Skip to content

Underfed 0.4.0

Choose a tag to compare

@PilaScat PilaScat released this 15 Sep 11:38
· 2 commits to master since this release
  • A source whose audio and video timestamps are far apart is switched too. reservoarr's ffmpeg
    then corrects every packet and swallows the gaps in the video, so the sound falls further
    behind with each one: on 14 September it was 7.8 s out after 20 minutes, and a viewer
    reopened the channel eight times. The sign is the timestamp discontinuity lines in
    delaybuf.log. At Timestamp discontinuities or above, per minute and 100 by default, for
    Confirm for, the channel moves to its next source with the guards and the hourly limit of a
    starving one. 0 turns it off.
  • From 5 to 15 September healthy feeds never logged more than 7 of those lines a minute. Replay
    over those days finds 60 triggers, all on the feed of 14 September, the first 45 s after its
    first packet; a 6-second burst on 9 September stays under the confirmation. The shortfall
    triggers are the same 103 with the rule and without it.
  • On a test bench, a source with its audio timestamps 300 s from its video made reservoarr log
    770 of those lines a minute, and the watcher moved the channel to its healthy source 47 s after
    the first one, without passing through the slate. The healthy source and a control channel
    logged none and were left alone.
  • Replay counts the two causes apart, Check status names the cause of each switch, and the
    journal carries it as cause.

Commits

  • b3e527e docs: why the confirmation stays at 45 s, and the monotonic "never"
  • 8110174 Switch a source whose audio and video timestamps are far apart
  • 57dbb94 Release 0.4.0
  • 2ffdc19 CHANGELOG 0.4.0: the bench run