Skip to content

Input redundancy

A headend must stay on air when a source dies. For that a stream is given several inputs: the station carries the air on the highest-priority live one, keeps an eye on the rest and returns to the primary when it comes back.

The typical pair on a station is a satellite feed as the primary input and a partner's multicast as the backup; or two groups from two headends.

Set several inputs

On the Sources tab add a second input with + Add source.

  • The order of the list is the priority: the higher an input is, the more important it is; the order is changed with the Up and Down arrows. The Priority field in an input's settings overrides the list order — a lower number is more important.
  • Set the "Backup recheck interval (s)" in the Input timeouts block. While a backup is not on air, the station probes it periodically, and in the list it shows as Ready — checked but not active. Without rechecking, the health of a backup stays a matter of faith until the failure itself, which is exactly the moment when finding out is too late; 0 turns rechecking off.

The input states in the list are Working, Ready, Starting, Degraded, Dead and Unchecked.

Dead and degraded rules

Every input has a Source settings plaque; empty fields inherit the stream's common rules from the Input timeouts block. That way the thresholds are set once per stream and overridden one by one, while a collapsed plaque shows with chips what exactly is set.

The fields are grouped in three, by what the station does when a rule is broken.

  • When the source is dead — hard failures, switch immediately: Source timeout (s) (no frames at all), Video timeout (no video), Max bitrate (bitrate above the ceiling — protection against garbage on the input).
  • When the source is degraded — soft failures, switch only if there is a better source: Audio timeout (audio gone while video lives), Min bitrate (kbit) (bitrate below the floor), CC error limit (MPEG-TS continuity violations within the clean window; the field exists only for transport protocols — UDP, HTTP MPEGTS, SRT, RIST).
  • Return and retries — Clean window (s), Max retry timeout and Priority.

Degradation is about the quality of the signal, and leaving a degraded input for a dead one makes no sense. So a soft rule does not drop the air, it only lowers the input in the queue. The station names the reason for a mark directly: "more CC errors than the limit", "bitrate below the floor", "no audio", "no frames".

The clean window

The clean window is the length of a stretch with no violations at all. An input counts as working if and only if it has broken no rule for that many seconds, and the same window is required of a revived input before it goes back on air.

This is the protection against a pendulum: a source that comes up for ten seconds and falls again no longer disturbs the air. The flip side is honest and predictable — the return to the primary is always delayed by the length of the window.

What happens at the moment of switching

  • An input becomes active only when it is ready. It must produce the stream parameters and the first keyframe; until then the current one keeps going on air. That is exactly why the switch does not give a black screen.
  • The join is not seamless. The new input's keyframe is the splice point, and the receiver will see it: a frame may repeat or go missing. Do not expect ST 2022-7 grade hitless switching from this mechanism.
  • A parameter mismatch is visible in advance. The station compares the inputs' parameters and shows the mismatch before the failure — a backup with a different resolution or a different track set is noticed early rather than at the moment of switching.
  • A failed input reconnects with a growing pause, and once all inputs are exhausted the rotation starts from the first one again.

The slate

The Slate on loss of sources block gives the stream a last line of defence: the Slate file on the streamer field takes the path to a prepared file, and Timeout (s) decides how many seconds after the last input has fallen it comes on.

For a headend this is not cosmetics: an empty multicast on the output makes receivers and modulators behave unpredictably, while a signal with a slate is still a signal.

The file must lie at that path on every machine the stream can move to: the slate is played by the machine the stream runs on at that moment.

What to look at afterwards

Per stream you can see how many seconds it lived on the primary input and how many on the backups, how long it had no data at all, how many switches and retries there were. A growing share of time on the backup means the primary input is systematically bad, even if the air was never interrupted.

Errors broken down by cause are on the Input errors page.

If the backup misbehaves

  • The air jumps between inputs. Increase the clean window: an input will have to prove its health for longer.
  • The backup is in the "Unchecked" state. The recheck interval is not set — the station does not go to it.
  • Switching feels slow. Lower the source timeout. Too small a threshold gives false positives on any network hiccup.
  • It never returns to the primary. The primary either has not come back yet, or cannot complete a clean window — that is, it is genuinely unhealthy.