Harbor.input && http.input : Buffer overrun looping after a some days

[yah00078]

Describe the bug

Hello !

I start liquidsoap, it run well during a while listening icecast stream and one day (can be from 24h to 7 days) all stop (sound, switch, fallback, encoders).

When I see logs I can find about 3500 line of '[http_9625:3] Buffer overrun: Dropping 0.03s.' (same for harbor.input) every 0.03s a new line is generated.

The only way to turn back available, is to restart liquidsoap.

Following many forums topics, I see there is a possible clock sync problem between encoder and liquidsoap.

I'am convinced there is an hardware un-sync but on many systems I never reached theses issues.

I can add info: more resources expensive is the code, more often it happens.

2020/08/04 09:39:51 [http_9625:3] New metadata chunk: ? -- -M- - Crois au Printemps.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.00s.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:57 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:58 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:58 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:58 [http_9625:3] Buffer overrun: Dropping 0.03s.
2020/08/04 09:39:58 [http_9625:3] Buffer overrun: Dropping 0.03s.
......

To Reproduce src = input.http("http://XXXXX") PGM = fallback([[strip_blank(max_blank=30.,master_src),ANOTHER_SOURCE)

cross fingers and wait ! :-)

Expected behavior

-> add a buffer adaptation on input

-> Is it possible to implement a resilience security action. as if there is more than XXsec of buffer overrun, drop all buffer and listen back?

Version details

• OS: DEBIAN 10
• Version: 1.3.3
• CPU : E5-1650 0 @ 3.20ghz

Install method packages

Thank you for help :-)

[yah00078]

Hello after some tests, I can add this : -> At 320 kbs/s it works about 5 days before drops. -> At 224 kbs/s it immediately buffer overrun

[yah00078]

I found a possibly clue to explain why there is no fallback in this case.

I tried direct input vs direct input + fallback with strip_blank function.

I think when harbor.input buffer overrun, stripblank function doesn't see it as sound blank, so fallback mechanism don't do anything.

[toots]

Hi! Sorry for the late response. I'll try to take a minute to test and give you a more informed response soon :-)

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[hydrosIII]

I am having similar problems with a stream that has a lot of latency. 150 ms from the server to input.http. And I have some crashes of liquidsoap, or the output just get silenced instead of dropping the buffer and reinitiating it. I have to reboot liquidsoap often. Sometimes it reinitiates the clocks, but sometimes just silently crashes.