Flushing

While the filter graph is running, arbitrary amounts of data can be moving through the graph. Some of it might be in queues, waiting to be delivered. There are times when the filter graph needs to remove this pending data as quickly as possible and replace it with new data. After a seek command, for example, the source filter generates samples from a new position in the source. To minimize latency, downstream filters should discard any samples that were created before the seek command. The process of discarding samples is called flushing. It enables the graph to be more responsive when events alter the normal data flow.

Flushing is handled slightly differently by the pull model than the push model. This article starts by describing the push model; then it describes the differences in the pull model.

Flushing happens in two stages.

• First, the source filter calls IPin::BeginFlush on the downstream filter's input pin. The downstream filter starts rejecting samples from upstream. It also discards any samples it is holding, and sends the BeginFlush call downstream to the next filter.
• When the source filter is ready to send new data, it calls IPin::EndFlush on the input pin. This signals the downstream filter that it can receive new samples. The downstream filter sends the EndFlush call to the next filter.

In the BeginFlush method, the input pin does the following:

1. Calls BeginFlush on downstream input pins.
2. Rejects any further calls that stream data, including Receive and EndOfStream.
3. Unblocks any upstream filters that are blocked waiting for a sample from the filter's allocator. Some filters decommit their allocators for this purpose.
4. Exits from any waits that block streaming. For example, renderer filters block when paused. They also block when they are waiting to render a sample at the correct stream time. The filter must unblock, so that samples queued upstream can be delivered and rejected. This step ensures that all upstream filters eventually unblock.

In the EndFlush method, the input pin does the following:

1. Waits for all queued samples to be discarded.
2. Frees any buffered data. This step can sometimes be performed in the BeginFlush method. However, BeginFlush is not synchronized with the streaming thread. The filter must not process or buffer any more data between the call to BeginFlush and the call to EndFlush.
3. Clears any pending EC_COMPLETE notifications.
4. Calls EndFlush downstream.

At this point, the filter can accept samples again. All samples are guaranteed to be more recent than the flush.

In the pull model, the parser filter initiates flushing, rather than the source filter. Not only does it call IPin::BeginFlush and IPin::EndFlush on the downstream filter, it also calls IAsyncReader::BeginFlush and IAsyncReader::EndFlush on the output pin of the source filter. If the source filter has pending read requests, it will discard them.