- Fixed bug in stream health function causing loop if track not active
- Fixed DTSC pulls ignoring data before the live point
- Improved async buffers (deque mode) to spread the tracks more fairly
- DTSC pull now implements "ping" and "error" commands
- DTSC pulls report suspicious keyframe intervals to the origin and ask for confirmation
- DTSC output now accepts these reports and disconnects if there is no match in keyframe intervals
- Outputs in async mode now keep the seek point in all tracks when reselecting
- Outputs in async mode now default to a starting position in each track that is at a keyframe roughly halfway in the buffer
- Outputs in async mode now ignore playback rate (always fastest possible)
- Removed code duplication in prepareNext function
- Reordered the prepareNext function somewhat to be easier to follow for humans
- DTSC output no longer overrides initialSeek function, now uses default implementation
- Sanitycheck output now supports both sync and async modes, supports printing multiple timestamps for multiple tracks
- Stream selecting now obeys new "maxdelay" capa entry
- Output::liveSeek now takes an optional bool argument to indicate only rate control should be applied (no seeking)
- dataWaitTimeout is now configurable per-output, defaults to the old 25s
- WS/MP4 uses the new liveSeek with rate-control only
- WS/MP4 uses the new dataWaitTimeout and sets it to 4.5s
- WS/MP4 uses the new maxdelay capa, sets it to 5s
- WS/MP4 will now auto-reselect tracks if a track is dropped for data wait timeout or disappeared from metadata reasons
- Added support for jitter information in WS/MP4 protocol
- Robustify sendWebsocketCodecData being ran when sendHeader is ran
- Fix race condition when switching video tracks before previous video track has sent a single frame
Removed curPage map from IO. bufferFrame now creates this variable locally and passes it to bufferStart, bufferFinalize and bufferNext
Fix keyNum selection with mixed live & VoD data
Fix bufferframe to handle mixed VoD and live
Added check to bufferFrame to not start the countdown timer for removing live pages
Fixed countdown timer being set using keyNum rather than pageNumber, which resulted in the wrong pages being deleted
livePage variable moved from static to private variable to correctly handle multithreaded inputs
# Conflicts:
# src/io.cpp
# src/output/output.cpp
- Optimize URIReader class to not close connections if not needed
- reConnector now works for non-GET requests with GET params
- Chunk sending mode correctly kept for HLS
- Removed lots of H.Clean() from HLS that weren't needed
- Improved HTTP output class request handling logic
- Removed firstRun from HTTP output class; no longer needed
All processes:
- Added process status system and relevant API calls
- Added ability to set track masks for input/output in processes
- Added support for unmasking tracks when there is a push target, by the "unmask" parameter.
- Added track unmasking support for processes on exit/error
- Make processes start faster, if possible, in the first few seconds of a stream
- Delay stream ready state if there are processes attempting to start
Livepeer process updates:
- Added Content-Resolution header to MistProcLivepeer as per Livepeer's request
- Renamed transcode from "Mist Transcode" to source stream name
- Added ability to send audio to livepeer
- Robustified livepeer timing code, shutdown code, and improved GUI
- Prevent "audio keyframes" from starting segments in MistProcLivepeer
- Multithreaded (2 upload threads) livepeer process
- Stricter downloader/uploader timeout behaviour
- Robustness improvements
- Fix small segment size 😒
- Streamname correction
- Prevent getting stuck when transcoding multiple qualities and they are not equal length
- Corrected log message print error
- Race condition fix
- Now always waits for at least 1 video track
Approx. 10% of code originally written by Roxlu, but keeping it split up during cleanup before merge proved practically impossible, so it's all merged into a single commit.
