Class ForwardingAsyncResultSet (6.74.1) 

public class ForwardingAsyncResultSet extends ForwardingResultSet implements AsyncResultSet

Forwarding implementation of AsyncResultSet that forwards all calls to a delegate.

Inheritance

java.lang.Object > ForwardingStructReader > ForwardingResultSet > ForwardingAsyncResultSet

Implements

AsyncResultSet

Constructors

ForwardingAsyncResultSet(AsyncResultSet delegate)

public ForwardingAsyncResultSet(AsyncResultSet delegate)
Parameter
Name Description
delegate AsyncResultSet

Methods

<T>toList(Function<StructReader,T> transformer)

public List<T> <T>toList(Function<StructReader,T> transformer)

Transforms the row cursor into an immutable list using the given transformer function. transformer will be called once per row, thus the returned list will contain one entry per row. This method will block until all the rows have been yielded by the cursor.

WARNING: This will result in consuming the entire list so this should be used judiciously after considering the memory requirements of the returned list.

WARNING: The RowBase object passed to transformer function is not immutable and is not guaranteed to remain valid after the transformer function returns. The same RowBase object might be passed multiple times to the transformer with different underlying data each time. So NEVER keep a reference to the RowBase outside of the transformer. Specifically do not use com.google.common.base.Functions#identity() function.

Parameter
Name Description
transformer com.google.common.base.Function<StructReader,T>
Returns
Type Description
List<T>
Exceptions
Type Description
SpannerException

<T>toListAsync(Function<StructReader,T> transformer, Executor executor)

public ApiFuture<List<T>> <T>toListAsync(Function<StructReader,T> transformer, Executor executor)

Transforms the row cursor into an immutable list using the given transformer function. transformer will be called once per row, thus the returned list will contain one entry per row. The returned future will throw a SpannerException if the row cursor encountered any error or if the transformer threw an exception on any row.

The transformer will be run on the supplied executor. The implementation may batch multiple transformer invocations together into a single Runnable when possible to increase efficiency. At any point in time, there will be at most one invocation of the transformer in progress.

WARNING: This will result in materializing the entire list so this should be used judiciously after considering the memory requirements of the returned list.

WARNING: The RowBase object passed to transformer function is not immutable and is not guaranteed to remain valid after the transformer function returns. The same RowBase object might be passed multiple times to the transformer with different underlying data each time. So NEVER keep a reference to the RowBase outside of the transformer. Specifically do not use com.google.common.base.Functions#identity() function.

Parameters
Name Description
transformer com.google.common.base.Function<StructReader,T>
executor Executor
Returns
Type Description
ApiFuture<List<T>>

cancel()

public void cancel()

Attempt to cancel this operation and free all resources. Non-blocking. This is a no-op for child row cursors and does not cancel the parent cursor.

resume()

public void resume()

Resume callbacks from the cursor. If there is more data available, a callback will be dispatched immediately. This can be called from any thread.

setCallback(Executor exec, AsyncResultSet.ReadyCallback cb)

public ApiFuture<Void> setCallback(Executor exec, AsyncResultSet.ReadyCallback cb)

Register a callback with the ResultSet to be made aware when more data is available, changing the usage pattern from sync to async. Details:

  • The callback will be called at least once.
  • The callback is run each time more results are available, or when we discover that there will be no more results. (unless paused, see below). Spurious callbacks are possible, see below.
  • Spanner guarantees that one callback is ever outstanding at a time. Also, future callbacks guarantee the "happens before" property with previous callbacks.
  • A callback normally consumes all available data in the ResultSet, and then returns CallbackResponse#CONTINUE.
  • If a callback returns CallbackResponse#CONTINUE with data still in the ResultSet, the callback is invoked again immediately!
  • Once a callback has returned CallbackResponse#PAUSE on the cursor no more callbacks will be run until a corresponding #resume().
  • Callback will stop being called once any of the following occurs:
    1. Callback returns CallbackResponse#DONE.
    2. ForwardingAsyncResultSet#tryNext() returns CursorState#DONE.
    3. ForwardingAsyncResultSet#tryNext() throws an exception.
  • Callback may possibly be invoked after a call to ForwardingAsyncResultSet#cancel() call, but the subsequent call to #tryNext() will yield a SpannerException.
  • Spurious callbacks are possible where cursors are not actually ready. Typically callback should return CallbackResponse#CONTINUE any time it sees CursorState#NOT_READY.

Flow Control

If no flow control is needed (say because result sizes are known in advance to be finite in size) then async processing is simple. The following is a code example that transfers work from the cursor to an upstream sink:


 @Override
 public CallbackResponse cursorReady(ResultSet cursor) {
   try {
     while (true) {
       switch (cursor.tryNext()) {
         case OK:    upstream.emit(cursor.getRow()); break;
         case DONE:  upstream.done(); return CallbackResponse.DONE;
         case NOT_READY:  return CallbackResponse.CONTINUE;
       }
     }
   } catch (SpannerException e) {
     upstream.doneWithError(e);
     return CallbackResponse.DONE;
   }
 }

Flow control may be needed if for example the upstream system may not always be ready to handle more data. In this case the app developer has two main options:

  • Semi-async: make upstream.emit() a blocking call. This will block the callback thread until progress is possible. When coding in this way the threads in the Executor provided to setCallback must be blockable without causing harm to progress in your system.
  • Full-async: call cursor.pause() and return from the callback with data still in the Cursor. Once in this state cursor waits until resume() is called before calling callback again.
Parameters
Name Description
exec Executor
cb AsyncResultSet.ReadyCallback
Returns
Type Description
ApiFuture<Void>

tryNext()

public AsyncResultSet.CursorState tryNext()

Non-blocking call that attempts to step the cursor to the next position in the stream. The cursor may be inspected only if the cursor returns CursorState.OK.

A caller will typically call tryNext in a loop inside the ReadyCallback, consuming all results available. For more information see #setCallback(Executor, ReadyCallback).

Currently this method may only be called if a ReadyCallback has been registered. This is for safety purposes only, and may be relaxed in future.

Returns
Type Description
AsyncResultSet.CursorState
Exceptions
Type Description
SpannerException