Liquidsoap stalling when buffer.adaptative gets emptied both in liquidsoap stable and rolling release

[hydrosIII]

We have a server playing a stream from the internet. We have some clock issues and latency issues so we have implemented buffer.adaptative as a solution however, sometimes the buffer gets emptied. That is normal, the only problem is that is not starting again, but it is stalling the entire program. So we have a script to restart liquidsoap every x hours. We are always trying to restart liquidsoap before the buffer gets full, so it does not gets emptied.

The number of hours when the buffers get emptied varies, but is after 3 hours of running. We are restarting every 4 hours liquidsoap.

The desired behaviour is that after the buffer gets emptied it restarts and the program continue to run, so we can at least restart the program every 12 hours or so. Or do not restart it at all.

Is there any mistake in the usage of the buffer? Maybe this is a config option? Is there a way to ensure that it will restart after it gets emptied?

The relevant lines of the log are:

mar 22 12:48:00 violetaradio liquidsoap[1458595]: 2023/03/22 12:48:00 [buffer.adaptative_producer_0:3] Buffer emptied, start buffering.>
mar 22 12:48:00 violetaradio liquidsoap[1458595]: 2023/03/22 12:48:00 [mksafe:3] Switch to safe_blank with transition.

We do not have more logs even if we set to the max debug.

Liquidsoap version: Liquidsoap 2.1.3. Installed trough .deb package. Debian buster.

The relevant script:

settings.frame.audio.samplerate.set(48000)

#### encodings de las canciones
settings.tag.encodings.set(["UTF-8", "ISO-8859-1"])

##### max latency
settings.root.max_latency.set(150.)

#### logs a stdout para systemd y a un archivo de log para backup
log.stdout.set(true)
settings.log.file.path.set("/home/violetaradio/violeta.log")

### debug mode 
settings.srt.log.level.set(5)
settings.srt.log.verbosity.set("debug")
settings.server.log.level.set(3)

### ffmpeg
settings.decoder.decoders.set(["FFMPEG"])
settings.ffmpeg.log.verbosity.set("debug")

settings.init.allow_root.set(true)

#### input 
radio = input.ffmpeg(max_buffer=60.0,self_sync=true,"https://b.stream.mayfirst.org/violetaradio.mp3")

###### buffer.adapatative
radio = buffer.adaptative(radio,max=240.,buffer=5.0,fallible=true,averaging=60., limit=1.1, reset=true)

## posprocessing
radio = compress.old(radio, attack = 8.0, gain = -2.0, knee = 1.0, ratio = 4.0, release = 600.0, threshold = -10.0)
radio = normalize(radio, target = -13.0, threshold = -25.0)

#### output directamente en alsa, pongo fallible true como fallback. 
output.alsa(id="alsa",fallible=true,mksafe(radio))

The relevant line is this:

###### buffer.adzptative
radio = buffer.adaptative(radio,max=240.,buffer=5.0,fallible=true,averaging=60., limit=1.1, reset=true)

[toots]

Hi and thanks for reporting. We are currently looking at this operator so hopefully we should be able to get back to you soon.

[hydrosIII]

Hi, We are facing this issue with the normal buffer operator also.

[hydrosIII]

Same issue with version 2.1.4 Liquidsoap 2.1.4

[hydrosIII]

I can confirm the same issue with rolling release liquidsoap 2.2.0+git@f3df74f81

I get:

 14:06:50 [adaptativo:3] Buffer emptied, start buffering...
 14:06:50 [mksafe.2:3] Switch to safe_blank.2 with transition.
 14:06:50 [source:4] Source replay_metadata gets down.

[hydrosIII]

Issue solved with new change introduced in liquidsoap version 2.2.4-1 about managing long running streams. @toots Thank you very much for your hard work. I made a small donation to the project.

[toots]

Thank you!