[
https://issues.apache.org/jira/browse/HDFS-17813?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18063866#comment-18063866
]
ASF GitHub Bot commented on HDFS-17813:
---------------------------------------
balodesecurity opened a new pull request, #8313:
URL: https://github.com/apache/hadoop/pull/8313
## Problem
In `NameCache.put()`, when a name's use count crosses the promotion
threshold, the cache was populated with the caller's `name` argument rather
than the original `useCount.value` that was stored in the transient map:
```java
// Before (buggy)
cache.put(name, name);
```
This means after promotion the cache holds a different object reference than
what was stored before promotion. Any caller that held a reference to the
pre-promotion value now has a stale reference, and two distinct `String`
objects exist for what should be a single cached identity — wasting memory and
breaking reference equality.
## Fix
Pass `useCount` into the `promote()` helper and store `useCount.value`
instead:
```java
// After
cache.put(name, useCount.value);
```
This preserves the original object reference across the promotion boundary.
## Testing
- Added `TestNameCache#testPromotionPreservesObjectIdentity`: creates a
`NameCache` with threshold 2, inserts a non-interned `String`, triggers
promotion, and uses `assertSame` to verify the returned reference is identical
before and after promotion.
- Test passes locally.
> NameCache promote wrong name to cache map
> -----------------------------------------
>
> Key: HDFS-17813
> URL: https://issues.apache.org/jira/browse/HDFS-17813
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Reporter: khazhen
> Priority: Minor
> Labels: pull-request-available
>
> The NameCache class is used to cache frequently used names in namenode, it
> promotes a name used more than useThreshold to the cache, the promote logic:
> {code:java}
> K put(final K name) {
> K internal = cache.get(name);
> if (internal != null) {
> lookups++;
> return internal;
> }
> // Track the usage count only during initialization
> if (!initialized) {
> UseCount useCount = transientMap.get(name);
> if (useCount != null) {
> useCount.increment();
> if (useCount.get() >= useThreshold) {
> promote(name); // name got promoted
> }
> return useCount.value;
> }
> useCount = new UseCount(name);
> transientMap.put(name, useCount);
> }
> return null;
> } {code}
> When promoting, the cache stores the `name` parameter from put() instead of
> the existing useCount.value. This causes the returned value to change after a
> name is promoted, resulting in memory duplication.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
