[
https://issues.apache.org/jira/browse/HBASE-4330?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13103010#comment-13103010
]
Li Pi commented on HBASE-4330:
------------------------------
The change in v7 that fixes things:
- scache.cacheBlock(blockName, cachedItem); // if this
- // fails, due to
- // block already
- // being there, exception will be thrown
- backingStore.put(blockName, scache);
+
+ /*This will throw a runtime exception if we try to cache the same value
twice*/
+ scache.cacheBlock(blockName, cachedItem);
+
+ /*Spinlock, if we're spinlocking, that means an eviction hasn't taken
place yet*/
+ while (backingStore.putIfAbsent(blockName, scache) != null) {
+ Thread.yield();
+ }
The test failed when the following occurred:
Invariant:
Both SSC and SC have the same contents:
Violation:
Item A is in both SSC and SC.
Thread A: evicts A from ssc.
Thread B: starts doing a put into thread SC, -
Thread B: gets directed into SSC, starts doing put in SSC,
Thread B: put goes through thanks to the occuring eviction.
Thread A: calls evictor on SC, removing the object from SlabCache.
result: Object is in SSC, but not SC.
> Fix races in slab cache
> -----------------------
>
> Key: HBASE-4330
> URL: https://issues.apache.org/jira/browse/HBASE-4330
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.92.0
> Reporter: Todd Lipcon
> Assignee: Li Pi
> Fix For: 0.92.0
>
> Attachments: hbase-4330.txt, hbase-4330.txt, hbase-4330v3.txt,
> hbase-4330v4.txt, hbase-4330v5.txt, hbase-4330v6.txt, hbase-4330v7.txt
>
>
> A few races are still lingering in the slab cache. Here are some tests and
> proposed fixes.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
