1. Grizzly 2.0 Initial propose overview
Oleksiy Stashok 1

2. Agenda Grizzly 2.0 goals Architecture
Grizzly 1.x architecture solutions to be changed
Abstractions overview
Examples

3. Agenda Grizzly 2.0 goals Architecture
Grizzly 1.x architecture solutions to be changed
Abstractions overview
Examples

4. 1. Grizzly 2.0 goals
Make framework core API more clear and useful, based on earlier experience
Performance. Make sure API changes will not affect the performance, which is still the main goal for Grizzly
Introduce unified Buffer/Memory management API
Unify ThreadPool implementation. Make it compatible with Thread pool abstractions, provided by JDK.

5. Agenda Grizzly 2.0 goals Architecture
Grizzly 1.x architecture solutions to be changed
Abstractions overview
Examples

6. Agenda Grizzly 2.0 goals Architecture
Grizzly 1.x architecture solutions to be changed
Abstractions overview
Examples

7. 2.1 Grizzly 1.x architecture solutions to be changed
Split different transports logic, as it makes mess
No Controller
NIO Transport <-1 : 1-> SelectorHandler <-1 : 1-> SelectionKeyHandler. No copies, no clones.
Unify SelectionKey events processing: client and server side connections will be processed the same way
Unify SelectionKey attachments

8. 2.1 Grizzly 1.x architecture solutions to be changed (cont.)
ConnectorHandler will be responsible just for client ”connect” phase.

9. Agenda Grizzly 2.0 goals Architecture
Grizzly 1.x architecture solutions to be changed
Abstractions overview
Examples

10. 2.2.1 Transport Represents one specific transport implementation
Configuration
Thread pool
Memory/ByteBuffer manager
Connection IO events processing logic
Life-cycle control: start, pause, resume, stop

11. 2.2.2 Connection
Represents connection unit in Grizzly, both server and client side
Implements basic IO operations: read, write, close
Hides protocol specifics, NIO complexity
Could be used to store connection state/attributes

12. 2.2.3 IOEventProcessor
Common interface for processing IO events, occurred on the connection
CallbackHandler and ProtocolChain implement IOEventProcessor
It's easy to write custom IOEventProcessor:
public void process(IOEventContext context) throws IOException;

13. 2.2.4 IOEventProcessorSelector
Incapsulates logic for selecting appropriate IOEventProcessor(s) to process specific IO event
Is associated with Transport, although each specific connection unit, can have its own selector, which has higher priority
public boolean injectEventProcessors(int connectionOp,
Connection connection,
IOEventProcessorChain eventProcessorChain);

14. 2.2.5 MemoryManager Manages memory allocation, release operations
Associated with a Transport
Default implementations: ByteBufferManager, ByteBufferViewManager
public interface MemoryManager<E> {
public E allocate(int size);
public E reallocate(E oldBuffer, int newSize);
public void release(E buffer); }

15. 2.2.6 ThreadPool Currently its API was taken from Grizzly 1.x
Associated with a Transport
Needs to be compatible with JDK ThreadPool?

16. 2.2.7 Attribute Typed Attribute implementation (proposed by Ken)
Compile time attribute types check
Example:
Attribute<Integer> lengthAttr;
AttributeHolder session;
public void setLength(int length) {
lengthAttr.set(session, length); }
public int getLength() {
return lengthAttr.get(session);

17. Agenda Grizzly 2.0 goals Architecture
Grizzly 1.x architecture solutions to be changed
Abstractions overview
Examples

18. 3.1 Start and stop the Transport
TCPNIOTransport transport = new TCPNIOTransport();
try {
transport.bind(7777);
transport.start();
} finally {
transport.stop(); }

19. 3.2 Blocking client write TCPNIOConnectorHandler connectorHandler =
new TCPNIOConnectorHandler(transport);
try {
IOFuture<NIOConnection> future =
connectorHandler.connectAsync("localhost", 7777);
NIOConnection connection =
future.get(10, TimeUnit.SECONDS);
connection.write(someBuffer);
} finally {
connectorHandler.close(); }

20. 3.3 Async client write TCPNIOConnectorHandler connectorHandler =
new TCPNIOConnectorHandler(transport);
IOFuture writeFuture = connection.writeAsync(someBuffer);
WriteResult writeResult =
writeFuture.get(10, TimeUnit.SECONDS);

21. 3.4 Filter chain write
public class CustomFilter extends AbstractFilter {
public boolean execute(IOEventContext ctx) throws
IOException {
ByteBuffer response = getResponse();
// Filter's filterWrite performs some Filter specific
// response transformation and call filterWrite
// on previous Filter in chain until it comes to
// TransportFilter, which will send response on a wire
IOFuture future = filterWriteAsync(ctx, response); }

22. 3.5 Filter chain write (cont.)
public class CustomFilter extends AbstractFilter {
public boolean execute(IOEventContext ctx) throws
IOException {
ByteBuffer response = getResponse();
// It's possible to send response directly on a wire
IOFuture future =
ctx.getConnection().writeAsync(response); }

23. Grizzly 2.0 Initial propose overview23