Using grafana with VLC

Observability tooling for long-running analysis

Alexandre Janniaux - Juin 2025

Type of debugging

Or printf, and build understanding of what might be the root cause.

1

Reproduce and use a debugger

Build understanding in code via a scenario while testing an hypothesis.

2

Reproduce in a regression test

Try to find any root cause because there is no ideas of why it fails yet.

3

Write a CSV serializer in logs

Trace API in VLC [1]

[1]: https://code.videolan.org/videolan/vlc/-/merge_requests/365

vlc_tracer_Trace(tracer,
                 VLC_TRACE("type", "DEMUX"),
                 VLC_TRACE("id", "input"),
                 VLC_TRACE_TICK_NS("buffering", -i_late),
                 VLC_TRACE_END);

  • From Nicolas Le Quec, 5 juillet 2021

  • Tracer modules can be implemented

  • Core API to forward tracer to VLC Objects

  • JSON tracer implementation

vlc-pa HTML

house-made timeseries visualizer reading JSON traces

vlc-pa

Python mathplotlib to visualize json traces

  • More systematic than copy-pasting logs in libreoffice
  • Common tools and presentation
  • Pandas integration, more advanced flaw research

Benefits

Pain points

  • Off-line captures of the points
  • Huge amount of data to proceed
  • Tool maintenance and missing features

Re-implement the traces

CFO

Realtime traces

Dashboard completeness

Tools / alerts integration

Telegraf

Time series aggregator

Influxdb

time series database, handling tags and metrics

Loki

Log aggregator and metadata indexer

Grafana

sysadmin tool to monitor server nodes

Docker compose

code.v.o/Garf/vlc-influxdb

docker compose up

firefox http://localhost:3000/

export VLC_TELEGRAF_ENDPOINT="tcp://localhost:8094"

vlc --tracer=telegraf

Clock interpolation

Clock interpolate with

    y = coeff * t + offset

 

This tracks the two parameters over time.

Clock context

Analyze asynchronous issues in the clock design.

 

Each line is a track or input.

Each number matches with a given clock context.

Log integration

Logging is synchronized with the selected time window.

 

Provide additional event feedback on what happened.

Input buffer

Input buffering metrics shows internal state of the input clock.

 

Highlights fluttering in the input, and long-run input drift issue.

Audio drift

Reports the "drift" value for each clock/track whenever they get updated.

 

Tracks avsync issue, long-run drift, resampling, etc... 

Audio latency

Tracks the reporting of audio latency, specifically for coreaudio here.

 

Useful in testing variation of output devices and its consequences.

Audio buffer

Tracks how much buffer if available in the audio output, after audio_Play().

 

When underrun happens, shows how much was missing on the right.

Audio buffer


from(bucket: v.defaultBucket)
  |> range(start: v.timeRangeStart, stop: v.timeRangeStop)
  |> filter(fn: (r) => r["_measurement"] == "measurement" and r.id == "coreaudio")
  |> yield(name: "mean")

VSYNC discrepancy

Analyze irregularities in the reporting of VSYNC (SK branch).

 

Gives hints about frame drops and is compatible with alerts.

Video drops

Shows a bar when a frame is drop and set the colour/values to how much it was late.

Examples of bugs

When do you need this ?

Your mileage may vary.

– Famous quote

Fast channel zapping (SK)

Symptoms:

- PCR gap, ES_OUT_RESET_PCR()

- Re-buffering at beginning

- Triggers clock context bug /o\

Fast channel zapping (SK)

Symptoms:

 - Playback way to late/early

 - Playback, then resampling,

    then silence

Fast channel zapping (SK)

Symptoms:

 - Lipsync issue, audio and video out of sync multiple secs

 - Clock confidently report drift

    value is 0ms drift /o\ 

Decoder out of sync (PDI)

Symptoms:

 - Video playback getting slower

 - Audio playback gets out of sync quickly

DEMO TIME

Maybe

Made with Slides.com