[
https://issues.apache.org/jira/browse/HADOOP-18291?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17738882#comment-17738882
]
ASF GitHub Bot commented on HADOOP-18291:
-----------------------------------------
virajjasani commented on code in PR #5754:
URL: https://github.com/apache/hadoop/pull/5754#discussion_r1247450167
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/impl/prefetch/SingleFilePerBlockCache.java:
##########
@@ -299,9 +395,62 @@ public void put(int blockNumber, ByteBuffer buffer,
Configuration conf,
// Update stream_read_blocks_in_cache stats only after blocks map is
updated with new file
// entry to avoid any discrepancy related to the value of
stream_read_blocks_in_cache.
// If stream_read_blocks_in_cache is updated before updating the blocks
map here, closing of
- // the input stream can lead to the removal of the cache file even before
blocks is added with
- // the new cache file, leading to incorrect value of
stream_read_blocks_in_cache.
+ // the input stream can lead to the removal of the cache file even before
blocks is added
+ // with the new cache file, leading to incorrect value of
stream_read_blocks_in_cache.
prefetchingStatistics.blockAddedToFileCache();
+ addToLinkedListAndEvictIfRequired(entry);
+ }
+
+ /**
+ * Add the given entry to the head of the linked list and if the LRU cache
size
+ * exceeds the max limit, evict tail of the LRU linked list.
+ *
+ * @param entry Block entry to add.
+ */
+ private void addToLinkedListAndEvictIfRequired(Entry entry) {
+ addToHeadOfLinkedList(entry);
+ blocksLock.writeLock().lock();
+ try {
+ if (blocks.size() > maxBlocksCount && !closed.get()) {
+ Entry elementToPurge = tail;
+ tail = tail.getPrevious();
+ if (tail == null) {
+ tail = head;
+ }
+ tail.setNext(null);
+ elementToPurge.setPrevious(null);
+ deleteBlockFileAndEvictCache(elementToPurge);
+ }
+ } finally {
+ blocksLock.writeLock().unlock();
+ }
+ }
+
+ /**
+ * Delete cache file as part of the block cache LRU eviction.
+ *
+ * @param elementToPurge Block entry to evict.
+ */
+ private void deleteBlockFileAndEvictCache(Entry elementToPurge) {
+ boolean lockAcquired =
+ elementToPurge.takeLock(Entry.LockType.WRITE,
PREFETCH_WRITE_LOCK_TIMEOUT,
+ PREFETCH_WRITE_LOCK_TIMEOUT_UNIT);
+ if (!lockAcquired) {
+ LOG.error("Cache file {} deletion would not be attempted as write lock
could not"
Review Comment:
if eviction misses it, stream close would be able to clean it up.
if stream close misses it, then it stays on disk and we might eventually
also come up with some "file last accessed" based check and maybe some crons
removing them eventually. not a bad idea IMO.
> S3A prefetch - Implement LRU cache for SingleFilePerBlockCache
> --------------------------------------------------------------
>
> Key: HADOOP-18291
> URL: https://issues.apache.org/jira/browse/HADOOP-18291
> Project: Hadoop Common
> Issue Type: Sub-task
> Affects Versions: 3.4.0
> Reporter: Ahmar Suhail
> Assignee: Viraj Jasani
> Priority: Major
> Labels: pull-request-available
>
> Currently there is no limit on the size of disk cache. This means we could
> have a large number of files on files, especially for access patterns that
> are very random and do not always read the block fully.
>
> eg:
> in.seek(5);
> in.read();
> in.seek(blockSize + 10) // block 0 gets saved to disk as it's not fully read
> in.read();
> in.seek(2 * blockSize + 10) // block 1 gets saved to disk
> .. and so on
>
> The in memory cache is bounded, and by default has a limit of 72MB (9
> blocks). When a block is fully read, and a seek is issued it's released
> [here|https://github.com/apache/hadoop/blob/feature-HADOOP-18028-s3a-prefetch/hadoop-tools/hadoop-aws/src/main/java/org/apache/hadoop/fs/s3a/read/S3CachingInputStream.java#L109].
> We can also delete the on disk file for the block here if it exists.
>
> Also maybe add an upper limit on disk space, and delete the file which stores
> data of the block furthest from the current block (similar to the in memory
> cache) when this limit is reached.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
