#serialization #migration #serde #flow


Simplifying migration for serde and zero-copy entities

4 stable releases

1.1.1 Mar 14, 2024
1.1.0 Mar 13, 2024
1.0.1 Mar 7, 2024

#365 in Encoding

Download history 221/week @ 2024-03-03 253/week @ 2024-03-10 61/week @ 2024-03-17 16/week @ 2024-03-31

150 downloads per month
Used in atomic-ops

MIT license

262 lines

Serde Flow - Compatibility Framework

build status

The library assists you in smoothly deserializing the earlier variants of your serialized information.


The Rust library that simplifies managing changes to serialized data formats during software development, enabling seamless file migration and maintaining version compatibility by supporting backward compatibility through versioning of serialized data.


  1. Versioning of serialize/deserialize entities
  2. Migration of serialized bytes
  3. Async migration
  4. Zerocopy deserialization
  5. Data Integrity Verification

Main Concepts

Serde Flow primarily consists of three major components:

  1. #[derive(Flow)]: To utilize Serde Flow, you must annotate your class with serde_flow::Flow. This annotation serves as a signal to the library that the class is eligible for data migration.
  2. #[flow(variant = N)]: Utilize this annotation to specify the version of the entity. Simply replace N with a u16 number that represents the version. This helps in managing different versions of your data structures efficiently.
    • variant = N - defines version of the struct with number(u16) N
    • file(option1, option2) or file - implements loading from file
      • blocking - (default) - normal blocking IO loading and deserialization
      • nonblocking - async IO loading and deserialization (it's possible to use blockin and nonblocking at the same time)
      • verify_write - verifies writted data by calculating checksum
    • zerocopy - Uses rkyv to perfome zerocopy deserialization.
    • bytes - Uses in memory migration without persising on the disk (just call encode()->Vec<u8> or decode(Vec<u8>)->T)
  3. #[variants(StructA, StructB, ...)] (Optional): This annotation is optional but highly recommended for comprehensive data migration management. Here, you list the structs that are essential for migrating into the struct highlighted with this annotation. To ensure, you need to implement From<VariantStruct> for all structs listed in #[variants(..)].

🛠️ Getting Started

serde_flow = "1.1.0"

Imagine you have a User struct that has evolved over time through versions UserV1 -> UserV2 -> User (current), while the previous versions UserV1 and UserV2 still exist elsewhere. To manage this scenario effectively, follow these steps:

  1. Versioning: Start by setting proper versioning from the beginning. The initial creation of the user should be annotated with #[flow(variant = 1)].
  2. Incremental Versioning: As you iterate and create subsequent versions, ensure to increment the version in the annotations, such as #[flow(variant = 2)] for the next version.
  3. Migration Preparation: When you're ready to migrate to a new version, add the #[variants(UserV1, UserV2)] annotation to the main User struct. It's essential to include all previous variants that you intend to migrate from.
  4. Implementation Scope: Variants must be implemented in the same file as the main variant to ensure proper management and accessibility during migration processes.

By adhering to these guidelines, you can effectively manage the evolution of your data structures while ensuring seamless migration across versions.

Setup - add attributes to your structs

use serde_flow::{Flow};
use serde::{Deserialize, Serialize};

// The last (variant 3) of the User
#[derive(Serialize, Deserialize, Flow)]
#[flow(variant = 3, file)]
#[variants(UserV1, UserV2)]
pub struct User {
    pub first_name: String,
    pub middle_name: String,
    pub last_name: String,

// previous variant
#[derive(Serialize, Deserialize, Flow)]
#[flow(variant = 2)]
pub struct UserV2 {
    pub first_name: String,
    pub last_name: String,

// the first variant of the User entity
#[derive(Serialize, Deserialize, Flow)]
#[flow(variant = 1)]
pub struct UserV1 {
    pub name: String,

// Migration from UserV1 and UserV2 for User
impl From<UserV1> for User { /* migration */ }
impl From<UserV2> for User { /* migration */ }


use serde_flow::{encoder::bincode, flow::File, flow::FileMigrate, Flow};

// create an old struct UserV2
let user_v2 = UserV2 {
    name: "John Adam Doe".to_string(),

// serialize to the disk

// deserialize as User
let user = User::load_from_path::<bincode::Encoder>(path.as_path()).unwrap();

// Migrate and load (loads, migrates, saves, and returns new entity)
let user User::load_and_migrate::<bincode::Encoder>(path.as_path()).unwrap();

// Just migrate (loads, migrates, and saves new entity)

📜 License

Serde-flow is open-source software, freely available under the MIT License.


~158K SLoC