daguimu opened a new pull request, #16171:
URL: https://github.com/apache/dubbo/pull/16171
## Problem
`Version.VERSION2INT` uses a plain `HashMap` that is accessed concurrently
from the RPC hot path:
```java
private static final Map<String, Integer> VERSION2INT = new HashMap<>(); //
NOT thread-safe
public static int getIntVersion(String version) {
Integer v = VERSION2INT.get(version); // concurrent read
if (v == null) {
v = parseInt(version);
VERSION2INT.put(version, v); // concurrent write
}
return v;
}
```
This is called via `isSupportResponseAttachment()` on **every RPC call** to
check protocol compatibility.
### Impact
Concurrent `put()` during `HashMap` internal resize can corrupt the hash
table's bucket chain, causing threads to **hang permanently in an infinite
loop** during `get()` — a well-documented JDK issue with unsynchronized
`HashMap` access. This is a classic production incident pattern that is
difficult to diagnose.
## Root Cause
`HashMap` was used where `ConcurrentHashMap` is required. The code comment
on the field even notes it is used for performance ("int compare expect to has
higher performance than string"), but the collection itself is not thread-safe.
## Fix
Replace `new HashMap<>()` with `new ConcurrentHashMap<>()`. The
check-then-act pattern in `getIntVersion()` is acceptable because:
1. The computation is **idempotent** (same version string always produces
the same int)
2. Worst case under concurrent miss: two threads compute and put the same
value — no data loss or corruption with `ConcurrentHashMap`
## Tests Added
- `testGetIntVersionConcurrency` — 10 threads × 1000 iterations calling
`getIntVersion` with various version strings concurrently, verifying consistent
results
All 9 tests in `VersionTest` pass.
## Impact
- 1-line production change: `HashMap` → `ConcurrentHashMap`
- Zero behavioral change
- Eliminates risk of thread hangs in the RPC protocol path
--
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]
