moonfire-nvr/server/src/stream.rs

324 lines
11 KiB
Rust
Raw Normal View History

// This file is part of Moonfire NVR, a security camera network video recorder.
// Copyright (C) 2016 The Moonfire NVR Authors; see AUTHORS and LICENSE.txt.
// SPDX-License-Identifier: GPL-v3.0-or-later WITH GPL-3.0-linking-exception.
Rust rewrite I should have submitted/pushed more incrementally but just played with it on my computer as I was learning the language. The new Rust version more or less matches the functionality of the current C++ version, although there are many caveats listed below. Upgrade notes: when moving from the C++ version, I recommend dropping and recreating the "recording_cover" index in SQLite3 to pick up the addition of the "video_sync_samples" column: $ sudo systemctl stop moonfire-nvr $ sudo -u moonfire-nvr sqlite3 /var/lib/moonfire-nvr/db/db sqlite> drop index recording_cover; sqlite3> create index ...rest of command as in schema.sql...; sqlite3> ^D Some known visible differences from the C++ version: * .mp4 generation queries SQLite3 differently. Before it would just get all video indexes in a single query. Now it leads with a query that should be satisfiable by the covering index (assuming the index has been recreated as noted above), then queries individual recording's indexes as needed to fill a LRU cache. I believe this is roughly similar speed for the initial hit (which generates the moov part of the file) and significantly faster when seeking. I would have done it a while ago with the C++ version but didn't want to track down a lru cache library. It was easier to find with Rust. * On startup, the Rust version cleans up old reserved files. This is as in the design; the C++ version was just missing this code. * The .html recording list output is a little different. It's in ascending order, with the most current segment shorten than an hour rather than the oldest. This is less ergonomic, but it was easy. I could fix it or just wait to obsolete it with some fancier JavaScript UI. * commandline argument parsing and logging have changed formats due to different underlying libraries. * The JSON output isn't quite right (matching the spec / C++ implementation) yet. Additional caveats: * I haven't done any proof-reading of prep.sh + install instructions. * There's a lot of code quality work to do: adding (back) comments and test coverage, developing a good Rust style. * The ffmpeg foreign function interface is particularly sketchy. I'd eventually like to switch to something based on autogenerated bindings. I'd also like to use pure Rust code where practical, but once I do on-NVR motion detection I'll need to existing C/C++ libraries for speed (H.264 decoding + OpenCL-based analysis).
2016-11-25 14:34:00 -08:00
use crate::h264;
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
use bytes::Bytes;
use failure::format_err;
use failure::{bail, Error};
use futures::StreamExt;
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
use retina::client::Demuxed;
use retina::codec::CodecItem;
use std::pin::Pin;
Rust rewrite I should have submitted/pushed more incrementally but just played with it on my computer as I was learning the language. The new Rust version more or less matches the functionality of the current C++ version, although there are many caveats listed below. Upgrade notes: when moving from the C++ version, I recommend dropping and recreating the "recording_cover" index in SQLite3 to pick up the addition of the "video_sync_samples" column: $ sudo systemctl stop moonfire-nvr $ sudo -u moonfire-nvr sqlite3 /var/lib/moonfire-nvr/db/db sqlite> drop index recording_cover; sqlite3> create index ...rest of command as in schema.sql...; sqlite3> ^D Some known visible differences from the C++ version: * .mp4 generation queries SQLite3 differently. Before it would just get all video indexes in a single query. Now it leads with a query that should be satisfiable by the covering index (assuming the index has been recreated as noted above), then queries individual recording's indexes as needed to fill a LRU cache. I believe this is roughly similar speed for the initial hit (which generates the moov part of the file) and significantly faster when seeking. I would have done it a while ago with the C++ version but didn't want to track down a lru cache library. It was easier to find with Rust. * On startup, the Rust version cleans up old reserved files. This is as in the design; the C++ version was just missing this code. * The .html recording list output is a little different. It's in ascending order, with the most current segment shorten than an hour rather than the oldest. This is less ergonomic, but it was easy. I could fix it or just wait to obsolete it with some fancier JavaScript UI. * commandline argument parsing and logging have changed formats due to different underlying libraries. * The JSON output isn't quite right (matching the spec / C++ implementation) yet. Additional caveats: * I haven't done any proof-reading of prep.sh + install instructions. * There's a lot of code quality work to do: adding (back) comments and test coverage, developing a good Rust style. * The ffmpeg foreign function interface is particularly sketchy. I'd eventually like to switch to something based on autogenerated bindings. I'd also like to use pure Rust code where practical, but once I do on-NVR motion detection I'll need to existing C/C++ libraries for speed (H.264 decoding + OpenCL-based analysis).
2016-11-25 14:34:00 -08:00
use std::result::Result;
use url::Url;
Rust rewrite I should have submitted/pushed more incrementally but just played with it on my computer as I was learning the language. The new Rust version more or less matches the functionality of the current C++ version, although there are many caveats listed below. Upgrade notes: when moving from the C++ version, I recommend dropping and recreating the "recording_cover" index in SQLite3 to pick up the addition of the "video_sync_samples" column: $ sudo systemctl stop moonfire-nvr $ sudo -u moonfire-nvr sqlite3 /var/lib/moonfire-nvr/db/db sqlite> drop index recording_cover; sqlite3> create index ...rest of command as in schema.sql...; sqlite3> ^D Some known visible differences from the C++ version: * .mp4 generation queries SQLite3 differently. Before it would just get all video indexes in a single query. Now it leads with a query that should be satisfiable by the covering index (assuming the index has been recreated as noted above), then queries individual recording's indexes as needed to fill a LRU cache. I believe this is roughly similar speed for the initial hit (which generates the moov part of the file) and significantly faster when seeking. I would have done it a while ago with the C++ version but didn't want to track down a lru cache library. It was easier to find with Rust. * On startup, the Rust version cleans up old reserved files. This is as in the design; the C++ version was just missing this code. * The .html recording list output is a little different. It's in ascending order, with the most current segment shorten than an hour rather than the oldest. This is less ergonomic, but it was easy. I could fix it or just wait to obsolete it with some fancier JavaScript UI. * commandline argument parsing and logging have changed formats due to different underlying libraries. * The JSON output isn't quite right (matching the spec / C++ implementation) yet. Additional caveats: * I haven't done any proof-reading of prep.sh + install instructions. * There's a lot of code quality work to do: adding (back) comments and test coverage, developing a good Rust style. * The ffmpeg foreign function interface is particularly sketchy. I'd eventually like to switch to something based on autogenerated bindings. I'd also like to use pure Rust code where practical, but once I do on-NVR motion detection I'll need to existing C/C++ libraries for speed (H.264 decoding + OpenCL-based analysis).
2016-11-25 14:34:00 -08:00
static RETINA_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(30);
2022-05-17 16:36:39 -07:00
pub struct Options {
pub session: retina::client::SessionOptions,
pub setup: retina::client::SetupOptions,
}
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
/// Opens a RTSP stream. This is a trait for test injection.
pub trait Opener: Send + Sync {
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
/// Opens the given RTSP URL.
///
/// Note: despite the blocking interface, this expects to be called from
/// the context of a multithreaded tokio runtime with IO and time enabled.
2022-05-17 16:36:39 -07:00
fn open(&self, label: String, url: Url, options: Options) -> Result<Box<dyn Stream>, Error>;
}
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
pub struct VideoFrame {
pub pts: i64,
/// An estimate of the duration of the frame, or zero.
/// This can be deceptive and is only used by some testing code.
pub duration: i32,
pub is_key: bool,
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
pub data: Bytes,
pub new_video_sample_entry: bool,
}
pub trait Stream: Send {
fn tool(&self) -> Option<&retina::client::Tool>;
fn video_sample_entry(&self) -> &db::VideoSampleEntryToInsert;
fn next(&mut self) -> Result<VideoFrame, Error>;
}
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
pub struct RealOpener;
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
pub const OPENER: RealOpener = RealOpener;
Rust rewrite I should have submitted/pushed more incrementally but just played with it on my computer as I was learning the language. The new Rust version more or less matches the functionality of the current C++ version, although there are many caveats listed below. Upgrade notes: when moving from the C++ version, I recommend dropping and recreating the "recording_cover" index in SQLite3 to pick up the addition of the "video_sync_samples" column: $ sudo systemctl stop moonfire-nvr $ sudo -u moonfire-nvr sqlite3 /var/lib/moonfire-nvr/db/db sqlite> drop index recording_cover; sqlite3> create index ...rest of command as in schema.sql...; sqlite3> ^D Some known visible differences from the C++ version: * .mp4 generation queries SQLite3 differently. Before it would just get all video indexes in a single query. Now it leads with a query that should be satisfiable by the covering index (assuming the index has been recreated as noted above), then queries individual recording's indexes as needed to fill a LRU cache. I believe this is roughly similar speed for the initial hit (which generates the moov part of the file) and significantly faster when seeking. I would have done it a while ago with the C++ version but didn't want to track down a lru cache library. It was easier to find with Rust. * On startup, the Rust version cleans up old reserved files. This is as in the design; the C++ version was just missing this code. * The .html recording list output is a little different. It's in ascending order, with the most current segment shorten than an hour rather than the oldest. This is less ergonomic, but it was easy. I could fix it or just wait to obsolete it with some fancier JavaScript UI. * commandline argument parsing and logging have changed formats due to different underlying libraries. * The JSON output isn't quite right (matching the spec / C++ implementation) yet. Additional caveats: * I haven't done any proof-reading of prep.sh + install instructions. * There's a lot of code quality work to do: adding (back) comments and test coverage, developing a good Rust style. * The ffmpeg foreign function interface is particularly sketchy. I'd eventually like to switch to something based on autogenerated bindings. I'd also like to use pure Rust code where practical, but once I do on-NVR motion detection I'll need to existing C/C++ libraries for speed (H.264 decoding + OpenCL-based analysis).
2016-11-25 14:34:00 -08:00
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
impl Opener for RealOpener {
fn open(
&self,
label: String,
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
url: Url,
2022-05-17 16:36:39 -07:00
mut options: Options,
) -> Result<Box<dyn Stream>, Error> {
2022-05-17 16:36:39 -07:00
options.session = options
.session
.user_agent(format!("Moonfire NVR {}", env!("CARGO_PKG_VERSION")));
let rt_handle = tokio::runtime::Handle::current();
let (inner, first_frame) = rt_handle
2022-04-13 13:22:26 -07:00
.block_on(rt_handle.spawn(tokio::time::timeout(
RETINA_TIMEOUT,
RetinaStreamInner::play(label, url, options),
)))
.expect("RetinaStream::play task panicked, see earlier error")??;
Ok(Box::new(RetinaStream {
2022-04-13 13:22:26 -07:00
inner: Some(inner),
rt_handle,
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
first_frame: Some(first_frame),
}))
Rust rewrite I should have submitted/pushed more incrementally but just played with it on my computer as I was learning the language. The new Rust version more or less matches the functionality of the current C++ version, although there are many caveats listed below. Upgrade notes: when moving from the C++ version, I recommend dropping and recreating the "recording_cover" index in SQLite3 to pick up the addition of the "video_sync_samples" column: $ sudo systemctl stop moonfire-nvr $ sudo -u moonfire-nvr sqlite3 /var/lib/moonfire-nvr/db/db sqlite> drop index recording_cover; sqlite3> create index ...rest of command as in schema.sql...; sqlite3> ^D Some known visible differences from the C++ version: * .mp4 generation queries SQLite3 differently. Before it would just get all video indexes in a single query. Now it leads with a query that should be satisfiable by the covering index (assuming the index has been recreated as noted above), then queries individual recording's indexes as needed to fill a LRU cache. I believe this is roughly similar speed for the initial hit (which generates the moov part of the file) and significantly faster when seeking. I would have done it a while ago with the C++ version but didn't want to track down a lru cache library. It was easier to find with Rust. * On startup, the Rust version cleans up old reserved files. This is as in the design; the C++ version was just missing this code. * The .html recording list output is a little different. It's in ascending order, with the most current segment shorten than an hour rather than the oldest. This is less ergonomic, but it was easy. I could fix it or just wait to obsolete it with some fancier JavaScript UI. * commandline argument parsing and logging have changed formats due to different underlying libraries. * The JSON output isn't quite right (matching the spec / C++ implementation) yet. Additional caveats: * I haven't done any proof-reading of prep.sh + install instructions. * There's a lot of code quality work to do: adding (back) comments and test coverage, developing a good Rust style. * The ffmpeg foreign function interface is particularly sketchy. I'd eventually like to switch to something based on autogenerated bindings. I'd also like to use pure Rust code where practical, but once I do on-NVR motion detection I'll need to existing C/C++ libraries for speed (H.264 decoding + OpenCL-based analysis).
2016-11-25 14:34:00 -08:00
}
}
Rust rewrite I should have submitted/pushed more incrementally but just played with it on my computer as I was learning the language. The new Rust version more or less matches the functionality of the current C++ version, although there are many caveats listed below. Upgrade notes: when moving from the C++ version, I recommend dropping and recreating the "recording_cover" index in SQLite3 to pick up the addition of the "video_sync_samples" column: $ sudo systemctl stop moonfire-nvr $ sudo -u moonfire-nvr sqlite3 /var/lib/moonfire-nvr/db/db sqlite> drop index recording_cover; sqlite3> create index ...rest of command as in schema.sql...; sqlite3> ^D Some known visible differences from the C++ version: * .mp4 generation queries SQLite3 differently. Before it would just get all video indexes in a single query. Now it leads with a query that should be satisfiable by the covering index (assuming the index has been recreated as noted above), then queries individual recording's indexes as needed to fill a LRU cache. I believe this is roughly similar speed for the initial hit (which generates the moov part of the file) and significantly faster when seeking. I would have done it a while ago with the C++ version but didn't want to track down a lru cache library. It was easier to find with Rust. * On startup, the Rust version cleans up old reserved files. This is as in the design; the C++ version was just missing this code. * The .html recording list output is a little different. It's in ascending order, with the most current segment shorten than an hour rather than the oldest. This is less ergonomic, but it was easy. I could fix it or just wait to obsolete it with some fancier JavaScript UI. * commandline argument parsing and logging have changed formats due to different underlying libraries. * The JSON output isn't quite right (matching the spec / C++ implementation) yet. Additional caveats: * I haven't done any proof-reading of prep.sh + install instructions. * There's a lot of code quality work to do: adding (back) comments and test coverage, developing a good Rust style. * The ffmpeg foreign function interface is particularly sketchy. I'd eventually like to switch to something based on autogenerated bindings. I'd also like to use pure Rust code where practical, but once I do on-NVR motion detection I'll need to existing C/C++ libraries for speed (H.264 decoding + OpenCL-based analysis).
2016-11-25 14:34:00 -08:00
2022-04-13 13:22:26 -07:00
/// Real stream, implemented with the Retina library.
///
/// Retina is asynchronous and tokio-based where currently Moonfire expects
/// a synchronous stream interface. This blocks on the tokio operations.
///
/// Experimentally, it appears faster to have one thread hand-off per frame via
/// `handle.block_on(handle.spawn(...))` rather than the same without the
/// `handle.spawn(...)`. See
/// [#206](https://github.com/scottlamb/moonfire-nvr/issues/206).
struct RetinaStream {
2022-04-13 13:22:26 -07:00
/// The actual stream details used from within the tokio reactor.
///
/// Spawned tokio tasks must be `'static`, so ownership is passed to the
/// task, and then returned when it completes.
inner: Option<Box<RetinaStreamInner>>,
rt_handle: tokio::runtime::Handle,
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
/// The first frame, if not yet returned from `next`.
///
/// This frame is special because we sometimes need to fetch it as part of getting the video
/// parameters.
first_frame: Option<retina::codec::VideoFrame>,
}
2022-04-13 13:22:26 -07:00
struct RetinaStreamInner {
label: String,
session: Demuxed,
video_sample_entry: db::VideoSampleEntryToInsert,
2022-04-13 13:22:26 -07:00
}
impl RetinaStreamInner {
/// Plays to first frame. No timeout; that's the caller's responsibility.
async fn play(
2022-04-13 13:22:26 -07:00
label: String,
url: Url,
2022-05-17 16:36:39 -07:00
options: Options,
) -> Result<(Box<Self>, retina::codec::VideoFrame), Error> {
2022-05-17 16:36:39 -07:00
let mut session = retina::client::Session::describe(url, options.session).await?;
2022-04-13 13:22:26 -07:00
log::debug!("connected to {:?}, tool {:?}", &label, session.tool());
2022-05-17 16:36:39 -07:00
let video_i = session
.streams()
.iter()
2022-05-17 16:36:39 -07:00
.position(|s| s.media() == "video" && s.encoding_name() == "h264")
.ok_or_else(|| format_err!("couldn't find H.264 video stream"))?;
2022-05-17 16:36:39 -07:00
session.setup(video_i, options.setup).await?;
let session = session.play(retina::client::PlayOptions::default()).await?;
2022-04-13 13:22:26 -07:00
let mut session = session.demuxed()?;
// First frame.
let first_frame = loop {
2022-04-13 13:22:26 -07:00
match Pin::new(&mut session).next().await {
None => bail!("stream closed before first frame"),
Some(Err(e)) => return Err(e.into()),
2022-05-17 16:36:39 -07:00
Some(Ok(CodecItem::VideoFrame(v))) => {
if v.is_random_access_point() {
break v;
}
}
Some(Ok(_)) => {}
}
};
2022-05-17 16:36:39 -07:00
let video_params = match session.streams()[video_i].parameters() {
Some(retina::codec::ParametersRef::Video(v)) => v.clone(),
Some(_) => unreachable!(),
None => bail!("couldn't find H.264 parameters"),
};
let video_sample_entry = h264::parse_extra_data(video_params.extra_data())?;
let self_ = Box::new(Self {
label,
session,
video_sample_entry,
});
Ok((self_, first_frame))
}
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
/// Fetches a non-initial frame.
async fn fetch_next_frame(
2022-04-13 13:22:26 -07:00
mut self: Box<Self>,
2022-05-17 16:36:39 -07:00
) -> Result<
(
Box<Self>,
retina::codec::VideoFrame,
Option<retina::codec::VideoParameters>,
),
Error,
> {
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
loop {
2022-04-13 13:22:26 -07:00
match Pin::new(&mut self.session).next().await.transpose()? {
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
None => bail!("end of stream"),
Some(CodecItem::VideoFrame(v)) => {
2022-05-17 16:36:39 -07:00
if v.loss() > 0 {
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
log::warn!(
"{}: lost {} RTP packets @ {}",
2022-04-13 13:22:26 -07:00
&self.label,
2022-05-17 16:36:39 -07:00
v.loss(),
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
v.start_ctx()
);
}
2022-05-17 16:36:39 -07:00
let p = if v.has_new_parameters() {
Some(match self.session.streams()[v.stream_id()].parameters() {
Some(retina::codec::ParametersRef::Video(v)) => v.clone(),
_ => unreachable!(),
})
} else {
None
};
return Ok((self, v, p));
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
}
Some(_) => {}
}
}
}
}
impl Stream for RetinaStream {
fn tool(&self) -> Option<&retina::client::Tool> {
2022-04-13 13:22:26 -07:00
self.inner.as_ref().unwrap().session.tool()
}
fn video_sample_entry(&self) -> &db::VideoSampleEntryToInsert {
&self.inner.as_ref().unwrap().video_sample_entry
}
fn next(&mut self) -> Result<VideoFrame, Error> {
let (frame, new_video_sample_entry) = self
.first_frame
.take()
.map(|f| Ok((f, false)))
.unwrap_or_else(move || {
let inner = self.inner.take().unwrap();
2022-05-17 16:36:39 -07:00
let (mut inner, frame, new_parameters) = self
.rt_handle
.block_on(self.rt_handle.spawn(tokio::time::timeout(
RETINA_TIMEOUT,
inner.fetch_next_frame(),
)))
.expect("fetch_next_frame task panicked, see earlier error")
.map_err(|_| format_err!("timeout getting next frame"))??;
let mut new_video_sample_entry = false;
2022-05-17 16:36:39 -07:00
if let Some(p) = new_parameters {
let video_sample_entry = h264::parse_extra_data(p.extra_data())?;
if video_sample_entry != inner.video_sample_entry {
log::debug!(
"{}: parameter change:\nold: {:?}\nnew: {:?}",
&inner.label,
&inner.video_sample_entry,
&video_sample_entry
);
inner.video_sample_entry = video_sample_entry;
new_video_sample_entry = true;
}
};
self.inner = Some(inner);
Ok::<_, failure::Error>((frame, new_video_sample_entry))
})?;
Ok(VideoFrame {
2022-05-17 16:36:39 -07:00
pts: frame.timestamp().elapsed(),
duration: 0,
2022-05-17 16:36:39 -07:00
is_key: frame.is_random_access_point(),
data: frame.into_data().into(),
new_video_sample_entry,
})
Rust rewrite I should have submitted/pushed more incrementally but just played with it on my computer as I was learning the language. The new Rust version more or less matches the functionality of the current C++ version, although there are many caveats listed below. Upgrade notes: when moving from the C++ version, I recommend dropping and recreating the "recording_cover" index in SQLite3 to pick up the addition of the "video_sync_samples" column: $ sudo systemctl stop moonfire-nvr $ sudo -u moonfire-nvr sqlite3 /var/lib/moonfire-nvr/db/db sqlite> drop index recording_cover; sqlite3> create index ...rest of command as in schema.sql...; sqlite3> ^D Some known visible differences from the C++ version: * .mp4 generation queries SQLite3 differently. Before it would just get all video indexes in a single query. Now it leads with a query that should be satisfiable by the covering index (assuming the index has been recreated as noted above), then queries individual recording's indexes as needed to fill a LRU cache. I believe this is roughly similar speed for the initial hit (which generates the moov part of the file) and significantly faster when seeking. I would have done it a while ago with the C++ version but didn't want to track down a lru cache library. It was easier to find with Rust. * On startup, the Rust version cleans up old reserved files. This is as in the design; the C++ version was just missing this code. * The .html recording list output is a little different. It's in ascending order, with the most current segment shorten than an hour rather than the oldest. This is less ergonomic, but it was easy. I could fix it or just wait to obsolete it with some fancier JavaScript UI. * commandline argument parsing and logging have changed formats due to different underlying libraries. * The JSON output isn't quite right (matching the spec / C++ implementation) yet. Additional caveats: * I haven't done any proof-reading of prep.sh + install instructions. * There's a lot of code quality work to do: adding (back) comments and test coverage, developing a good Rust style. * The ffmpeg foreign function interface is particularly sketchy. I'd eventually like to switch to something based on autogenerated bindings. I'd also like to use pure Rust code where practical, but once I do on-NVR motion detection I'll need to existing C/C++ libraries for speed (H.264 decoding + OpenCL-based analysis).
2016-11-25 14:34:00 -08:00
}
}
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
#[cfg(test)]
pub mod testutil {
use super::*;
use std::convert::TryFrom;
use std::io::Cursor;
pub struct Mp4Stream {
reader: mp4::Mp4Reader<Cursor<Vec<u8>>>,
h264_track_id: u32,
next_sample_id: u32,
video_sample_entry: db::VideoSampleEntryToInsert,
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
}
impl Mp4Stream {
/// Opens a stream, with a return matching that expected by [`Opener`].
pub fn open(path: &str) -> Result<Self, Error> {
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
let f = std::fs::read(path)?;
let len = f.len();
let reader = mp4::Mp4Reader::read_header(Cursor::new(f), u64::try_from(len)?)?;
let h264_track = match reader
.tracks()
.values()
.find(|t| matches!(t.media_type(), Ok(mp4::MediaType::H264)))
{
None => bail!("expected a H.264 track"),
Some(t) => t,
};
let video_sample_entry = h264::parse_extra_data(&h264_track.extra_data()?[..])?;
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
let h264_track_id = h264_track.track_id();
let stream = Mp4Stream {
reader,
h264_track_id,
next_sample_id: 1,
video_sample_entry,
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
};
Ok(stream)
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
}
pub fn duration(&self) -> u64 {
self.reader.moov.mvhd.duration
}
/// Returns the edit list from the H.264 stream, if any.
pub fn elst(&self) -> Option<&mp4::mp4box::elst::ElstBox> {
let h264_track = self.reader.tracks().get(&self.h264_track_id).unwrap();
h264_track
.trak
.edts
.as_ref()
.and_then(|edts| edts.elst.as_ref())
}
}
impl Stream for Mp4Stream {
fn tool(&self) -> Option<&retina::client::Tool> {
None
}
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
fn next(&mut self) -> Result<VideoFrame, Error> {
let sample = self
.reader
.read_sample(self.h264_track_id, self.next_sample_id)?
.ok_or_else(|| format_err!("End of file"))?;
self.next_sample_id += 1;
Ok(VideoFrame {
pts: sample.start_time as i64,
duration: sample.duration as i32,
is_key: sample.is_sync,
data: sample.bytes,
new_video_sample_entry: false,
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
})
}
fn video_sample_entry(&self) -> &db::VideoSampleEntryToInsert {
&self.video_sample_entry
}
drop ffmpeg support * switch the config interface over to use Retina and make the test button honor rtsp_transport = udp. * adjust the threading model of the Retina streaming code. Before, it spawned a background future that read from the runtime and wrote to a channel. Other calls read from this channel. After, it does work directly from within the block_on calls (no channels). The immediate motivation was that the config interface didn't have another runtime handy. And passing in a current thread runtime deadlocked. I later learned this is a difference between Runtime::block_on and Handle::block_on. The former will drive IO and timers; the latter will not. But this is also more efficient to avoid so many thread hand-offs. Both the context switches and the extra spinning that tokio appears to do as mentioned here: https://github.com/scottlamb/retina/issues/5#issuecomment-871971550 This may not be the final word on the threading model. Eventually I may not have per-stream writing threads at all. But I think it will be easier to look at this after getting rid of the separate `moonfire-nvr config` subcommand in favor of a web interface. * in tests, read `.mp4` files via the `mp4` crate rather than ffmpeg. The annoying part is that this doesn't parse edit lists; oh well. * simplify the `Opener` interface. Formerly, it'd take either a RTSP URL or a path to a `.mp4` file, and they'd share some code because they both sometimes used ffmpeg. Now, they're totally different libraries (`retina` vs `mp4`). Pull the latter out to a `testutil` module with a different interface that exposes more of the `mp4` stuff. Now `Opener` is just for RTSP. * simplify the h264 module. It had a lot of logic to deal with Annex B. Retina doesn't use this encoding. Fixes #36 Fixes #126
2022-03-18 10:30:23 -07:00
}
}