apache-avro
A library for working with Apache Avro in Rust language.
Please check our documentation for examples, tutorials and API reference.
Apache Avro is a data serialization system which provides rich data structures and a compact, fast, binary data format.
All data in Avro is schematized, as in the following example:
{
"type": "record",
"name": "test",
"fields": [
{"name": "a", "type": "long", "default": 42},
{"name": "b", "type": "string"}
]
}
There are basically two ways of handling Avro data in Rust:
- as Avro-specialized data types based on an Avro schema;
- as generic Rust serde-compatible types implementing/deriving
Serialize
andDeserialize
;
apache-avro provides a way to read and write both these data representations easily and efficiently.
Installing the library
Add to your Cargo.toml
:
[]
= "x.y"
Or in case you want to leverage the Snappy codec:
[]
= "x.y"
= ["snappy"]
Or in case you want to leverage the Zstandard codec:
[]
= "x.y"
= ["zstandard"]
Or in case you want to leverage the Bzip2 codec:
[]
= "x.y"
= ["bzip"]
Or in case you want to leverage the Xz codec:
[]
= "x.y"
= ["xz"]
Upgrading to a newer minor version
The library is still in beta, so there might be backward-incompatible changes between minor versions. If you have troubles upgrading, check the version upgrade guide.
Defining a schema
An Avro data cannot exist without an Avro schema. Schemas must be used while writing and can be used while reading and they carry the information regarding the type of data we are handling. Avro schemas are used for both schema validation and resolution of Avro data.
Avro schemas are defined in JSON format and can just be parsed out of a raw string:
use Schema;
let raw_schema = r#"
{
"type": "record",
"name": "test",
"fields": [
{"name": "a", "type": "long", "default": 42},
{"name": "b", "type": "string"}
]
}
"#;
// if the schema is not valid, this function will return an error
let schema = parse_str.unwrap;
// schemas can be printed for debugging
println!;
Additionally, a list of of definitions (which may depend on each other) can be given and all of them will be parsed into the corresponding schemas.
use Schema;
let raw_schema_1 = r#"{
"name": "A",
"type": "record",
"fields": [
{"name": "field_one", "type": "float"}
]
}"#;
// This definition depends on the definition of A above
let raw_schema_2 = r#"{
"name": "B",
"type": "record",
"fields": [
{"name": "field_one", "type": "A"}
]
}"#;
// if the schemas are not valid, this function will return an error
let schemas = parse_list.unwrap;
// schemas can be printed for debugging
println!;
N.B. It is important to note that the composition of schema definitions requires schemas with names. For this reason, only schemas of type Record, Enum, and Fixed should be input into this function.
The library provides also a programmatic interface to define schemas without encoding them in JSON (for advanced use), but we highly recommend the JSON interface. Please read the API reference in case you are interested.
For more information about schemas and what kind of information you can encapsulate in them, please refer to the appropriate section of the Avro Specification.
Writing data
Once we have defined a schema, we are ready to serialize data in Avro, validating them against the provided schema in the process. As mentioned before, there are two ways of handling Avro data in Rust.
NOTE: The library also provides a low-level interface for encoding a single datum in Avro
bytecode without generating markers and headers (for advanced use), but we highly recommend the
Writer
interface to be totally Avro-compatible. Please read the API reference in case you are
interested.
The avro way
Given that the schema we defined above is that of an Avro Record, we are going to use the associated type provided by the library to specify the data we want to serialize:
use Record;
use Writer;
#
// a writer needs a schema and something to write to
let mut writer = new;
// the Record type models our Record schema
let mut record = new.unwrap;
record.put;
record.put;
// schema validation happens here
writer.append.unwrap;
// this is how to get back the resulting avro bytecode
// this performs a flush operation to make sure data has been written, so it can fail
// you can also call `writer.flush()` yourself without consuming the writer
let encoded = writer.into_inner.unwrap;
The vast majority of the times, schemas tend to define a record as a top-level container
encapsulating all the values to convert as fields and providing documentation for them, but in
case we want to directly define an Avro value, the library offers that capability via the
Value
interface.
use Value;
let mut value = String;
The serde way
Given that the schema we defined above is an Avro Record, we can directly use a Rust struct
deriving Serialize
to model our data:
use Writer;
// a writer needs a schema and something to write to
let mut writer = new;
// the structure models our Record schema
let test = Test ;
// schema validation happens here
writer.append_ser.unwrap;
// this is how to get back the resulting avro bytecode
// this performs a flush operation to make sure data is written, so it can fail
// you can also call `writer.flush()` yourself without consuming the writer
let encoded = writer.into_inner;
The vast majority of the times, schemas tend to define a record as a top-level container
encapsulating all the values to convert as fields and providing documentation for them, but in
case we want to directly define an Avro value, any type implementing Serialize
should work.
let mut value = "foo".to_string;
Using codecs to compress data
Avro supports three different compression codecs when encoding data:
- Null: leaves data uncompressed;
- Deflate: writes the data block using the deflate algorithm as specified in RFC 1951, and typically implemented using the zlib library. Note that this format (unlike the "zlib format" in RFC 1950) does not have a checksum.
- Snappy: uses Google's Snappy compression library. Each
compressed block is followed by the 4-byte, big-endianCRC32 checksum of the uncompressed data in
the block. You must enable the
snappy
feature to use this codec. - Zstandard: uses Facebook's Zstandard compression library.
You must enable the
zstandard
feature to use this codec. - Bzip2: uses BZip2 compression library.
You must enable the
bzip
feature to use this codec. - Xz: uses xz2 compression library.
You must enable the
xz
feature to use this codec.
To specify a codec to use to compress data, just specify it while creating a Writer
:
use Writer;
use Codec;
#
let mut writer = with_codec;
Reading data
As far as reading Avro encoded data goes, we can just use the schema encoded with the data to read them. The library will do it automatically for us, as it already does for the compression codec:
use Reader;
#
// reader creation can fail in case the input to read from is not Avro-compatible or malformed
let reader = new.unwrap;
In case, instead, we want to specify a different (but compatible) reader schema from the schema the data has been written with, we can just do as the following:
use Schema;
use Reader;
#
let reader_raw_schema = r#"
{
"type": "record",
"name": "test",
"fields": [
{"name": "a", "type": "long", "default": 42},
{"name": "b", "type": "string"},
{"name": "c", "type": "long", "default": 43}
]
}
"#;
let reader_schema = parse_str.unwrap;
// reader creation can fail in case the input to read from is not Avro-compatible or malformed
let reader = with_schema.unwrap;
The library will also automatically perform schema resolution while reading the data.
For more information about schema compatibility and resolution, please refer to the Avro Specification.
As usual, there are two ways to handle Avro data in Rust, as you can see below.
NOTE: The library also provides a low-level interface for decoding a single datum in Avro
bytecode without markers and header (for advanced use), but we highly recommend the Reader
interface to leverage all Avro features. Please read the API reference in case you are
interested.
The avro way
We can just read directly instances of Value
out of the Reader
iterator:
use Reader;
#
let reader = new.unwrap;
// value is a Result of an Avro Value in case the read operation fails
for value in reader
The serde way
Alternatively, we can use a Rust type implementing Deserialize
and representing our schema to
read the data into:
use Reader;
use from_value;
let reader = new.unwrap;
// value is a Result in case the read operation fails
for value in reader
Putting everything together
The following is an example of how to combine everything showed so far and it is meant to be a quick reference of the library interface:
use ;
use ;
apache-avro
also supports the logical types listed in the Avro specification:
Decimal
using thenum_bigint
crate- UUID using the
uuid
crate - Date, Time (milli) as
i32
and Time (micro) asi64
- Timestamp (milli and micro) as
i64
- Duration as a custom type with
months
,days
andmillis
accessor methods each of which returns ani32
Note that the on-disk representation is identical to the underlying primitive/complex type.
Read and write logical types
use ;
use ToBigInt;
Calculate Avro schema fingerprint
This library supports calculating the following fingerprints:
- SHA-256
- MD5
- Rabin
An example of fingerprinting for the supported fingerprints:
use Rabin;
use ;
use Md5;
use Sha256;
Ill-formed data
In order to ease decoding, the Binary Encoding specification of Avro data requires some fields to have their length encoded alongside the data.
If encoded data passed to a Reader
has been ill-formed, it can happen that
the bytes meant to contain the length of data are bogus and could result
in extravagant memory allocation.
To shield users from ill-formed data, apache-avro
sets a limit (default: 512MB)
to any allocation it will perform when decoding data.
If you expect some of your data fields to be larger than this limit, be sure
to make use of the max_allocation_bytes
function before reading any data
(we leverage Rust's std::sync::Once
mechanism to initialize this value, if
any call to decode is made before a call to max_allocation_bytes
, the limit
will be 512MB throughout the lifetime of the program).
use max_allocation_bytes;
max_allocation_bytes; // 2GB
// ... happily decode large data
Check schemas compatibility
This library supports checking for schemas compatibility.
Examples of checking for compatibility:
- Compatible schemas
Explanation: an int array schema can be read by a long array schema- an int (32bit signed integer) fits into a long (64bit signed integer)
use ;
let writers_schema = parse_str.unwrap;
let readers_schema = parse_str.unwrap;
assert_eq!;
- Incompatible schemas (a long array schema cannot be read by an int array schema)
Explanation: a long array schema cannot be read by an int array schema- a long (64bit signed integer) does not fit into an int (32bit signed integer)
use ;
let writers_schema = parse_str.unwrap;
let readers_schema = parse_str.unwrap;
assert_eq!;
Minimal supported Rust version
1.60.0
License
This project is licensed under Apache License 2.0.
Contributing
Everyone is encouraged to contribute! You can contribute by forking the GitHub repo and making a pull request or opening an issue. All contributions will be licensed under Apache License 2.0.
Please consider adding documentation and tests!
If you introduce a backward-incompatible change, please consider adding instruction to migrate in the Migration Guide
If you modify the crate documentation in lib.rs
, run make readme
to sync the README file.