org.apache.flink.runtime.blob

Class BlobCacheSizeTracker

java.lang.Object

org.apache.flink.runtime.blob.BlobCacheSizeTracker

public class BlobCacheSizeTracker
extends Object

BlobCacheSizeTracker uses LinkedHashMap to maintain the LRU order for the files in the cache. When new files are intended to be put into cache, checkLimit is called to query the files should be removed. This tracker maintains a lock to avoid concurrent modification. To avoid the inconsistency, make sure that hold the READ/WRITE lock in PermanentBlobCache first and then hold the lock here.

Field Summary

Fields

Modifier and Type	Field and Description
protected long	sizeLimit

Constructor Summary

Constructors

Constructor and Description
BlobCacheSizeTracker(long sizeLimit)

Method Summary

Modifier and Type	Method and Description
List<Tuple2<JobID,BlobKey>>	checkLimit(long size) Check the size limit and return the BLOBs to delete.
void	track(JobID jobId, BlobKey blobKey, long size) Register the BLOB to the tracker.
void	untrack(Tuple2<JobID,BlobKey> key) Remove the BLOB from the tracker.
void	untrackAll(JobID jobId) Unregister all the tracked BLOBs related to the given job.
void	update(JobID jobId, BlobKey blobKey) Update the least used index for the BLOBs so that the tracker can easily find out the least recently used BLOBs.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

sizeLimit

protected final long sizeLimit

Constructor Detail

BlobCacheSizeTracker

public BlobCacheSizeTracker(long sizeLimit)

Method Detail

checkLimit

public List<Tuple2<JobID,BlobKey>> checkLimit(long size)

Check the size limit and return the BLOBs to delete.

Parameters:

size - size of the BLOB intended to put into the cache

Returns:

list of BLOBs to delete before putting into the target BLOB

track

public void track(JobID jobId,
                  BlobKey blobKey,
                  long size)

Register the BLOB to the tracker.

untrack

public void untrack(Tuple2<JobID,BlobKey> key)

Remove the BLOB from the tracker.

update

public void update(JobID jobId,
                   BlobKey blobKey)

Update the least used index for the BLOBs so that the tracker can easily find out the least recently used BLOBs.

untrackAll

public void untrackAll(JobID jobId)

Unregister all the tracked BLOBs related to the given job.