
672            "false",
673            r#""int""#,
674            expect![[r#"
675                FieldEncodeError {
676                    message: "cannot encode boolean column as \"int\" field",
677                    rev_path: [],
678                }
679            "#]],
680        );
681        test_v2("boolean", "true", r#""boolean""#, expect!["Boolean(true)"]);
682
683        test_v2(
684            "map(varchar,varchar)",
685            "{1:1,2:2,3:3}",
686            r#"{"type": "map","values": "string"}"#,
687            expect![[r#"Map({1: String("1"), 2: String("2"), 3: String("3"), })"#]],
688        );
689
690        test_v2(
691            "map(varchar,varchar)",
692            "{1:1,2:NULL,3:3}",
693            r#"{"type": "map","values": "string"}"#,
694            expect![[r#"
695                FieldEncodeError {
696                    message: "found null but required",
697                    rev_path: [],
698                }
699            "#]],
700        );
701
702        test_v2(
703            "map(varchar,varchar)",
704            "{1:1,2:NULL,3:3}",
705            r#"{"type": "map","values": ["null", "string"]}"#,
706            expect![[
707                r#"Map({1: Union(1, String("1")), 2: Union(0, Null), 3: Union(1, String("3")), })"#
708            ]],
709        );
710
711        test_v2(
712            "map(int,varchar)",
713            "{1:1,2:NULL,3:3}",
714            r#"{"type": "map","values": ["null", "string"]}"#,
715            expect![[r#"
716                FieldEncodeError {
717                    message: "cannot encode map(integer,character varying) column as {\"type\":\"map\",\"values\":[\"null\",\"string\"]} field",
718                    rev_path: [],
719                }
720            "#]],
721        );
722    }
723
724    #[test]
725    fn test_encode_avro_ok() {
726        test_ok(
727            &DataType::Boolean,
728            Some(ScalarImpl::Bool(false)),
729            r#""boolean""#,
730            Value::Boolean(false),
731        );
732
733        test_ok(
734            &DataType::Varchar,
735            Some(ScalarImpl::Utf8("RisingWave".into())),
736            r#""string""#,
737            Value::String("RisingWave".into()),
738        );
739
740        test_ok(
741            &DataType::Bytea,
742            Some(ScalarImpl::Bytea([0xbe, 0xef].into())),
743            r#""bytes""#,
744            Value::Bytes([0xbe, 0xef].into()),
745        );
746
747        test_ok(
748            &DataType::Float32,
749            Some(ScalarImpl::Float32(3.5f32.into())),
750            r#""float""#,
751            Value::Float(3.5f32),
752        );
753
754        test_ok(
755            &DataType::Float64,
756            Some(ScalarImpl::Float64(4.25f64.into())),
757            r#""double""#,
758            Value::Double(4.25f64),
759        );
760
761        test_ok(
762            &DataType::Int32,
763            Some(ScalarImpl::Int32(16)),
764            r#""int""#,
765            Value::Int(16),
766        );
767
768        test_ok(
769            &DataType::Int64,
770            Some(ScalarImpl::Int64(i64::MAX)),
771            r#""long""#,
772            Value::Long(i64::MAX),
773        );
774
775        test_ok(
776            &DataType::Serial,
777            Some(ScalarImpl::Serial(i64::MAX.into())),
778            r#""long""#,
779            Value::Long(i64::MAX),
780        );
781
782        let tstz = "2018-01-26T18:30:09.453Z".parse().unwrap();
783        test_ok(
784            &DataType::Timestamptz,
785            Some(ScalarImpl::Timestamptz(tstz)),
786            r#"{"type": "long", "logicalType": "timestamp-micros"}"#,
787            Value::TimestampMicros(tstz.timestamp_micros()),
788        );
789        test_ok(
790            &DataType::Timestamptz,
791            Some(ScalarImpl::Timestamptz(tstz)),
792            r#"{"type": "long", "logicalType": "timestamp-millis"}"#,
793            Value::TimestampMillis(tstz.timestamp_millis()),
794        );
795
796        test_ok(
797            &DataType::Date,
798            Some(ScalarImpl::Date(Date::from_ymd_uncheck(1970, 1, 2))),
799            r#"{"type": "int", "logicalType": "date"}"#,
800            Value::Date(1),
801        );
802
803        let tm = Time::from_num_seconds_from_midnight_uncheck(1000, 0);
804        test_ok(
805            &DataType::Time,
806            Some(ScalarImpl::Time(tm)),
807            r#"{"type": "long", "logicalType": "time-micros"}"#,
808            Value::TimeMicros(1000 * 1_000_000),
809        );
810        test_ok(
811            &DataType::Time,
812            Some(ScalarImpl::Time(tm)),
813            r#"{"type": "int", "logicalType": "time-millis"}"#,
814            Value::TimeMillis(1000 * 1000),
815        );
816
817        test_ok(
818            &DataType::Int16,
819            Some(ScalarImpl::Int16(i16::MAX)),
820            r#""int""#,
821            Value::Int(i16::MAX as i32),
822        );
823
824        test_ok(
825            &DataType::Int16,
826            Some(ScalarImpl::Int16(i16::MIN)),
827            r#""int""#,
828            Value::Int(i16::MIN as i32),
829        );
830
831        test_ok(
832            &DataType::Jsonb,
833            Some(ScalarImpl::Jsonb(
834                JsonbVal::from_str(r#"{"a": 1}"#).unwrap(),
835            )),
836            r#""string""#,
837            Value::String(r#"{"a": 1}"#.into()),
838        );
839
840        test_ok(
841            &DataType::Interval,
842            Some(ScalarImpl::Interval(Interval::from_month_day_usec(
843                13, 2, 1000000,
844            ))),
845            r#"{"type": "fixed", "name": "Duration", "size": 12, "logicalType": "duration"}"#,
846            Value::Duration(apache_avro::Duration::new(
847                apache_avro::Months::new(13),
848                apache_avro::Days::new(2),
849                apache_avro::Millis::new(1000),
850            )),
851        );
852
853        let mut inner_map_array_builder = MapArrayBuilder::with_type(
854            2,
855            DataType::Map(MapType::from_kv(DataType::Varchar, DataType::Int32)),
856        );
857        inner_map_array_builder.append(Some(
858            MapValue::try_from_kv(
859                ListValue::from_iter(["a", "b"]),
860                ListValue::from_iter([1, 2]),
861            )
862            .unwrap()
863            .as_scalar_ref(),
864        ));
865        inner_map_array_builder.append(Some(
866            MapValue::try_from_kv(
867                ListValue::from_iter(["c", "d"]),
868                ListValue::from_iter([3, 4]),
869            )
870            .unwrap()
871            .as_scalar_ref(),
872        ));
873        let inner_map_array = inner_map_array_builder.finish();
874        test_ok(
875            &DataType::Map(MapType::from_kv(
876                DataType::Varchar,
877                DataType::Map(MapType::from_kv(DataType::Varchar, DataType::Int32)),
878            )),
879            Some(ScalarImpl::Map(
880                MapValue::try_from_kv(
881                    ListValue::from_iter(["k1", "k2"]),
882                    ListValue::new(inner_map_array.into()),
883                )
884                .unwrap(),
885            )),
886            r#"{"type": "map","values": {"type": "map","values": "int"}}"#,
887            Value::Map(HashMap::from_iter([
888                (
889                    "k1".into(),
890                    Value::Map(HashMap::from_iter([
891                        ("a".into(), Value::Int(1)),
892                        ("b".into(), Value::Int(2)),
893                    ])),
894                ),
895                (
896                    "k2".into(),
897                    Value::Map(HashMap::from_iter([
898                        ("c".into(), Value::Int(3)),
899                        ("d".into(), Value::Int(4)),
900                    ])),
901                ),
902            ])),
903        );
904
905        test_ok(
906            &DataType::Struct(StructType::new(vec![
907                (
908                    "p",
909                    DataType::Struct(StructType::new(vec![
910                        ("x", DataType::Int32),
911                        ("y", DataType::Int32),
912                    ])),
913                ),
914                (
915                    "q",
916                    DataType::Struct(StructType::new(vec![
917                        ("x", DataType::Int32),
918                        ("y", DataType::Int32),
919                    ])),
920                ),
921            ])),
922            Some(ScalarImpl::Struct(StructValue::new(vec![
923                Some(ScalarImpl::Struct(StructValue::new(vec![
924                    Some(ScalarImpl::Int32(-2)),
925                    Some(ScalarImpl::Int32(-1)),
926                ]))),
927                Some(ScalarImpl::Struct(StructValue::new(vec![
928                    Some(ScalarImpl::Int32(2)),
929                    Some(ScalarImpl::Int32(1)),
930                ]))),
931            ]))),
932            r#"{
933                "type": "record",
934                "name": "Segment",
935                "fields": [
936                    {
937                        "name": "p",
938                        "type": {
939                            "type": "record",
940                            "name": "Point",
941                            "fields": [
942                                {
943                                    "name": "x",
944                                    "type": "int"
945                                },
946                                {
947                                    "name": "y",
948                                    "type": "int"
949                                }
950                            ]
951                        }
952                    },
953                    {
954                        "name": "q",
955                        "type": "Point"
956                    }
957                ]
958            }"#,
959            Value::Record(vec![
960                (
961                    "p".to_owned(),
962                    Value::Record(vec![
963                        ("x".to_owned(), Value::Int(-2)),
964                        ("y".to_owned(), Value::Int(-1)),
965                    ]),
966                ),
967                (
968                    "q".to_owned(),
969                    Value::Record(vec![
970                        ("x".to_owned(), Value::Int(2)),
971                        ("y".to_owned(), Value::Int(1)),
972                    ]),
973                ),
974            ]),
975        );
976
977        // Test complex JSON with nested structures - using serde_json::Value comparison
978        let complex_json = r#"{
979            "person": {
980                "name": "John Doe",
981                "age": 30,
982                "address": {
983                    "street": "123 Main St.",
984                    "city": "New York",
985                    "coordinates": [40.7128, -74.0060]
986                },
987                "contacts": [
988                    {"type": "email", "value": "john@example.com"},
989                    {"type": "phone", "value": "+1-555-123-4567"}
990                ],
991                "active": true,
992                "preferences": {
993                    "notifications": true,
994                    "theme": "dark",
995                    "languages": ["en", "es"],
996                    "lastLogin": null
997                },
998                "tags": ["premium", "verified"],
999                "unicode_test": "Hello, 世界! 🌍"
1000            }
1001        }"#;
1002
1003        let input_json = JsonbVal::from_str(complex_json).unwrap();
1004        let result = on_field(
1005            &DataType::Jsonb,
1006            Some(ScalarImpl::Jsonb(input_json.clone())).to_datum_ref(),
1007            &AvroSchema::parse_str(r#""string""#).unwrap(),
1008            &NamesRef::new(&AvroSchema::parse_str(r#""string""#).unwrap()).unwrap(),
1009        )
1010        .unwrap();
1011
1012        // Compare as parsed JSON values to handle key order randomness
1013        if let Value::String(result_str) = result {
1014            let expected_json: serde_json::Value = serde_json::from_str(complex_json).unwrap();
1015            let actual_json: serde_json::Value = serde_json::from_str(&result_str).unwrap();
1016            assert_eq!(
1017                expected_json, actual_json,
1018                "JSON values should be equivalent regardless of key order"
1019            );
1020        } else {
1021            panic!("Expected String value");
1022        };
1023    }
1024
1025    #[test]
1026    fn test_encode_avro_err() {
1027        test_err(
1028            &DataType::Interval,
1029            Some(ScalarRefImpl::Interval(Interval::from_month_day_usec(
1030                -1,
1031                -1,
1032                i64::MAX,
1033            ))),
1034            r#"{"type": "fixed", "name": "Duration", "size": 12, "logicalType": "duration"}"#,
1035            "encode '' error: -1 mons -1 days +2562047788:00:54.775807 overflows avro duration",
1036        );
1037
1038        let avro_schema = AvroSchema::parse_str(
1039            r#"{"type": "record", "name": "Root", "fields": [
1040                {"name": "f0", "type": "int"}
1041            ]}"#,
1042        )
1043        .unwrap();
1044        let mut record = Record::new(&avro_schema).unwrap();
1045        record.put("f0", Value::String("2".into()));
1046        let res: SinkResult<Vec<u8>> = AvroEncoded {
1047            value: Value::from(record),
1048            schema: Arc::new(avro_schema),
1049            header: AvroHeader::ConfluentSchemaRegistry(42),
1050        }
1051        .ser_to();
1052        assert_eq!(
1053            res.unwrap_err().to_string(),
1054            "Encode error: Value does not match schema"
1055        );
1056    }
1057
1058    #[test]
1059    fn test_encode_avro_record() {
1060        let avro_schema = AvroSchema::parse_str(
1061            r#"{
1062                "type": "record",
1063                "name": "Root",
1064                "fields": [
1065                    {"name": "req", "type": "int"},
1066                    {"name": "opt", "type": ["null", "long"]}
1067                ]
1068            }"#,
1069        )
1070        .unwrap();
1071        let avro_schema = Arc::new(avro_schema);
1072        let header = AvroHeader::None;
1073
1074        let schema = Schema::new(vec![
1075            Field::with_name(DataType::Int64, "opt"),
1076            Field::with_name(DataType::Int32, "req"),
1077        ]);
1078        let row = OwnedRow::new(vec![
1079            Some(ScalarImpl::Int64(31)),
1080            Some(ScalarImpl::Int32(15)),
1081        ]);
1082        let encoder = AvroEncoder::new(schema, None, avro_schema.clone(), header).unwrap();
1083        let actual = encoder.encode(row).unwrap();
1084        assert_eq!(
1085            actual.value,
1086            Value::Record(vec![
1087                ("req".into(), Value::Int(15)),
1088                ("opt".into(), Value::Union(1, Value::Long(31).into())),
1089            ])
1090        );
1091
1092        let schema = Schema::new(vec![Field::with_name(DataType::Int32, "req")]);
1093        let row = OwnedRow::new(vec![Some(ScalarImpl::Int32(15))]);
1094        let encoder = AvroEncoder::new(schema, None, avro_schema.clone(), header).unwrap();
1095        let actual = encoder.encode(row).unwrap();
1096        assert_eq!(
1097            actual.value,
1098            Value::Record(vec![
1099                ("req".into(), Value::Int(15)),
1100                ("opt".into(), Value::Union(0, Value::Null.into())),
1101            ])
1102        );
1103
1104        let schema = Schema::new(vec![Field::with_name(DataType::Int64, "opt")]);
1105        let Err(err) = AvroEncoder::new(schema, None, avro_schema.clone(), header) else {
1106            panic!()
1107        };
1108        assert_eq!(
1109            err.to_string(),
1110            "Encode error: encode 'req' error: field not present but required"
1111        );
1112
1113        let schema = Schema::new(vec![
1114            Field::with_name(DataType::Int64, "opt"),
1115            Field::with_name(DataType::Int32, "req"),
1116            Field::with_name(DataType::Varchar, "extra"),
1117        ]);
1118        let Err(err) = AvroEncoder::new(schema, None, avro_schema.clone(), header) else {
1119            panic!()
1120        };
1121        assert_eq!(
1122            err.to_string(),
1123            "Encode error: encode 'extra' error: field not in avro"
1124        );
1125
1126        let avro_schema = AvroSchema::parse_str(r#"["null", "long"]"#).unwrap();
1127        let schema = Schema::new(vec![Field::with_name(DataType::Int64, "opt")]);
1128        let Err(err) = AvroEncoder::new(schema, None, avro_schema.into(), header) else {
1129            panic!()
1130        };
1131        assert_eq!(
1132            err.to_string(),
1133            r#"Encode error: encode '' error: expect avro record but got ["null","long"]"#
1134        );
1135
1136        test_err(
1137            &DataType::Struct(StructType::new(vec![("f0", DataType::Boolean)])),
1138            (),
1139            r#"{"type": "record", "name": "T", "fields": [{"name": "f0", "type": "int"}]}"#,
1140            "encode 'f0' error: cannot encode boolean column as \"int\" field",
1141        );
1142    }
1143
1144    #[test]
1145    fn test_encode_avro_array() {
1146        let avro_schema = r#"{
1147            "type": "array",
1148            "items": "int"
1149        }"#;
1150
1151        test_ok(
1152            &DataType::List(DataType::Int32.into()),
1153            Some(ScalarImpl::List(ListValue::from_iter([4, 5]))),
1154            avro_schema,
1155            Value::Array(vec![Value::Int(4), Value::Int(5)]),
1156        );
1157
1158        test_err(
1159            &DataType::List(DataType::Int32.into()),
1160            Some(ScalarImpl::List(ListValue::from_iter([Some(4), None]))).to_datum_ref(),
1161            avro_schema,
1162            "encode '' error: found null but required",
1163        );
1164
1165        test_ok(
1166            &DataType::List(DataType::Int32.into()),
1167            Some(ScalarImpl::List(ListValue::from_iter([Some(4), None]))),
1168            r#"{
1169                "type": "array",
1170                "items": ["null", "int"]
1171            }"#,
1172            Value::Array(vec![
1173                Value::Union(1, Value::Int(4).into()),
1174                Value::Union(0, Value::Null.into()),
1175            ]),
1176        );
1177
1178        test_ok(
1179            &DataType::List(DataType::List(DataType::Int32.into()).into()),
1180            Some(ScalarImpl::List(ListValue::from_iter([
1181                ListValue::from_iter([26, 29]),
1182                ListValue::from_iter([46, 49]),
1183            ]))),
1184            r#"{
1185                "type": "array",
1186                "items": {
1187                    "type": "array",
1188                    "items": "int"
1189                }
1190            }"#,
1191            Value::Array(vec![
1192                Value::Array(vec![Value::Int(26), Value::Int(29)]),
1193                Value::Array(vec![Value::Int(46), Value::Int(49)]),
1194            ]),
1195        );
1196
1197        test_err(
1198            &DataType::List(DataType::Boolean.into()),
1199            (),
1200            r#"{"type": "array", "items": "int"}"#,
1201            "encode '' error: cannot encode boolean column as \"int\" field",
1202        );
1203    }
1204
1205    #[test]
1206    fn test_encode_avro_union() {
1207        let t = &DataType::Timestamptz;
1208        let datum = Some(ScalarImpl::Timestamptz(Timestamptz::from_micros(1500)));
1209        let opt_micros = r#"["null", {"type": "long", "logicalType": "timestamp-micros"}]"#;
1210        let opt_millis = r#"["null", {"type": "long", "logicalType": "timestamp-millis"}]"#;
1211        let both = r#"[{"type": "long", "logicalType": "timestamp-millis"}, {"type": "long", "logicalType": "timestamp-micros"}]"#;
1212        let empty = "[]";
1213        let one = r#"[{"type": "long", "logicalType": "timestamp-millis"}]"#;
1214        let right = r#"[{"type": "long", "logicalType": "timestamp-millis"}, "null"]"#;
1215
1216        test_ok(
1217            t,
1218            datum.clone(),
1219            opt_micros,
1220            Value::Union(1, Value::TimestampMicros(1500).into()),
1221        );
1222        test_ok(t, None, opt_micros, Value::Union(0, Value::Null.into()));
1223        test_ok(
1224            t,
1225            datum.clone(),
1226            opt_millis,
1227            Value::Union(1, Value::TimestampMillis(1).into()),
1228        );
1229        test_ok(t, None, opt_millis, Value::Union(0, Value::Null.into()));
1230
1231        test_err(
1232            t,
1233            datum.to_datum_ref(),
1234            both,
1235            r#"encode '' error: cannot encode timestamp with time zone column as [{"type":"long","logicalType":"timestamp-millis"},{"type":"long","logicalType":"timestamp-micros"}] field"#,
1236        );
1237
1238        test_err(
1239            t,
1240            datum.to_datum_ref(),
1241            empty,
1242            "encode '' error: cannot encode timestamp with time zone column as [] field",
1243        );
1244
1245        test_ok(
1246            t,
1247            datum.clone(),
1248            one,
1249            Value::Union(0, Value::TimestampMillis(1).into()),
1250        );
1251        test_err(t, None, one, "encode '' error: found null but required");
1252
1253        test_ok(
1254            t,
1255            datum.clone(),
1256            right,
1257            Value::Union(0, Value::TimestampMillis(1).into()),
1258        );
1259        test_ok(t, None, right, Value::Union(1, Value::Null.into()));
1260    }
1261
1262    /// This just demonstrates bugs of the upstream [`apache_avro`], rather than our encoder.
1263    /// The encoder is not using these buggy calls and is already tested above.
1264    #[test]
1265    fn test_encode_avro_lib_bug() {
1266        use apache_avro::{Reader, Writer};
1267
1268        // a record with 2 optional int fields
1269        let avro_schema = AvroSchema::parse_str(
1270            r#"{
1271                "type": "record",
1272                "name": "Root",
1273                "fields": [
1274                    {
1275                        "name": "f0",
1276                        "type": ["null", "int"]
1277                    },
1278                    {
1279                        "name": "f1",
1280                        "type": ["null", "int"]
1281                    }
1282                ]
1283            }"#,
1284        )
1285        .unwrap();
1286
1287        let mut writer = Writer::new(&avro_schema, Vec::new());
1288        let mut record = Record::new(writer.schema()).unwrap();
1289        // f0 omitted, f1 = Int(3)
1290        record.put("f1", Value::Int(3));
1291        writer.append(record).unwrap();
1292        let encoded = writer.into_inner().unwrap();
1293        // writing produced no error, but read fails
1294        let reader = Reader::new(encoded.as_slice()).unwrap();
1295        for value in reader {
1296            assert_eq!(
1297                value.unwrap_err().to_string(),
1298                "Union index 3 out of bounds: 2"
1299            );
1300        }
1301
1302        let mut writer = Writer::new(&avro_schema, Vec::new());
1303        let mut record = Record::new(writer.schema()).unwrap();
1304        // f0 omitted, f1 = Union(1, Int(3))
1305        record.put("f1", Value::Union(1, Value::Int(3).into()));
1306        writer.append(record).unwrap();
1307        let encoded = writer.into_inner().unwrap();
1308        // writing produced no error, but read returns wrong value
1309        let reader = Reader::new(encoded.as_slice()).unwrap();
1310        for value in reader {
1311            assert_eq!(
1312                value.unwrap(),
1313                Value::Record(vec![
1314                    ("f0".into(), Value::Union(1, Value::Int(3).into())),
1315                    ("f1".into(), Value::Union(0, Value::Null.into())),
1316                ])
1317            );
1318        }
1319    }
1320}
risingwave_connector/sink/encoder/avro.rs

risingwave_connector/sink/encoder/
avro.rs
