Skip to content

WebDAV

WebDAV authentication and streaming/connection behavior for playback mounts.

Headless ENV

Map config keys below to NZBDAV_CONFIG__... with the naming algorithm (webdav.pass → NZBDAV_CONFIG__WEBDAV__PASS).

Control Config key Default Effect
WebDAV User webdav.user admin / WEBDAV_USER Alphanumeric + _ -
WebDAV Password webdav.pass env WEBDAV_PASSWORD Required for rclone/clients
Queue Download Connections usenet.max-queue-connections blank = all Cap queue NNTP use
Queue connection preset since 0.9.1 usenet.max-queue-connections-preset blank = all low/medium/high/max = 25/50/75/100% of the pooled total, so one value fits any provider count. Ignored when the absolute setting above is set
Concurrent Queue Downloads since 0.9.0 queue.worker-count 1 Concurrent NZB imports (1–4); the oldest item is preferred while other workers share the connection budget
Enable Segment Cache usenet.segment-cache.enabled off Disk cache; restart required
Cache path usenet.segment-cache.path /config/segment-cache
Maximum size (GB) usenet.segment-cache.max-gb 10
Max Download Connections usenet.max-download-connections 0 (auto = pool) Streaming budget
Apply limit per stream usenet.max-download-connections-per-stream off Per-stream budget
Per-stream performance usenet.max-download-connections-per-stream-preset high low/medium/high/max
Streaming Priority (vs Queue) since 0.9.0 usenet.streaming-priority 80 High vs Low admission odds at each saturated provider connection pool (playback High, queue/health Low). Spare capacity is never held idle for High
Streaming Segment Timeout usenet.streaming-segment-timeout-seconds 8 2–40s
Streaming Read Timeout since 0.9.0 usenet.streaming-read-timeout-seconds 30 5–120s initial backend wait to open a GET/range (semaphore + pool + first segment). Cleared once body bytes flow; mid-stream stalls use the per-segment timeout. Unstarted responses return 503 with Retry-After — rclone/FUSE may still show client-side errors
Streaming Segment Retries usenet.streaming-segment-retries 3 0–5
Article Buffer Size usenet.article-buffer-size 40 Articles buffered ahead per stream (count bound)
In-flight article budget (MiB) since 0.8.2 usenet.in-flight-article-budget-mb auto (≈25% of container memory, max 512) Host-wide cap on decoded article bytes in RAM (64–8192); empty = derived; distinct from per-stream article buffer count
Idle connection timeout usenet.idle-connection-timeout-seconds 60 15–300; pool rebuild/restart
Pipelined article downloads usenet.pipelined-body-requests on WebDAV BODY batches
Container-aware gap fill since 0.10.0 usenet.container-aware-fill off Experimental; MPEG-TS null packets instead of zeros for confirmed missing articles
Enforce Read-Only webdav.enforce-readonly on /content readonly
Show hidden files webdav.show-hidden-files off Dot-prefixed names in Explore
Preview par2 files webdav.preview-par2-files off Render as text
Sanitize paths for Windows webdav.windows-safe-paths on New mounts only

Speed tuning

Raise Max Download Connections until throughput plateaus without pegging CPU. Baseline with a host speed test, then time a /view download from inside the container against the backend.

Article Buffer Size and adaptive prefetch since 0.9.0

usenet.article-buffer-size bounds how many decoded articles a stream may keep ahead of the consumer — and therefore per-stream memory — not how many provider connections playback may use.

When Pipelined article downloads is on, WebDAV BODY requests start in batches of up to four articles on one connection. If playback starves waiting for the next segment, InfiniDysk automatically narrows that batch width (4 → 2 → 1) so more connections can work in parallel at the same buffer depth. When the consumer stays ahead, batch width recovers gradually. Connection and host-wide byte budgets still apply.

Experimental container-aware gap fill since 0.10.0

When a confirmed-missing or persistently corrupt article cannot be recovered from any provider or fallback Message-ID, InfiniDysk normally emits the same number of zero bytes to keep every later file offset correct. Enable Container-aware gap fill to emit format-native discard markers instead for supported direct files:

  • MPEG-TS (.ts, .m2ts, .mts): packet-aligned null packets when exact segment offsets are available.

This may help compatible players resynchronize sooner, but it cannot restore the missing audio or video data. Matroska, MP4/MOV, and archive-backed files retain zero-fill because arbitrary article boundaries do not provide safe container-element boundaries.

The setting is experimental and defaults to off. It does not affect transient transport failures: after their retries are exhausted, the current HTTP response fails so the player can request the range again.

Capturing a buffering support pack

Playback stalls are only diagnosable if the evidence is still in the log buffer when you collect the pack:

  1. Set LOG_LEVEL=INFO. At DEBUG, routine background activity can fill the whole buffer within hours and evict the streaming events. logs/warnings.log in the pack keeps the last 500 warnings and errors regardless, so check it first.
  2. Enable Developer stream tracing on Settings → Support for 15–60 minutes and pick a capacity that covers the whole reproduction (default 100,000 events; or set STREAM_TRACE_EVENTS=100000 and restart). After downloading the pack, check manifest.json → streamTraces.overflowed — if true, increase capacity and reproduce again.
  3. Reproduce the stall — play the file from Explore so the read goes straight to /view, skipping rclone and your media server.
  4. Download the pack from Settings → Support immediately afterwards, while tracing is still on. The buffer is in-memory and is cleared on restart.

See Logs and crash dumps.

Streaming · NNTP pipelining