Github user StefanRRichter commented on a diff in the pull request:
https://github.com/apache/flink/pull/5518#discussion_r169077434
--- Diff:
flink-contrib/flink-statebackend-rocksdb/src/main/java/org/apache/flink/contrib/streaming/state/RocksDBKeyedStateBackend.java
---
@@ -1991,43 +1999,71 @@ public int numStateEntries() {
return count;
}
+ /**
+ * This class is not thread safety.
+ */
private static class RocksIteratorWrapper<K> implements Iterator<K> {
private final RocksIterator iterator;
private final String state;
private final TypeSerializer<K> keySerializer;
private final int keyGroupPrefixBytes;
+ private final byte[] namespaceBytes;
+ private K nextKey;
public RocksIteratorWrapper(
RocksIterator iterator,
String state,
TypeSerializer<K> keySerializer,
- int keyGroupPrefixBytes) {
+ int keyGroupPrefixBytes,
+ byte[] namespaceBytes) {
this.iterator = Preconditions.checkNotNull(iterator);
this.state = Preconditions.checkNotNull(state);
this.keySerializer =
Preconditions.checkNotNull(keySerializer);
this.keyGroupPrefixBytes =
Preconditions.checkNotNull(keyGroupPrefixBytes);
+ this.namespaceBytes =
Preconditions.checkNotNull(namespaceBytes);
+ this.nextKey = null;
}
@Override
public boolean hasNext() {
- return iterator.isValid();
+ final int namespaceBytesLength = namespaceBytes.length;
+ while (nextKey == null && iterator.isValid()) {
+ try {
+ boolean namespaceValid = true;
+ byte[] key = iterator.key();
+ if (key.length >= namespaceBytesLength
+ keyGroupPrefixBytes) {
+ for (int i = 1; i <=
namespaceBytesLength; ++i) {
+ if (key[key.length - i]
!= namespaceBytes[namespaceBytesLength - i]) {
+ namespaceValid
= false;
+ break;
+ }
+ }
+ if (namespaceValid) {
+
DataInputViewStreamWrapper dataInput = new DataInputViewStreamWrapper(
+ new
ByteArrayInputStreamWithPos(key, keyGroupPrefixBytes, key.length -
keyGroupPrefixBytes));
+ K value =
keySerializer.deserialize(dataInput);
+ if
(dataInput.available() == namespaceBytesLength) {
--- End diff --
What is the purpose of this `if`? It looks more like a sanity check, but at
the same time will silently drop data if it is triggered.
---
