- The HLS output's segments are now kept in memory inside the container (
/dev/shm/streamhls) instead of in/data/streamhls. Liquidsoap writes a segment every two seconds from the thread that keeps the station's audio on time, so a data volume that paused, as an Unraid user share does when the array is busy, paused the stream too: the log filled with "we must catchup" and, now and then, "Too much latency! Resetting active sources". The segments take a few megabytes of the 64 MB Docker gives every container there, so no--shm-sizeis needed. The small file that keeps segment numbering between restarts moves to/data/streamstate/hls.json. On the first start after upgrading, the station carries that numbering across and removes/data/streamhls, leaving the folder only if it holds something the station did not write. A container restart now empties the segments: anyone listening over HLS at that moment reconnects, as they already had to while the station restarted, and for the first few seconds after a restart a newly arriving player may be told a segment is missing and ask again. Nothing changes for the Icecast mounts or for anyone not using HLS.