bitflicker64 opened a new issue, #3067:
URL: https://github.com/apache/hugegraph/issues/3067
## Source
- **File:**
`hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/io/HugeGryoModule.java`
- **References:** L151-L152
---
Now I have a comprehensive picture of the issue. Let me summarize what I
found:
1. In `HugeGryoModule.java`, there are four Kryo serializer inner classes
for schema types:
- `PropertyKeyKryoSerializer`
- `VertexLabelKryoSerializer`
- `EdgeLabelKryoSerializer`
- `IndexLabelKryoSerializer`
2. Each of these has a `write()` method that is fully implemented (using
`GraphSONSchemaSerializer` to convert the schema object to a `Map<HugeKeys,
Object>`, then writing it via `writeEntry()`), and a `read()` method that is a
stub returning `null` with a `// TODO` comment.
3. The `readEntry()` helper method exists in `HugeGryoModule` but is marked
`@SuppressWarnings("unused")` — it reads back the `Map<HugeKeys, Object>` from
the Kryo stream, but is never called by any of the `read()` methods.
4. `GraphSONSchemaSerializer` only has `write*` methods — there are no
corresponding `read*` methods in it. The `SchemaSerializer` interface (in the
backend serializer package) does have `readEdgeLabel`, `readVertexLabel`, etc.,
but those take a `HugeGraph` and a `BackendEntry`, not a `Map<HugeKeys,
Object>`.
5. The TODO affects all 4 schema serializer `read()` methods in the same
file.
The issue is that Gryo (Kryo-based binary serialization used by TinkerPop)
can serialize schema objects out but cannot deserialize them back. Any code
path that tries to read a Gryo-serialized `EdgeLabel`, `VertexLabel`,
`PropertyKey`, or `IndexLabel` from a binary stream will silently get `null`
instead of the actual object.
The scope of the fix is confined to `HugeGryoModule.java` (1 file), but may
also require adding read methods to `GraphSONSchemaSerializer` (1 file), for a
total of 1-2 files affected.
## The Issue
The TODO affects **all four** schema Kryo serializer inner classes in
`HugeGryoModule.java`, not just `EdgeLabelKryoSerializer`. Each has a
fully-implemented `write()` but a stub `read()` that unconditionally returns
`null`:
```
PropertyKeyKryoSerializer.read() → // TODO, return null (line 151-154)
VertexLabelKryoSerializer.read() → // TODO, return null (line 165-168)
EdgeLabelKryoSerializer.read() → // TODO, return null (line 179-182)
IndexLabelKryoSerializer.read() → // TODO, return null (line 193-196)
```
### What exactly is broken
The `write()` side works: it calls
`GraphSONSchemaSerializer.writeEdgeLabel()` to produce a `Map<HugeKeys,
Object>`, then serializes it to the Kryo binary stream via `writeEntry()`.
The inverse path — `readEntry()` — already exists and correctly reads back
the `Map<HugeKeys, Object>` from the stream, but it is annotated
`@SuppressWarnings("unused")` because **nothing calls it**:
The missing piece is: after calling `readEntry()` to get the map back, there
is no code to reconstruct an `EdgeLabel` (or any other schema object) from that
map. `GraphSONSchemaSerializer` only has `write*` methods — it has no `read*`
counterparts:
The backend `SchemaSerializer` interface does define
`readEdgeLabel(HugeGraph, BackendEntry)` etc., but those operate on
`BackendEntry` objects (storage-layer format), not on the `Map<HugeKeys,
Object>` that the Gryo path produces:
### Consequence
Any Gryo (TinkerPop binary) deserialization of schema objects — e.g. reading
an `EdgeLabel` back from a Gryo stream — silently returns `null`. This would
cause NPEs or incorrect behavior in any code path that round-trips schema
objects through Gryo I/O (e.g. Gremlin result serialization over the binary
protocol, graph export/import via Gryo).
### What code needs to change
**1 primary file** needs changes:
-
`hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/io/HugeGryoModule.java`
— The four `read()` stubs must be implemented. Each should call
`readEntry()` to recover the `Map<HugeKeys, Object>`, then reconstruct the
schema object from that map.
**1 supporting file** likely also needs changes:
-
`hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/io/GraphSONSchemaSerializer.java`
— Needs `readEdgeLabel(Map<HugeKeys,Object>, HugeGraph)`,
`readVertexLabel(...)`, `readPropertyKey(...)`, and `readIndexLabel(...)`
methods added, mirroring the existing `write*` methods in reverse. (The
`HugeGryoModule` serializers need a `HugeGraph` reference to reconstruct schema
objects, which is a complication — the Kryo `Serializer` API does not pass one
in, so the serializer classes would need to hold a graph reference injected at
registration time.)
**Total: 2 files** directly affected, with the core logic gap being the
missing `read*` methods in `GraphSONSchemaSerializer` and the unimplemented
`read()` bodies in `HugeGryoModule`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
