Even with the filter changes made, there's still some cases where
filter chain hangs can result from filter reconfigurations. To solve the
issue completely, I have split out pre-processing video files to
separate ffmpeg processes, so that the filter chain for compositing will
not ever be reconfigured.
Each input video now has a separate ffmpeg process run for it which
does the scaling, padding, and video extending steps. To avoid issues
with disk space usage and extra cpu usage or quality loss, the output
from these separate processes is sent to the compositing ffmpeg process
as uncompressed video in a pipe. To simplify the setup, named pipes
(special files) are used rather than setting up pipes in the ruby code
programmatically.
The extra ffmpeg processes are configured to log to files, and when
complete their log output is copied to the recording processing log.
Processes are joined to ensure zombie processes are not created, and
the return codes of all the processes are checked so errors can be
detected.
Due to the overhead of transferring video through pipes, this might
be a bit slower than the 2.4 recording processing - but on the other
hand, some of the video decoding and scaling happens in parallel, so it
might balance out.
Even with the filter changes made, there's still some cases where
filter chain hangs can result from filter reconfigurations. To solve the
issue completely, I have split out pre-processing video files to
separate ffmpeg processes, so that the filter chain for compositing will
not ever be reconfigured.
Each input video now has a separate ffmpeg process run for it which
does the scaling, padding, and video extending steps. To avoid issues
with disk space usage and extra cpu usage or quality loss, the output
from these separate processes is sent to the compositing ffmpeg process
as uncompressed video in a pipe. To simplify the setup, named pipes
(special files) are used rather than setting up pipes in the ruby code
programmatically.
The extra ffmpeg processes are configured to log to files, and when
complete their log output is copied to the recording processing log.
Processes are joined to ensure zombie processes are not created, and
the return codes of all the processes are checked so errors can be
detected.
Due to the overhead of transferring video through pipes, this might
be a bit slower than the 2.4 recording processing - but on the other
hand, some of the video decoding and scaling happens in parallel, so it
might balance out.
Because the input videos for BigBlueButton recording processing switch
resolution and aspect ratio, the filter chain gets re-initialized, and
any state in the filters is lost. This causes problems with the
following filters:
`color`: Timestamps restart from 0, rather than continuing at the point
where they left off.
`fps=start_time=12.345`: After reset, the fps filter thinks it's at the
start of the file again, so the next frame it sees gets duplicated
output for timestamps from the `start_time` until it catches back up.
`setpts=PTS-STARTPTS`: The 'STARTPTS' is re-read as the first pts the
filter sees after reinitialization, so timestamp of the next frame is
reset to 0. (In practise, this didn't cause any problems because the
duplicate frames created by the fps filter had the original start time.)
The end result of all of these issues is that a lot of duplicate frames
were created with invalid timestamps, which eventually get discarded
by ffmpeg. But a lot of time is wasted, causing recordings to sometimes
take hours to process when they should be ready in minutes.
The fixes are as follows:
* The `color` filters are used to generate the background and
substitutes for missing videos. Move them out to separate filter
chains by using the 'lavfi' input format, which lets you use a filter
as if it was an input file.
* Rather than use the `fps` filter's `start_time` feature, use the
`trim` filter to remove early frames.
* The actual start pts is already known by the script, so replace
`setpts=PTS-STARTPTS` with `setpts=PTS-12.345/TB`, substituting in the
absolute time.
Because the input videos for BigBlueButton recording processing switch
resolution and aspect ratio, the filter chain gets re-initialized, and
any state in the filters is lost. This causes problems with the
following filters:
`color`: Timestamps restart from 0, rather than continuing at the point
where they left off.
`fps=start_time=12.345`: After reset, the fps filter thinks it's at the
start of the file again, so the next frame it sees gets duplicated
output for timestamps from the `start_time` until it catches back up.
`setpts=PTS-STARTPTS`: The 'STARTPTS' is re-read as the first pts the
filter sees after reinitialization, so timestamp of the next frame is
reset to 0. (In practise, this didn't cause any problems because the
duplicate frames created by the fps filter had the original start time.)
The end result of all of these issues is that a lot of duplicate frames
were created with invalid timestamps, which eventually get discarded
by ffmpeg. But a lot of time is wasted, causing recordings to sometimes
take hours to process when they should be ready in minutes.
The fixes are as follows:
* The `color` filters are used to generate the background and
substitutes for missing videos. Move them out to separate filter
chains by using the 'lavfi' input format, which lets you use a filter
as if it was an input file.
* Rather than use the `fps` filter's `start_time` feature, use the
`trim` filter to remove early frames.
* The actual start pts is already known by the script, so replace
`setpts=PTS-STARTPTS` with `setpts=PTS-12.345/TB`, substituting in the
absolute time.
When converting from using the 'movie' source filter to using separate
ffmpeg command line inputs, it wasn't taken into account that the
'movie' filter passes through timestamps from the source file, while the
ffmpeg input adjusts timestamps so that '0' is the selected seek point.
The easiest fix is to add an option to the ffmpeg command to disable the
timestamp adjustments.
Fixes#15644 (This needs to go into BigBlueButton 2.5 and 2.6)
- Update the version to 2.6.0 to ease the detection of old/new whiteboard events
- Fix recording cursor when there is no pan/zoom and annotations in tldraw
- Don't generate slides pngs for 2.6.0, they are not used anymore in playback (svgs instead).
Changed the names of tldraw record events to differentiate from before.
Publish tldraw.json file with all shape information during the meeting to be used in playback.
Adapted cursor.xml and panzoom.xml to store tldraw data.
Publish slides svgs to be used by playback's tldraw component (otherwise we have different image sizes in pngs and thus messing the coordinates).
Retro-compatible with old recordings.
Based on work done by Tiago Jacobs and Guilherme Pereira Leme to
investigate the behaviour of ffmpeg on videos with changing input size.
Rather than having the EDL code set fixed sizes for the scale and pad
filters, instead use ffmpeg's built in features to calculate the scale
and pad automatically, dynamically updating if the input video size
changes.
In the version of ffmpeg in Ubuntu 20.04, the 'movie' input filter is
buggy and doesn't correctly handle the video size changing. Instead of
using the movie filter, use separate inputs to the ffmpeg command (the
design used here is based on the code in audio.rb). The filter chain is
now stored into a file (using -filter_complex_script) to reduce problems
with the command line getting too long.
We are using a version of ffmpeg that's new enough to have the tpad
filter now too, so use it to extend the video if needed instead of an
extra concat.
In BBB 2.5, we switched the recording system to use bundled gems
included privately in the recording package, rather than installed
system-wide. The rap-enqueue.rb script needs to be updated to load the
bundler gems.
According to bundler devs, setting the BUNDLE_GEMFILE environement
variable is the supported way to tell bundler where to find it
(otherwise bundler will search starting at the current working directory
- which in the case of rap-enqueue.rb is probably nowhere near the
Gemfile).
Use a relative path from the directory where the script is located so it
can be run both when installed and from a development environment.
Switch the script interpreter to use /usr/bin/env to load ruby from the
path. Doesn't make a difference in the installed package, but it makes
testing on development systems with multiple ruby environments easier.
Fixes#15085
* fix unit name: the unit name on Ubuntu is `redis-server.service`
* services which need a working redis require both After= and Wants=
See the description in the `systemd.unit` man page.
In some cases with incomplete/partially corrupt files, the input video
file can be shorter than the displayed time. If there is a badly timed
cut, this can result in a seek being generated to a point where ffmpeg
is unable to start at.
Add a detection for this situation, and replace with a blank video.
The rap scripts might load or run some scripts using relative paths from
the scripts directory, so restore that.
Bundler automatically looks up in parent dirs to find the Gemfile, so
loading gems will work correctly.
Several scripts internally run bundler setup, so no explicit bundler
command is needed. For the others, start up using /usr/bin/bundle
(installed by ruby-bundler) to load the environment.
Now uses Ubuntu's bundler version to install all dependencies at build time
rather than install time. Gems are also now vendored, and no longer pollute the
operating system.
Move all Etherpad's access control from Meteor to a separated [Node application](https://github.com/bigbluebutton/bbb-pads).
This new app uses [Etherpad's API](https://etherpad.org/doc/v1.8.4/#index_overview)
to create groups and manage session tokens for users to access them. Each group
represents one distinct pad at the html5 client.
- Removed locked users' access to pads: replaced readOnly pad's access with a new pad's content sharing routine
- Pad's access is now controlled by [Etherpad's API](https://etherpad.org/doc/v1.8.4/#index_overview)
- Closed captions edited content now reflects at it's live feedback
- Improved closed caption's dictation mode live feedback
- Moved all Etherpad's API control from Meteor to a separated [app](https://github.com/bigbluebutton/bbb-pads)
- Included access control both in akka-apps and bbb-pads
There's an issue sometimes where when processing the audio from a
desktop sharing file, the STARTPTS value is invalid somehow, and this
results in bad timestamps in the output stream. Depending on the
selected codec/container combination, this might result in errors such
as spam of the message:
Application provided invalid, non monotonically increasing dts to muxer in stream
or simply a hang during processing.
Since we're already using aresample in async mode to correct timestamp
gaps, we can use "asetpts=N" here to simply set the audio frame pts
values to the sample number directly, giving proper monotonic timestamps
with no gaps.
I was going to resort to some trickery to make mediasoup raw files end up in the
same directory as KMS to reduce changes, but it ended up being too dirty.
I am adding a third directory (/var/mediasoup) to be tracked by the rap scripts
which is where the new raw files end up in.
Move the handling of chat events into the shared library so it can be
used by multiple recording formats.
The anonymization of names is based on the external user id, if
available, so users have a consistent name through the meeting. Note
that no effort is made to edit chat messages - if someone is mentioned
by name in a chat message, that will still be visible.
Default settings for anonymization can be controlled in
bigbluebutton.yml, and per-meeting overrides can be done using meta
parameters on the create call.
Disabling audio or video processing isn't really something that's part
of the working format (and at some point we might want to combine
audio+video processing together).
Move the setting of the '-an' and '-vn' options to where they're
required - as a detail of the EDL processing for video-only and
audio-only components of the output.
Honestly, the main reason for this change is that when testing alternate
working formats, I had accidentally dropped the '-vn' option from the
FFMPEG_WF_ARGS variable without noticing.
Generate a external_videos.json file at the recordings with an array of
played external videos url and timestamp.
This file will be published along with the other presentation format files
and can be used to display at the playback.
Previously, bbb-record --rebuild was restarting recording processing
from scratch by creating the .../recording/<meeting_id>.done file. This
causes the recording to be reprocessed starting at the archive step.
However, re-running the archive step for an existing meeting is not
really supported! Ever since the segmented recording code was added, it
shouldn't /corrupt/ the recording files, but it's still not good.
And as a side-effect, re-running the archive step will re-create the
.norecord file for meetings without recording marks, meaning that you
cannot use bbb-record --rebuild to force a recording without marks to be
processed.
Switch bbb-record to restart recording processing at the sanity stage to
match the BBB 2.2 behaviour. Rather than have it insert tasks directly
into resque via redis-cli, it goes through a ruby wrapper that performs
input validation and uses the resque apis.
In the case where a meeting had recording enabled (record=true on create
call) but the presenter did not start recording during the meeting,
recording processing needs to be stopped after the meeting data is
archived, but before the recording formats are processed.
In the current 2.3 code, processing is halted after the "sanity" step.
However, the 2.2 code stopped processing after the "archive" step
instead. The main difference is that the scripts in the "post_archive"
directory (which are actually post_sanity scripts) did not get run on
non-recorded meetings for 2.2. This behaviour should be preserved for
compatibility.
I have added a special exception to trigger halting processing for a
recording job without causing the entire resque job to be marked as
failed. It only causes the `schedule_next_step` method to be skipped, so
following jobs won't get automatically run. This fixes#11877
This function is useful any place you want the matched recording marks
with timestamps rebased so 0 is the start of the meeting, I've used it
for chat analysis, for example.
There is no functional change here, it only exposes the extra function
for recording scripts or dropin/post scripts to use.
Since Meteor was split in multiple process and events started to be
filtered by instances, all Etherpad's Redis events were being discarded.
Etherpad has a Redis' publisher plugin that is unaware of BigBlueButton's
existence. All the communication between them is kept simple with minimal
of internal data exchange. The concept of distincts subscribers at Meteor's
side broke part of this simplicity and, now, Etherpad has to know which
instance must receive it's messages. To provide such information I decided
to include Meteor's instance as part of the pad's id. Should look like:
- [instanceId]padId for the shared notes
- [instanceId]padId_cc_(locale) for the closed captions
With those changes the pad id generation made at the recording scripts had to
be re-done because there is no instance id available. Pad id is now recorded at
akka-apps and queried while archiving the shared notes.
