HBASE-27590 Change Iterable to List in SnapshotFileCache (#4995)

Signed-off-by: Duo Zhang <[email protected]>
(cherry picked from commit d2c5af1)

Author: petersomogyi

hbase-server/src/main/java/org/apache/hadoop/hbase/master/snapshot/SnapshotFileCache.java

@@ -178,15 +178,11 @@ public synchronized void triggerCacheRefreshForTesting() {
    * at that point, cache will still think the file system contains that file and return
    * <tt>true</tt>, even if it is no longer present (false positive). However, if the file never was
    * on the filesystem, we will never find it and always return <tt>false</tt>.
-   * @param files file to check, NOTE: Relies that files are loaded from hdfs before method is
-   *              called (NOT LAZY)
+   * @param files file to check
    * @return <tt>unReferencedFiles</tt> the collection of files that do not have snapshot references
    * @throws IOException if there is an unexpected error reaching the filesystem.
    */
-  // XXX this is inefficient to synchronize on the method, when what we really need to guard against
-  // is an illegal access to the cache. Really we could do a mutex-guarded pointer swap on the
-  // cache, but that seems overkill at the moment and isn't necessarily a bottleneck.
-  public Iterable<FileStatus> getUnreferencedFiles(Iterable<FileStatus> files,
+  public Iterable<FileStatus> getUnreferencedFiles(List<FileStatus> files,
     final SnapshotManager snapshotManager) throws IOException {
     List<FileStatus> unReferencedFiles = Lists.newArrayList();
     List<String> snapshotsInProgress = null;

hbase-server/src/main/java/org/apache/hadoop/hbase/master/snapshot/SnapshotHFileCleaner.java

@@ -20,7 +20,10 @@
 import java.io.IOException;
 import java.util.Collection;
 import java.util.Collections;
+import java.util.List;
 import java.util.Map;
+import java.util.stream.Collectors;
+import java.util.stream.StreamSupport;
 import org.apache.hadoop.conf.Configuration;
 import org.apache.hadoop.fs.FileStatus;
 import org.apache.hadoop.fs.FileSystem;
@@ -64,8 +67,15 @@ public class SnapshotHFileCleaner extends BaseHFileCleanerDelegate {
 
   @Override
   public Iterable<FileStatus> getDeletableFiles(Iterable<FileStatus> files) {
+    // The Iterable is lazy evaluated, so if we just pass this Iterable in, we will access the HFile
+    // storage inside the snapshot lock, which could take a lot of time (for example, several
+    // seconds), and block all other operations, especially other cleaners.
+    // So here we convert it to List first, to force it evaluated before calling
+    // getUnreferencedFiles, so we will not hold snapshot lock for a long time.
+    List<FileStatus> filesList =
+      StreamSupport.stream(files.spliterator(), false).collect(Collectors.toList());
     try {
-      return cache.getUnreferencedFiles(files, master.getSnapshotManager());
+      return cache.getUnreferencedFiles(filesList, master.getSnapshotManager());
     } catch (CorruptedSnapshotException cse) {
       LOG.debug("Corrupted in-progress snapshot file exception, ignored ", cse);
     } catch (IOException e) {