InputConnectionHealth.StreamHealth
Per-stream wallclock-vs-media health for this connection.
Field |
Type |
Repeated |
Description |
stream_key |
|||
frames_for_sample |
|||
bits_for_sample |
|||
realtime_delta_ms |
now - media time of the last frame (+ = behind realtime) |
||
realtime_delta_slope_ms_per_s |
trend of realtime_delta over the window (ms per s; + = falling further behind) |
||
media_advance_ms |
how much media time advanced over the sample window |
||
last_media_age_ms |
now - arrival (wallclock) of the last frame — the stall detector |
||
nominal_frame_rate |
|||
delivery_rate |
Smoothed media-time-per-wallclock-second (EWMA). ~1 = realtime; > 1 sustained = source running above realtime; the startup burst is smoothed out and only meaningful once settled. |
||
settled |
False during the startup window(s); once true the initial producer burst is discounted and delivery_rate reflects steady state. |
||
seamless |
ST 2022-7 per-path health, set only for a stream configured with a
redundant path (input.rtp, where protection is per stream). The udpTs
input has one stream pair per input and reports it on |
||
timestamp_resets_for_sample |
Frames of this stream over the sample window that the timestamp corrector re-anchored on (the first after each of its discontinuities): how often the source’s timestamps failed to hold a timeline. 0 on a healthy source; a source whose timestamps are garbage (two senders on one UDP port) resets on nearly every frame. One re-anchor counts on every stream of the program; unstable_timestamps judges the busiest stream’s count, so it counts once. |