mmustafasenoglu opened a new pull request, #16384:
URL: https://github.com/apache/dubbo/pull/16384
## What is the purpose of the change
Fix a race condition in `FastJson2ObjectInput` where concurrent
deserialization of JSONB bytes containing `$ref` references produces null
fields.
## Brief changelog
- Add striped locks (64 stripes) to `FastJson2ObjectInput` keyed by target
class
- Wrap `JSONB.parseObject` calls in `synchronized (getStripe(cls))` blocks
- Add `getStripe(Class<?>)` helper method
## Related issue
Fixes #16368
## Detailed explanation
### Root cause
When multiple objects reference the same shared singleton (e.g.,
`List.of()`), fastjson2 serialization writes `$ref` references instead of
duplicating the value. During concurrent deserialization:
1. Multiple threads encounter an ObjectReader cache miss for the same type
2. fastjson2's `ObjectReaderProvider.getObjectReaderInternal` has a
non-atomic check-then-act: `if (objectReader == null) {
createObjectReader(...); putIfAbsent(...); }`
3. Concurrent `createObjectReader` calls produce corrupted ObjectReader
instances (incomplete FieldReader arrays)
4. `$ref` resolution via `fieldReader.accept()` fails silently → fields
remain null
### Fix
Striped locks keyed by target class serialize `JSONB.parseObject` calls per
type. Concurrent deserialization of **different** types is unaffected; only
**same-type** concurrent deserialization is serialized, preventing the
fastjson2 race condition.
The 64-stripe design provides sufficient concurrency for typical
applications while bounding memory usage.
### How to reproduce
See the reproduction test in the issue: 200 threads with CyclicBarrier,
cache cleared each round, ~1% of tasks produce null fields.
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
