akalash commented on a change in pull request #17946:
URL: https://github.com/apache/flink/pull/17946#discussion_r760299634



##########
File path: 
flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/ExecutionAttemptMappingProvider.java
##########
@@ -38,32 +38,34 @@
     private final List<ExecutionVertex> tasks;
 
     /** The cached mapping, which would only be updated on miss. */
-    private final LinkedHashMap<ExecutionAttemptID, ExecutionVertex> 
cachedTasksById;
+    private volatile Map<ExecutionAttemptID, ExecutionVertex> cachedTasksById;
 
     public ExecutionAttemptMappingProvider(Iterable<ExecutionVertex> 
tasksIterable) {
         this.tasks = new ArrayList<>();
         tasksIterable.forEach(this.tasks::add);
 
-        this.cachedTasksById =
-                new LinkedHashMap<ExecutionAttemptID, 
ExecutionVertex>(tasks.size()) {
-
-                    @Override
-                    protected boolean removeEldestEntry(
-                            Map.Entry<ExecutionAttemptID, ExecutionVertex> 
eldest) {
-                        return size() > tasks.size();
-                    }
-                };
+        this.cachedTasksById = new ConcurrentHashMap<>(tasks.size());
     }
 
     public Optional<ExecutionVertex> getVertex(ExecutionAttemptID id) {
-        if (!cachedTasksById.containsKey(id)) {
-            cachedTasksById.putAll(getCurrentAttemptMappings());
-            if (!cachedTasksById.containsKey(id)) {
-                // the task probably gone after a restart
-                cachedTasksById.put(id, null);
+        ExecutionVertex vertex = cachedTasksById.get(id);
+        if (vertex != null) {
+            return Optional.of(vertex);
+        }
+
+        return updateAndGet(id);
+    }
+
+    private Optional<ExecutionVertex> updateAndGet(ExecutionAttemptID id) {
+        synchronized (tasks) {

Review comment:
       Synchronization on `this` is always dangerous since it can lead to 
deadlock if somebody decides to synchronize on this object outside. So in my 
case, I can create the new private lock object or I can use some other private 
variable in this class. I decided to choose the second option because actually, 
it looks pretty reasonable for me - synchronization on tasks shows that only 
one thread can have access to tasks which is true.




-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to