Log a warn message when the size of buffers passes the limit of Netty's maxOrder and the allocation moves form pooled to unpooled

[andsel]

In V2Batch there is an accumulation of raw payloads of events, that later are lazily decoded to Logstash events.

logstash-input-beats/src/main/java/org/logstash/beats/V2Batch.java

Lines 89 to 91 in 4c6fd82

	if (internalBuffer.writableBytes() < size + (2 * SIZE_OF_INT)){
	internalBuffer.capacity(internalBuffer.capacity() + size + (2 * SIZE_OF_INT));
	}

This accumulation is one point where unpooled Netty's buffers are requested.
Update the implementation of the method to log a warn message, when the requested size requires an un pooled allocation. The log happens the first time the unpooled allocation is requested and prints also the expected maxOrder that would solve the problem.

void addMessage(int sequenceNumber, ByteBuf buffer, int size) {
...
        if (internalBuffer.writableBytes() < size + (2 * SIZE_OF_INT)) {
            int expectedSize = internalBuffer.capacity() + size + (2 * SIZE_OF_INT)
            int idealMaxOrder = computeIdealMaxOrder(expectedSize);
            if (warnNeverLogged(idealMaxOrder)) {
                LOGGER.warn("Got a batch size of {} bytes, while this instance expects batches up to {}, please bump maxOrder to {}.", size, allocator.chunkSize, idealMaxOrder);
        this.warningsDone.add(idealMaxOrder);
            }

            internalBuffer.capacity(internalBuffer.capacity() + size + (2 * SIZE_OF_INT));
        }
...
}