Github user jasobrown commented on a diff in the pull request:
https://github.com/apache/cassandra/pull/184#discussion_r160520881
--- Diff:
src/java/org/apache/cassandra/net/CompactEndpointSerializationHelper.java ---
@@ -21,28 +21,142 @@
import java.net.Inet4Address;
import java.net.Inet6Address;
import java.net.InetAddress;
+import java.nio.ByteBuffer;
-public class CompactEndpointSerializationHelper
+import org.apache.cassandra.io.IVersionedSerializer;
+import org.apache.cassandra.io.util.DataInputBuffer;
+import org.apache.cassandra.io.util.DataInputPlus;
+import org.apache.cassandra.io.util.DataOutputPlus;
+import org.apache.cassandra.locator.InetAddressAndPort;
+import org.apache.cassandra.streaming.messages.StreamMessage;
+
+/*
+ * As of version 4.0 the endpoint description includes a port number as an
unsigned short
+ */
+public class CompactEndpointSerializationHelper implements
IVersionedSerializer<InetAddressAndPort>
{
- public static void serialize(InetAddress endpoint, DataOutput out)
throws IOException
+ public static final IVersionedSerializer<InetAddressAndPort> instance
= new CompactEndpointSerializationHelper();
+
+ /**
+ * Streaming uses it's own version numbering so we need to map those
versions to the versions used my regular messaging.
+ * There are only two variants of the serialization currently so a
simple mapping around pre vs post 4.0 is fine.
+ */
+ public static final IVersionedSerializer<InetAddressAndPort>
streamingInstance = new IVersionedSerializer<InetAddressAndPort>()
{
- byte[] buf = endpoint.getAddress();
- out.writeByte(buf.length);
- out.write(buf);
+ public void serialize(InetAddressAndPort inetAddressAndPort,
DataOutputPlus out, int version) throws IOException
+ {
+ if (version < StreamMessage.VERSION_40)
+ {
+ instance.serialize(inetAddressAndPort, out,
MessagingService.VERSION_30);
+ }
+ else
+ {
+ instance.serialize(inetAddressAndPort, out,
MessagingService.VERSION_40);
+ }
+ }
+
+ public InetAddressAndPort deserialize(DataInputPlus in, int
version) throws IOException
+ {
+ if (version < StreamMessage.VERSION_40)
+ {
+ return instance.deserialize(in,
MessagingService.VERSION_30);
+ }
+ else
+ {
+ return instance.deserialize(in,
MessagingService.VERSION_40);
+ }
+ }
+
+ public long serializedSize(InetAddressAndPort inetAddressAndPort,
int version)
+ {
+ if (version < StreamMessage.VERSION_40)
+ {
+ return instance.serializedSize(inetAddressAndPort,
MessagingService.VERSION_30);
+ }
+ else
+ {
+ return instance.serializedSize(inetAddressAndPort,
MessagingService.VERSION_40);
+ }
+ }
+ };
+
+ private CompactEndpointSerializationHelper() {}
+
+ public void serialize(InetAddressAndPort endpoint, DataOutputPlus out,
int version) throws IOException
+ {
+ if (version >= MessagingService.VERSION_40)
+ {
+ byte[] buf = endpoint.address.getAddress();
--- End diff --
Since serializing/deserializing IP addrs is on the hot path (literally used
everywhere for internode messaging), I've long wondered if we can avoid the
byte array allocation and instead write to an int (for IPv4) or two longs (for
IPv6) as those would be stack allocated, and thus no extra garbage. We could
choose to pass the `DataOutputPlus` to a new function on `InetAddressAndPort`,
as well.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: pr-unsubscr...@cassandra.apache.org
For additional commands, e-mail: pr-h...@cassandra.apache.org
