Schema Evolution
Schema evolution lets different versions of your service exchange messages safely — a v2 writer can produce a message a v1 reader still understands, and vice versa.
Compatible And Schema-Consistent Evolution
- Compatible mode (default): writes extra field metadata so readers can skip unknown fields and tolerate missing ones. Good for independent deployments, rolling upgrades, and xlang services.
- Schema-consistent mode: more compact, but both sides must have exactly the same schema. Use it only when schemas do not change, or when all services update together.
Default Compatible Mode
const fory = new Fory();
Use this when:
- services deploy schema changes independently
- older readers may see newer payloads
- newer readers may see older payloads from before a field was added
Example
Writer schema:
const writerType = Type.struct(
{ typeId: 1001 },
{
name: Type.string(),
age: Type.int32(),
},
);
Reader schema with fewer fields:
const readerType = Type.struct(
{ typeId: 1001 },
{
name: Type.string(),
},
);
With compatible mode, the reader ignores fields it does not know about, and fills unknown fields with default values.
Opting Out of Evolution for One Struct
You can disable evolution metadata for a specific struct even inside a compatible: true instance:
const fixedType = Type.struct(
{ typeId: 1002, evolving: false },
{
name: Type.string(),
},
);
evolving: false produces smaller messages for that struct, but both the writer and reader must agree on this setting. If one side writes with evolving: false and the other reads expecting compatible metadata, deserialization will fail.
When to Use Each Mode
| Schema-consistent | Compatible | |
|---|---|---|
| Services always update together | best choice | works, but wasteful |
| Independent deployments | will break | best choice |
| Smallest possible messages | best choice | slightly larger |
| Rolling upgrades | risky | safe |
Xlang Requirement
Compatible mode only protects you from schema differences in the fields of a type. You still need the same type identity (same numeric ID or same namespace + typeName) on every side. See Xlang Serialization.