[
https://issues.apache.org/jira/browse/HADOOP-18740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17728181#comment-17728181
]
ASF GitHub Bot commented on HADOOP-18740:
-----------------------------------------
virajjasani commented on code in PR #5675:
URL: https://github.com/apache/hadoop/pull/5675#discussion_r1212472590
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/impl/prefetch/SingleFilePerBlockCache.java:
##########
@@ -268,12 +310,15 @@ public void close() throws IOException {
int numFilesDeleted = 0;
for (Entry entry : blocks.values()) {
+ entry.takeLock(Entry.LockType.WRITE);
Review Comment:
> also, L303: should closed be atomic?
+1 to this suggestion, let me create a separate patch with HADOOP-18756 to
better track it.
> good: no race condition in close
> bad) the usual
sounds reasonable, let me try setting timeout
> s3a prefetch cache blocks should be accessed by RW locks
> --------------------------------------------------------
>
> Key: HADOOP-18740
> URL: https://issues.apache.org/jira/browse/HADOOP-18740
> Project: Hadoop Common
> Issue Type: Sub-task
> Reporter: Viraj Jasani
> Assignee: Viraj Jasani
> Priority: Major
> Labels: pull-request-available
>
> In order to implement LRU or LFU based cache removal policies for s3a
> prefetched cache blocks, it is important for all cache reader threads to
> acquire read lock and similarly cache file removal mechanism (fs close or
> cache eviction) to acquire write lock before accessing the files.
> As we maintain the block entries in an in-memory map, we should be able to
> introduce read-write lock per cache file entry, we don't need coarse-grained
> lock shared by all entries.
>
> This is a prerequisite to HADOOP-18291.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
