srt2hls - how to generate index.m3u8 from multiple inputs without re-encoding?

[LifelessMuffin]

Hi guys, I would like to transmux the 3 inputs (with different bitrates) into 3 individual playlists (high.m3u8; mid.m3u8; low.m3u8) which are then indexed into the master playlist index.m3u8. When re-encoding a single input as shown in the examples it works fine, the index,m3u8 file is generated along the other playlists (ABR). But how to achieve the same without re-encoding and with 3 inputs?
This is my script:

settings.root.max_latency.set(5.)
settings.clock.allow_streaming_errors.set(false)

input_a = input.srt(port=9000, mode="listener")
input_b = input.srt(port=9001, mode="listener")
input_c = input.srt(port=9002, mode="listener")

transmux_high = %ffmpeg(format="mpegts", %audio.copy, %video.copy)
transmux_medium = %ffmpeg(format="mpegts", %audio.copy, %video.copy)
transmux_low = %ffmpeg(format="mpegts", %audio.copy, %video.copy)

high_stream = [("transmux_high",transmux_high)]
medium_stream = [("transmux_medium",transmux_medium)]
low_stream = [("transmux_low",transmux_low)]

def segment_name(~position,~extname,stream_name) =
  timestamp = int_of_float(time())
  duration = 2
  "#{stream_name}_#{duration}_#{timestamp}_#{position}.#{extname}"
end

output.file.hls(
                playlist="high.m3u8",
                fallible=true,
                segment_duration=2.0,
                segments=5,
                segments_overhead=5,
                segment_name=segment_name,
                "/var/www/liquidsoap/hls",
                high_stream,
                input_a
               )

output.file.hls(
                playlist="medium.m3u8",
                fallible=true,
                segment_duration=2.0,
                segments=5,
                segments_overhead=5,
                segment_name=segment_name,
                "/var/www/liquidsoap/hls",
                medium_stream,
                input_b
               )

output.file.hls(
                playlist="low.m3u8",
                fallible=true,
                segment_duration=2.0,
                segments=5,
                segments_overhead=5,
                segment_name=segment_name,
                "/var/www/liquidsoap/hls",
                low_stream,
                input_c
                )

[toots]

Well that was a great question and I found not one but two ways of doing this! They should be both working with the latest main. I wanted to backport to rolling-release-v2.2.x but this seems like too much changes.

Method 1: muxing

This one works by muxing all the sources together into a single multitrack source and the selectively encoding subsets in each HLS streams. The plus side is that you don't have anything special to do on the HLS side. However, when using SRT, you have to tell liquidsoap to only account for the synchronization coming from one of the multiple SRT sources. If they come from different machines, this can be an issue long term..

Here's some code:

input_a = input.srt(port=9000, mode="listener")
input_b = input.srt(clock_safe=false, port=9001, mode="listener")
input_c = input.srt(clock_safe=false, port=9002, mode="listener")

let {video = video_a, audio = audio_a} = source.tracks(input_a)
let {video = video_b, audio = audio_b} = source.tracks(input_b)
let {video = video_c, audio = audio_c} = source.tracks(input_c)

muxed_input =
  source(
    {
      audio_a=audio_a,
      audio_b=audio_b,
      audio_c=audio_c,
      video_a=video_a,
      video_b=video_b,
      video_c=video_c
    }
  )

streams =
  [
    (
      "transmux_high",
      %ffmpeg(
        format = "mpegts",
        %audio_a.copy,
        %audio_b.drop,
        %audio_c.drop,
        %video_a.copy,
        %video_b.drop,
        %video_c.drop
      )
    ),
    (
      "transmux_medium",
      %ffmpeg(
        format = "mpegts",
        %audio_a.drop,
        %audio_b.copy,
        %audio_c.drop,
        %video_a.drop,
        %video_b.copy,
        %video_c.drop
      )
    ),
    (
      "transmux_low",
      %ffmpeg(
        format = "mpegts",
        %audio_a.drop,
        %audio_b.drop,
        %audio_c.copy,
        %video_a.drop,
        %video_b.drop,
        %video_c.copy
      )
    )
  ]

def segment_name(~position, ~extname, stream_name) =
  timestamp = int_of_float(time())
  duration = 2
  "#{stream_name}_#{duration}_#{timestamp}_#{position}.#{extname}"
end

output.file.hls(
  fallible=true,
  segment_duration=2.0,
  segments=5,
  segments_overhead=5,
  segment_name=segment_name,
  "/tmp/hls",
  streams,
  muxed_input
)

Method 2: manual main playlist

With this method, you define 3 HLS outputs but none of them write the main playlist and you do it yourself instead. This is a more straightforward method but a little more advanced.

Here's some code:

input_a = input.srt(port=9000, mode="listener")
input_b = input.srt(clock_safe=false, port=9001, mode="listener")
input_c = input.srt(clock_safe=false, port=9002, mode="listener")

encoder = %ffmpeg(format = "mpegts", %audio.copy, %video.copy)

# We collect the stream information from each individual output
# and write the main playlist when we have them all:

streams = ref([])
def main_playlist_writer(~extra_tags=_, ~version=_, ~prefix=_, s) =
  streams := [...streams(), ...s]

  if
    list.length(streams()) == 3
  then
    file.write(data=hls.playlist.main(streams()), "/tmp/hls/streams.m3u8")
  end

  # Return null to prevent any of the individual output
  # from writting the main playlist
  null()
end

def segment_name(~position, ~extname, stream_name) =
  timestamp = int_of_float(time())
  duration = 2
  "#{stream_name}_#{duration}_#{timestamp}_#{position}.#{extname}"
end

output.file.hls(
  fallible=true,
  segment_duration=2.0,
  segments=5,
  segments_overhead=5,
  segment_name=segment_name,
  main_playlist_writer=main_playlist_writer,
  "/tmp/hls",
  [("transmux_high", encoder)],
  input_a
)

output.file.hls(
  fallible=true,
  segment_duration=2.0,
  segments=5,
  segments_overhead=5,
  segment_name=segment_name,
  main_playlist_writer=main_playlist_writer,
  "/tmp/hls",
  [("transmux_medium", encoder)],
  input_b
)

output.file.hls(
  fallible=true,
  segment_duration=2.0,
  segments=5,
  segments_overhead=5,
  segment_name=segment_name,
  main_playlist_writer=main_playlist_writer,
  "/tmp/hls",
  [("transmux_low", encoder)],
  input_c
)

[LifelessMuffin]

Thank you!
I am going to try it when I update my Ubuntu machine :D , there is no .deb installer yet, right?
Up to now I was using something similar to the second method by making a manual index, indeed using the single encoder option as you did makes it way better.
The only "breaking point" at the moment is the video codec issue (as in the other current open ticket).

[LifelessMuffin]

So I installed the latest .deb file

INFO: Loading Sdl_image, Target = linux
INFO: Loading Sdl_ttf, Target = linux
Liquidsoap 2.2.2+git@5c0b67524
Copyright (c) 2003-2023 Savonet team
Liquidsoap is open-source software, released under GNU General Public License.
See <http://liquidsoap.info> for more information.

But still getting an error, I made a copy and paste of the method 1 just to test but this is the output:

INFO: Loading Sdl_image, Target = linux
INFO: Loading Sdl_ttf, Target = linux
At /etc/liquidsoap/test3.liq, line 77, char 0-0:

Error 9: Failure: invalid content field
Raised at Stdlib.failwith in file "stdlib.ml", line 29, characters 17-33
Called from Lang_ffmpeg.parse_encoder_name in file "src/core/encoder/lang/lang_ffmpeg.ml", line 160, characters 13-45
Called from Lang_ffmpeg.type_of_encoder.(fun) in file "src/core/encoder/lang/lang_ffmpeg.ml", line 202, characters 30-53
Called from Stdlib__List.fold_left in file "list.ml", line 121, characters 24-34
Called from Lang_encoder.type_of_encoder in file "src/core/encoder/lang_encoder.ml", line 131, characters 15-32
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 194, characters 16-45
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Stdlib__List.iter in file "list.ml", line 110, characters 12-15
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 208, characters 10-52
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Stdlib__List.iter in file "list.ml", line 110, characters 12-15
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 203, characters 10-52
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 345, characters 10-43
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 389, characters 10-48
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 389, characters 10-48
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 389, characters 10-48
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 389, characters 10-48
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 389, characters 10-48
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 389, characters 10-48
Called from Liquidsoap_lang__Typechecking.check.(fun).check in file "src/lang/typechecking.ml", line 141, characters 4-39
Called from Liquidsoap_lang__Typechecking.check.(fun) in file "src/lang/typechecking.ml", line 389, characters 10-48
Called from Liquidsoap_lang__Typechecking.check in file "src/lang/typechecking.ml", line 419, characters 4-48
Re-raised at Liquidsoap_lang__Typechecking.check in file "src/lang/typechecking.ml", line 428, characters 4-38
Called from Liquidsoap_lang__Startup.time in file "src/lang/startup.ml", line 30, characters 12-16
Called from Liquidsoap_lang__Runtime.type_and_run.(fun) in file "src/lang/runtime.ml", line 33, characters 9-105
Called from Stdlib__Fun.protect in file "fun.ml", line 33, characters 8-15
Re-raised at Stdlib__Fun.protect in file "fun.ml", line 38, characters 6-52
Called from Liquidsoap_lang__Runtime.type_and_run in file "src/lang/runtime.ml", line 30, characters 4-612
Called from Liquidsoap_lang__Runtime.report in file "src/lang/runtime.ml" (inlined), line 215, characters 12-23
Called from Liquidsoap_lang__Runtime.from_lexbuf in file "src/lang/runtime.ml", line 229, characters 2-159

[toots]

Could you paste your script?

[LifelessMuffin]

Yes sure! It is an exact copy of your code of Method 1 as above, I tried it some days ago as a first run after updating Liquidsoap to version 2.2.2 (on Ubuntu 22.04 using the .deb installer):

input_a = input.srt(port=9000, mode="listener")
input_b = input.srt(clock_safe=false, port=9001, mode="listener")
input_c = input.srt(clock_safe=false, port=9002, mode="listener")

let {video = video_a, audio = audio_a} = source.tracks(input_a)
let {video = video_b, audio = audio_b} = source.tracks(input_b)
let {video = video_c, audio = audio_c} = source.tracks(input_c)

muxed_input =
  source(
    {
      audio_a=audio_a,
      audio_b=audio_b,
      audio_c=audio_c,
      video_a=video_a,
      video_b=video_b,
      video_c=video_c
    }
  )

streams =
  [
    (
      "transmux_high",
      %ffmpeg(
        format = "mpegts",
        %audio_a.copy,
        %audio_b.drop,
        %audio_c.drop,
        %video_a.copy,
        %video_b.drop,
        %video_c.drop
      )
    ),
    (
      "transmux_medium",
      %ffmpeg(
        format = "mpegts",
        %audio_a.drop,
        %audio_b.copy,
        %audio_c.drop,
        %video_a.drop,
        %video_b.copy,
        %video_c.drop
      )
    ),
    (
      "transmux_low",
      %ffmpeg(
        format = "mpegts",
        %audio_a.drop,
        %audio_b.drop,
        %audio_c.copy,
        %video_a.drop,
        %video_b.drop,
        %video_c.copy
      )
    )
  ]

def segment_name(~position, ~extname, stream_name) =
  timestamp = int_of_float(time())
  duration = 2
  "#{stream_name}_#{duration}_#{timestamp}_#{position}.#{extname}"
end

output.file.hls(
  fallible=true,
  segment_duration=2.0,
  segments=5,
  segments_overhead=5,
  segment_name=segment_name,
  "/tmp/hls",
  streams,
  muxed_input
)

[toots]

You need the latest main for this feature I'm afraid. You can grab the build assets from any of the recent actions run for instance: https://github.com/savonet/liquidsoap/actions/runs/6659013067

[LifelessMuffin]

I was also wondering if maybe there is something with the srt lib, suddenly something happens and this is the output.
This is version 2.2.1 on Ubuntu 22.04:

Error while executing OCaml read callback: Srt.Error(-1013981956, "Non-blocking call failure: transmission timed out")
2023/10/25 15:40:25 [input.srt.2:2] Feeding failed: Avutil.Error(Generic error in an external library)
2023/10/25 15:40:25 [/var/www/liquidsoap/hls/medium_m3u8:3] Source failed (no more tracks) stopping output...
2023/10/25 15:40:25 [/var/www/liquidsoap/hls/medium_m3u8:3] Could not remove file /var/www/liquidsoap/hls/transmux_medium.m3u8: No such file or directory
2023/10/25 15:40:25 [/var/www/liquidsoap/hls/medium_m3u8:3] Could not remove file /var/www/liquidsoap/hls/medium.m3u8: No such file or directory
Error while executing OCaml read callback: Srt.Error(-1013981956, "Non-blocking call failure: transmission timed out")
2023/10/25 15:40:25 [input.srt:2] Feeding failed: Avutil.Error(Generic error in an external library)
Error while executing OCaml read callback: Srt.Error(-1013981956, "Non-blocking call failure: transmission timed out")
2023/10/25 15:40:26 [input.srt.3:2] Feeding failed: Avutil.Error(Generic error in an external library)
2023/10/25 15:40:26 [/var/www/liquidsoap/hls/low_m3u8:3] Source failed (no more tracks) stopping output...
2023/10/25 15:40:26 [/var/www/liquidsoap/hls/low_m3u8:3] Could not remove file /var/www/liquidsoap/hls/transmux_low.m3u8: No such file or directory
2023/10/25 15:40:26 [/var/www/liquidsoap/hls/low_m3u8:3] Could not remove file /var/www/liquidsoap/hls/low.m3u8: No such file or directory
2023/10/25 15:40:26 [clock.main:2] We must catchup 1.96 seconds!

and also (from syslog):

Oct 26 13:39:48 origin-test liquidsoap[12205]: 2023/10/26 13:39:48 [input.srt.2:2] Feeding failed: File "src/core/stream/content_base.ml", line 285, characters 4-10: Assertion failed
Oct 26 13:39:48 origin-test liquidsoap[12205]: 2023/10/26 13:39:48 [clock.main:2] We must catchup 1.89 seconds!
Oct 26 13:39:48 origin-test liquidsoap[12205]: 2023/10/26 13:39:48 [input.srt:2] Feeding failed: File "src/core/stream/content_base.ml", line 285, characters 4-10: Assertion failed

it started when I added clock_safe=false on the input 2 and 3. It gets this behavior after some hours
can it be something related to the clock?

Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [input.srt:2] Feeding failed: File "src/core/stream/content_base.ml", line 285, characters 4-10: Assertion failed
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Source /var/www/liquidsoap/hls/high_m3u8 failed while streaming: Sys_error("/var/www/liquidsoap/tmp/liq2d2419tmp: Too many open files")!
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Raised by primitive operation at Stdlib.open_out_gen in file "stdlib.ml", line 331, characters 29-55
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Called from Stdlib__Filename.open_temp_file.try_name in file "filename.ml", line 357, characters 7-73
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Re-raised at Stdlib__Filename.open_temp_file.try_name in file "filename.ml", line 359, characters 30-37
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Called from Hls_output.hls_output#open_out in file "src/core/outputs/hls_output.ml", line 548, characters 8-66
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Called from Hls_output.hls_output#open_segment in file "src/core/outputs/hls_output.ml", line 622, characters 24-46
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Called from Stdlib__List.iter in file "list.ml", line 110, characters 12-15
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Called from Hls_output.hls_output#start in file "src/core/outputs/hls_output.ml", line 788, characters 6-41
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Called from Start_stop.base#transition_to in file "src/core/tools/start_stop.ml", line 53, characters 12-22
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Called from Output.output#output in file "src/core/outputs/output.ml", line 164, characters 8-41
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2] Called from Clock.MkClock.clock#end_tick.(fun) in file "src/core/clock.ml", line 318, characters 16-24
Oct 26 03:39:11 origin-test liquidsoap[6157]: 2023/10/26 03:39:11 [clock.input.srt:2]
Oct 26 03:39:12 origin-test liquidsoap[6157]: 2023/10/26 03:39:12 [input.srt:2] Feeding failed: File "src/core/stream/content_base.ml", line 285, characters 4-10: Assertion failed
Oct 26 03:39:12 origin-test liquidsoap[6157]: 2023/10/26 03:39:12 [clock.input.srt:2] We must catchup 1.24 seconds!
Oct 26 13:27:04 origin-test liquidsoap[12205]: 2023/10/26 13:27:04 [input.srt:2] Feeding failed: File "src/core/stream/content_base.ml", line 285, characters 4-10: Assertion failed
Oct 26 13:27:05 origin-test liquidsoap[12205]: 2023/10/26 13:27:05 [input.srt:2] Feeding failed: File "src/core/stream/content_base.ml", line 285, characters 4-10: Assertion failed

[toots]

I think that we're seeing two things here:

1. A failure in the content, looking like a timeout at the moment.
2. As a consequence of the multiple restart, a file descriptor leak that causes too many files from being opened.

2 is a consequence of 1. I have a pushed a potential fix for it.

As for 1, I think the best approach would be to run the test script with a single SRT input to try to de-corelate wether this is problem on a single stream or on the 3 streams when combined.