[accumulo] branch 1.10 updated: fixes #3260 by making cached split info a softref (#3261)

[kturner]

This is an automated email from the ASF dual-hosted git repository.

kturner pushed a commit to branch 1.10
in repository https://gitbox.apache.org/repos/asf/accumulo.git


The following commit(s) were added to refs/heads/1.10 by this push:
     new d925db128b fixes #3260 by making cached split info a softref (#3261)
d925db128b is described below

commit d925db128b568d61fd67647e5e033dfccf26f8dc
Author: Keith Turner <ktur...@apache.org>
AuthorDate: Thu Mar 30 19:54:50 2023 -0400

    fixes #3260 by making cached split info a softref (#3261)
    
    The changes in #3249 introduce per tablet caching of split information so 
that it does not need to be recomputed. If a tablet happens to have large keys 
and those end up getting cached by the new code in #3249 then it could cause 
memory pressure or OOMEs in the tablet server.  This commit attempts to avoid 
those problems by using a softref to store the cached information.
---
 .../java/org/apache/accumulo/tserver/tablet/Tablet.java   | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

diff --git 
a/server/tserver/src/main/java/org/apache/accumulo/tserver/tablet/Tablet.java 
b/server/tserver/src/main/java/org/apache/accumulo/tserver/tablet/Tablet.java
index 580616ee3f..9cd5cbd968 100644
--- 
a/server/tserver/src/main/java/org/apache/accumulo/tserver/tablet/Tablet.java
+++ 
b/server/tserver/src/main/java/org/apache/accumulo/tserver/tablet/Tablet.java
@@ -24,6 +24,7 @@ import java.io.ByteArrayInputStream;
 import java.io.DataInputStream;
 import java.io.FileNotFoundException;
 import java.io.IOException;
+import java.lang.ref.SoftReference;
 import java.util.ArrayList;
 import java.util.Collection;
 import java.util.Collections;
@@ -1821,8 +1822,12 @@ public class Tablet implements TabletCommitter {
     }
   }
 
-  private AtomicReference<SplitComputations> lastSplitComputation = new 
AtomicReference<>();
-  private Lock splitComputationLock = new ReentrantLock();
+  // The following caches keys from users files needed to compute a tablets 
split point. This cached
+  // data could potentially be large and is therefore stored using a soft 
refence so the Java GC can
+  // release it if needed. If the cached information is not there it can 
always be recomputed.
+  private volatile SoftReference<SplitComputations> lastSplitComputation =
+      new SoftReference<>(null);
+  private final Lock splitComputationLock = new ReentrantLock();
 
   /**
    * Computes split point information from files when a tablets set of files 
changes. Do not call
@@ -1870,16 +1875,16 @@ public class Tablet implements TabletCommitter {
         }
 
         newComputation = new SplitComputations(files, midpoint, lastRow);
+
+        lastSplitComputation = new SoftReference<>(newComputation);
       } catch (IOException e) {
-        lastSplitComputation.set(null);
+        lastSplitComputation.clear();
         log.error("Failed to compute split information from files " + 
e.getMessage());
         return Optional.absent();
       } finally {
         splitComputationLock.unlock();
       }
 
-      lastSplitComputation.set(newComputation);
-
       return Optional.of(newComputation);
     } else {
       // some other thread seems to be working on split, let the other thread 
work on it