Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
// Copyright 2017 Google Inc. All rights reserved.
|
|
|
|
//
|
|
|
|
// Use of this source code is governed by a BSD-style
|
|
|
|
// license that can be found in the LICENSE file or at
|
|
|
|
// https://developers.google.com/open-source/licenses/bsd
|
|
|
|
|
|
|
|
#include "packager/media/chunking/chunking_handler.h"
|
|
|
|
|
2018-01-09 01:18:15 +00:00
|
|
|
#include <algorithm>
|
|
|
|
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
#include "packager/base/logging.h"
|
|
|
|
#include "packager/base/threading/platform_thread.h"
|
|
|
|
#include "packager/media/base/media_sample.h"
|
|
|
|
|
|
|
|
namespace {
|
|
|
|
int64_t kThreadIdUnset = -1;
|
|
|
|
} // namespace
|
|
|
|
|
|
|
|
namespace shaka {
|
|
|
|
namespace media {
|
|
|
|
|
2017-07-27 17:23:27 +00:00
|
|
|
ChunkingHandler::ChunkingHandler(const ChunkingParams& chunking_params)
|
2018-01-09 01:18:15 +00:00
|
|
|
: chunking_params_(chunking_params),
|
|
|
|
thread_id_(kThreadIdUnset),
|
|
|
|
media_sample_comparator_(this),
|
|
|
|
cached_media_sample_stream_data_(media_sample_comparator_) {
|
2017-07-27 17:23:27 +00:00
|
|
|
CHECK_NE(chunking_params.segment_duration_in_seconds, 0u);
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
ChunkingHandler::~ChunkingHandler() {}
|
|
|
|
|
|
|
|
Status ChunkingHandler::InitializeInternal() {
|
|
|
|
segment_info_.resize(num_input_streams());
|
|
|
|
subsegment_info_.resize(num_input_streams());
|
|
|
|
time_scales_.resize(num_input_streams());
|
|
|
|
last_sample_end_timestamps_.resize(num_input_streams());
|
2018-01-09 01:18:15 +00:00
|
|
|
num_cached_samples_.resize(num_input_streams());
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
return Status::OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status ChunkingHandler::Process(std::unique_ptr<StreamData> stream_data) {
|
|
|
|
switch (stream_data->stream_data_type) {
|
2018-02-12 20:19:34 +00:00
|
|
|
case StreamDataType::kStreamInfo:
|
|
|
|
return OnStreamInfo(stream_data->stream_index, stream_data->stream_info);
|
2017-12-11 16:59:40 +00:00
|
|
|
case StreamDataType::kScte35Event: {
|
2018-01-12 00:55:43 +00:00
|
|
|
if (stream_data->stream_index != main_stream_index_) {
|
2017-12-11 16:59:40 +00:00
|
|
|
VLOG(3) << "Dropping scte35 event from non main stream.";
|
|
|
|
return Status::OK;
|
|
|
|
}
|
|
|
|
scte35_events_.push(std::move(stream_data));
|
|
|
|
return Status::OK;
|
|
|
|
}
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
case StreamDataType::kSegmentInfo:
|
2017-12-11 16:59:40 +00:00
|
|
|
VLOG(3) << "Droppping existing segment info.";
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
return Status::OK;
|
|
|
|
case StreamDataType::kMediaSample: {
|
2017-03-03 00:10:30 +00:00
|
|
|
const size_t stream_index = stream_data->stream_index;
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
DCHECK_NE(time_scales_[stream_index], 0u)
|
|
|
|
<< "kStreamInfo should arrive before kMediaSample";
|
2018-01-09 01:18:15 +00:00
|
|
|
|
|
|
|
if (stream_index != main_stream_index_ &&
|
|
|
|
!stream_data->media_sample->is_key_frame()) {
|
|
|
|
return Status(error::CHUNKING_ERROR,
|
|
|
|
"All non video samples should be key frames.");
|
|
|
|
}
|
|
|
|
// The streams are expected to be roughly synchronized, so we don't expect
|
|
|
|
// to see a lot of samples from one stream but no samples from another
|
|
|
|
// stream.
|
|
|
|
// The value is kind of arbitrary here. For a 24fps video, it is ~40s.
|
|
|
|
const size_t kMaxCachedSamplesPerStream = 1000u;
|
|
|
|
if (num_cached_samples_[stream_index] >= kMaxCachedSamplesPerStream) {
|
|
|
|
LOG(ERROR) << "Streams are not synchronized:";
|
|
|
|
for (size_t i = 0; i < num_cached_samples_.size(); ++i)
|
|
|
|
LOG(ERROR) << " [Stream " << i << "] " << num_cached_samples_[i];
|
|
|
|
return Status(error::CHUNKING_ERROR, "Streams are not synchronized.");
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
|
|
|
|
2018-01-09 01:18:15 +00:00
|
|
|
cached_media_sample_stream_data_.push(std::move(stream_data));
|
|
|
|
++num_cached_samples_[stream_index];
|
|
|
|
|
|
|
|
// If we have cached samples from every stream, the first sample in
|
|
|
|
// |cached_media_samples_stream_data_| is guaranteed to be the earliest
|
|
|
|
// sample. Extract and process that sample.
|
|
|
|
if (std::all_of(num_cached_samples_.begin(), num_cached_samples_.end(),
|
|
|
|
[](size_t num_samples) { return num_samples > 0; })) {
|
|
|
|
while (true) {
|
|
|
|
const size_t top_stream_index =
|
|
|
|
cached_media_sample_stream_data_.top()->stream_index;
|
|
|
|
Status status = ProcessMediaSampleStreamData(
|
|
|
|
*cached_media_sample_stream_data_.top());
|
|
|
|
if (!status.ok())
|
|
|
|
return status;
|
|
|
|
cached_media_sample_stream_data_.pop();
|
|
|
|
if (--num_cached_samples_[top_stream_index] == 0)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return Status::OK;
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
|
|
|
default:
|
|
|
|
VLOG(3) << "Stream data type "
|
|
|
|
<< static_cast<int>(stream_data->stream_data_type) << " ignored.";
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return Dispatch(std::move(stream_data));
|
|
|
|
}
|
|
|
|
|
2017-03-03 00:10:30 +00:00
|
|
|
Status ChunkingHandler::OnFlushRequest(size_t input_stream_index) {
|
2018-01-09 01:18:15 +00:00
|
|
|
// Process all cached samples.
|
|
|
|
while (!cached_media_sample_stream_data_.empty()) {
|
|
|
|
Status status =
|
|
|
|
ProcessMediaSampleStreamData(*cached_media_sample_stream_data_.top());
|
|
|
|
if (!status.ok())
|
|
|
|
return status;
|
|
|
|
--num_cached_samples_[cached_media_sample_stream_data_.top()->stream_index];
|
|
|
|
cached_media_sample_stream_data_.pop();
|
|
|
|
}
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
if (segment_info_[input_stream_index]) {
|
|
|
|
auto& segment_info = segment_info_[input_stream_index];
|
|
|
|
if (segment_info->start_timestamp != -1) {
|
|
|
|
segment_info->duration = last_sample_end_timestamps_[input_stream_index] -
|
|
|
|
segment_info->start_timestamp;
|
2018-01-09 01:18:15 +00:00
|
|
|
Status status =
|
|
|
|
DispatchSegmentInfo(input_stream_index, std::move(segment_info));
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
if (!status.ok())
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
}
|
2017-03-17 19:45:28 +00:00
|
|
|
const size_t output_stream_index = input_stream_index;
|
2017-02-22 20:14:26 +00:00
|
|
|
return FlushDownstream(output_stream_index);
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
|
|
|
|
2018-02-12 20:19:34 +00:00
|
|
|
Status ChunkingHandler::OnStreamInfo(uint64_t stream_index,
|
|
|
|
std::shared_ptr<const StreamInfo> info) {
|
|
|
|
// Make sure the inputs come from the same thread.
|
|
|
|
const int64_t thread_id =
|
|
|
|
static_cast<int64_t>(base::PlatformThread::CurrentId());
|
|
|
|
|
|
|
|
int64_t expected = kThreadIdUnset;
|
|
|
|
if (!thread_id_.compare_exchange_strong(expected, thread_id) &&
|
|
|
|
expected != thread_id) {
|
|
|
|
return Status(error::CHUNKING_ERROR,
|
|
|
|
"Inputs should come from the same thread.");
|
|
|
|
}
|
|
|
|
|
|
|
|
const auto time_scale = info->time_scale();
|
|
|
|
time_scales_[stream_index] = time_scale;
|
|
|
|
|
|
|
|
// The video stream is treated as the main stream. If there is only one
|
|
|
|
// stream, it is the main stream.
|
|
|
|
const bool is_main_stream =
|
|
|
|
main_stream_index_ == kInvalidStreamIndex &&
|
|
|
|
(info->stream_type() == kStreamVideo || num_input_streams() == 1);
|
|
|
|
if (is_main_stream) {
|
|
|
|
main_stream_index_ = stream_index;
|
|
|
|
segment_duration_ =
|
|
|
|
chunking_params_.segment_duration_in_seconds * time_scale;
|
|
|
|
subsegment_duration_ =
|
|
|
|
chunking_params_.subsegment_duration_in_seconds * time_scale;
|
|
|
|
} else if (info->stream_type() == kStreamVideo) {
|
|
|
|
return Status(error::CHUNKING_ERROR,
|
|
|
|
"Only one video stream is allowed per chunking handler.");
|
|
|
|
}
|
|
|
|
|
|
|
|
return DispatchStreamInfo(stream_index, std::move(info));
|
|
|
|
}
|
|
|
|
|
2018-01-09 01:18:15 +00:00
|
|
|
Status ChunkingHandler::ProcessMainMediaSample(const MediaSample* sample) {
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
const bool is_key_frame = sample->is_key_frame();
|
|
|
|
const int64_t timestamp = sample->dts();
|
|
|
|
// Check if we need to terminate the current (sub)segment.
|
|
|
|
bool new_segment = false;
|
|
|
|
bool new_subsegment = false;
|
2018-01-03 00:10:33 +00:00
|
|
|
std::shared_ptr<CueEvent> cue_event;
|
2017-07-27 17:23:27 +00:00
|
|
|
if (is_key_frame || !chunking_params_.segment_sap_aligned) {
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
const int64_t segment_index = timestamp / segment_duration_;
|
|
|
|
if (segment_index != current_segment_index_) {
|
|
|
|
current_segment_index_ = segment_index;
|
2017-02-24 01:17:47 +00:00
|
|
|
// Reset subsegment index.
|
|
|
|
current_subsegment_index_ = 0;
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
new_segment = true;
|
|
|
|
}
|
2017-12-12 16:51:50 +00:00
|
|
|
// We use 'while' instead of 'if' to make sure to pop off multiple SCTE35
|
|
|
|
// events that may be very close to each other.
|
|
|
|
while (!scte35_events_.empty() &&
|
|
|
|
(scte35_events_.top()->scte35_event->start_time <= timestamp)) {
|
2018-01-03 00:10:33 +00:00
|
|
|
// For simplicity, don't change |current_segment_index_|.
|
|
|
|
current_subsegment_index_ = 0;
|
|
|
|
new_segment = true;
|
|
|
|
|
|
|
|
cue_event = std::make_shared<CueEvent>();
|
|
|
|
// Use PTS instead of DTS for cue event timestamp.
|
|
|
|
cue_event->timestamp = sample->pts();
|
|
|
|
cue_event->cue_data = scte35_events_.top()->scte35_event->cue_data;
|
2018-01-12 00:55:43 +00:00
|
|
|
LOG(INFO) << "Chunked at " << timestamp << " for Ad Cue.";
|
2018-01-03 00:10:33 +00:00
|
|
|
|
2017-12-12 16:51:50 +00:00
|
|
|
scte35_events_.pop();
|
|
|
|
}
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
|
|
|
if (!new_segment && subsegment_duration_ > 0 &&
|
2017-07-27 17:23:27 +00:00
|
|
|
(is_key_frame || !chunking_params_.subsegment_sap_aligned)) {
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
const int64_t subsegment_index =
|
|
|
|
(timestamp - segment_info_[main_stream_index_]->start_timestamp) /
|
|
|
|
subsegment_duration_;
|
|
|
|
if (subsegment_index != current_subsegment_index_) {
|
|
|
|
current_subsegment_index_ = subsegment_index;
|
|
|
|
new_subsegment = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
Status status;
|
|
|
|
if (new_segment) {
|
|
|
|
status.Update(DispatchSegmentInfoForAllStreams());
|
|
|
|
segment_info_[main_stream_index_]->start_timestamp = timestamp;
|
2018-01-03 00:10:33 +00:00
|
|
|
|
|
|
|
if (cue_event)
|
|
|
|
status.Update(DispatchCueEventForAllStreams(std::move(cue_event)));
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
|
|
|
if (subsegment_duration_ > 0 && (new_segment || new_subsegment)) {
|
|
|
|
status.Update(DispatchSubsegmentInfoForAllStreams());
|
|
|
|
subsegment_info_[main_stream_index_]->start_timestamp = timestamp;
|
|
|
|
}
|
2018-01-09 01:18:15 +00:00
|
|
|
return status;
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
|
|
|
|
2018-01-09 01:18:15 +00:00
|
|
|
Status ChunkingHandler::ProcessMediaSampleStreamData(
|
|
|
|
const StreamData& media_sample_stream_data) {
|
|
|
|
const size_t stream_index = media_sample_stream_data.stream_index;
|
|
|
|
const auto sample = std::move(media_sample_stream_data.media_sample);
|
|
|
|
|
|
|
|
if (stream_index == main_stream_index_) {
|
|
|
|
Status status = ProcessMainMediaSample(sample.get());
|
|
|
|
if (!status.ok())
|
|
|
|
return status;
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
2018-01-09 01:18:15 +00:00
|
|
|
|
|
|
|
VLOG(3) << "Stream index: " << stream_index << " "
|
|
|
|
<< "Sample ts: " << sample->dts() << " "
|
|
|
|
<< " duration: " << sample->duration()
|
|
|
|
<< " scale: " << time_scales_[stream_index] << "\n"
|
|
|
|
<< " scale: " << time_scales_[main_stream_index_]
|
|
|
|
<< (segment_info_[stream_index] ? " dispatch " : " discard ");
|
|
|
|
// Discard samples before segment start. If the segment has started,
|
|
|
|
// |segment_info_[stream_index]| won't be null.
|
|
|
|
if (!segment_info_[stream_index])
|
|
|
|
return Status::OK;
|
|
|
|
if (segment_info_[stream_index]->start_timestamp == -1)
|
|
|
|
segment_info_[stream_index]->start_timestamp = sample->dts();
|
|
|
|
if (subsegment_info_[stream_index] &&
|
|
|
|
subsegment_info_[stream_index]->start_timestamp == -1) {
|
|
|
|
subsegment_info_[stream_index]->start_timestamp = sample->dts();
|
|
|
|
}
|
|
|
|
last_sample_end_timestamps_[stream_index] =
|
|
|
|
sample->dts() + sample->duration();
|
|
|
|
return DispatchMediaSample(stream_index, std::move(sample));
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
Status ChunkingHandler::DispatchSegmentInfoForAllStreams() {
|
|
|
|
Status status;
|
2017-03-03 00:10:30 +00:00
|
|
|
for (size_t i = 0; i < segment_info_.size() && status.ok(); ++i) {
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
if (segment_info_[i] && segment_info_[i]->start_timestamp != -1) {
|
|
|
|
segment_info_[i]->duration =
|
|
|
|
last_sample_end_timestamps_[i] - segment_info_[i]->start_timestamp;
|
|
|
|
status.Update(DispatchSegmentInfo(i, std::move(segment_info_[i])));
|
|
|
|
}
|
|
|
|
segment_info_[i].reset(new SegmentInfo);
|
|
|
|
subsegment_info_[i].reset();
|
|
|
|
}
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status ChunkingHandler::DispatchSubsegmentInfoForAllStreams() {
|
|
|
|
Status status;
|
2017-03-03 00:10:30 +00:00
|
|
|
for (size_t i = 0; i < subsegment_info_.size() && status.ok(); ++i) {
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
if (subsegment_info_[i] && subsegment_info_[i]->start_timestamp != -1) {
|
|
|
|
subsegment_info_[i]->duration =
|
|
|
|
last_sample_end_timestamps_[i] - subsegment_info_[i]->start_timestamp;
|
|
|
|
status.Update(DispatchSegmentInfo(i, std::move(subsegment_info_[i])));
|
|
|
|
}
|
|
|
|
subsegment_info_[i].reset(new SegmentInfo);
|
|
|
|
subsegment_info_[i]->is_subsegment = true;
|
|
|
|
}
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2018-01-03 00:10:33 +00:00
|
|
|
Status ChunkingHandler::DispatchCueEventForAllStreams(
|
|
|
|
std::shared_ptr<CueEvent> cue_event) {
|
|
|
|
Status status;
|
|
|
|
for (size_t i = 0; i < segment_info_.size() && status.ok(); ++i) {
|
|
|
|
std::shared_ptr<CueEvent> new_cue_event(new CueEvent(*cue_event));
|
|
|
|
new_cue_event->timestamp = cue_event->timestamp * time_scales_[i] /
|
|
|
|
time_scales_[main_stream_index_];
|
|
|
|
status.Update(DispatchCueEvent(i, std::move(new_cue_event)));
|
|
|
|
}
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2018-01-09 01:18:15 +00:00
|
|
|
ChunkingHandler::MediaSampleTimestampGreater::MediaSampleTimestampGreater(
|
|
|
|
const ChunkingHandler* const chunking_handler)
|
|
|
|
: chunking_handler_(chunking_handler) {}
|
|
|
|
|
|
|
|
bool ChunkingHandler::MediaSampleTimestampGreater::operator()(
|
|
|
|
const std::unique_ptr<StreamData>& lhs,
|
|
|
|
const std::unique_ptr<StreamData>& rhs) const {
|
|
|
|
DCHECK(lhs);
|
|
|
|
DCHECK(rhs);
|
|
|
|
return GetSampleTimeInSeconds(*lhs) > GetSampleTimeInSeconds(*rhs);
|
|
|
|
}
|
|
|
|
|
|
|
|
double ChunkingHandler::MediaSampleTimestampGreater::GetSampleTimeInSeconds(
|
|
|
|
const StreamData& media_sample_stream_data) const {
|
|
|
|
const size_t stream_index = media_sample_stream_data.stream_index;
|
|
|
|
const auto& sample = media_sample_stream_data.media_sample;
|
|
|
|
DCHECK(sample);
|
|
|
|
// Order main samples by left boundary and non main samples by mid-point. This
|
|
|
|
// ensures non main samples are properly chunked, i.e. if the portion of the
|
|
|
|
// sample in the next chunk is bigger than the portion of the sample in the
|
|
|
|
// previous chunk, the sample is placed in the next chunk.
|
|
|
|
const uint64_t timestamp =
|
|
|
|
stream_index == chunking_handler_->main_stream_index_
|
|
|
|
? sample->dts()
|
|
|
|
: (sample->dts() + sample->duration() / 2);
|
|
|
|
return static_cast<double>(timestamp) /
|
|
|
|
chunking_handler_->time_scales_[stream_index];
|
|
|
|
}
|
|
|
|
|
|
|
|
bool ChunkingHandler::Scte35EventTimestampGreater::operator()(
|
|
|
|
const std::unique_ptr<StreamData>& lhs,
|
|
|
|
const std::unique_ptr<StreamData>& rhs) const {
|
|
|
|
DCHECK(lhs);
|
|
|
|
DCHECK(rhs);
|
|
|
|
DCHECK(lhs->scte35_event);
|
|
|
|
DCHECK(rhs->scte35_event);
|
|
|
|
return lhs->scte35_event->start_time > rhs->scte35_event->start_time;
|
|
|
|
}
|
|
|
|
|
Implement ChunkingHandler
This handler is a multi-in multi-out handler. If more than one input is
provided, there should be one and only one video stream; also, all inputs
should come from the same thread and are synchronized.
There can be multiple chunking handler running in different threads or even
different processes, we use the "consistent chunking algorithm" to make sure
the chunks in different streams are aligned without explicit communcating
with each other - which is not efficient and often difficult.
Consistent Chunking Algorithm:
1. Find the consistent chunkable boundary
Let the timestamps for video frames be (t1, t2, t3, ...). Then a
consistent chunkable boundary is simply the first chunkable boundary after
(tk / N) != (tk-1 / N), where '/' denotes integer division, and N is the
intended chunk duration.
2. Chunk only at the consistent chunkable boundary
This algorithm will make sure the chunks from different video streams are
aligned if they have aligned GoPs. However, this algorithm will only work
for video streams. To be able to chunk non video streams at similar
positions as video streams, ChunkingHandler is designed to accept one video
input and multiple non video inputs, the non video inputs are chunked when
the video input is chunked. If the inputs are synchronized - which is true
if the inputs come from the same demuxer, the video and non video chunks
are aligned.
Change-Id: Id3bad51ab14f311efdb8713b6cd36d36cf9e4639
2017-02-07 18:58:47 +00:00
|
|
|
} // namespace media
|
|
|
|
} // namespace shaka
|