Stream failover: staying on air when your connection drops

A dropped stream is not just a gap in the video. The platform ends the broadcast, the VOD splits in two, your raid target is gone, and the people who were watching get an offline screen and go somewhere else. Failover is the machinery that stops a bad thirty seconds from becoming that. This page explains why streams drop, what platforms do about it (very little), and the two mechanisms that actually work.

Why your stream keeps dropping

Almost always it is the uplink, not OBS. In rough order of how often we see it:

• Your upload collapses, briefly. Cellular upload is not a flat number. It sags when the cell gets busy, when you walk indoors, when you turn a corner. If your encoder is set to send more than the link can carry at its worst, the buffer fills and the connection dies.
• The handover between cells. Moving between towers, or between 5G and LTE, produces a gap of a few seconds. It is invisible when you browse and fatal to an RTMP session with no buffer.
• RTMP has no room to recover. RTMP rides on TCP with no latency budget of its own: a lost stretch of packets stalls the session and the server hangs up. SRT holds a retransmission buffer — typically 2 to 4 seconds — precisely so a short gap is invisible.
• One connection, one point of failure. A single SIM card with no fallback means every dead spot on your route is a dead stream. This is the problem bonding solves, and nothing else does.
• The encoder is over-configured. 1080p60 at 8,000 kbps looks excellent right up to the moment it does not fit. Most IRL drops we look at come from an encoder set for the best case rather than the worst one.

What the platform does about it

The two mechanisms that work

Different products, and they solve the problem in different places. It is worth knowing which one you are buying:

• A relay that holds the connection and splices. Your video goes to a server that stays connected to the platform. When your feed stops arriving, the server keeps talking to the platform and sends a placeholder — a screen, a loop — until you come back. The platform never sees an interruption, so the broadcast is never ended.
• Automatic scene switching inside OBS. Something watches the incoming bitrate and, when it collapses, tells OBS to cut to your offline scene, then back when the signal recovers. Because the fallback is a scene, it can be anything OBS can build: a countdown, chat on screen, the last frame, music. The stream itself never stops.

They are complementary, not rival. Scene switching gives you a better-looking fallback; the relay splice covers the case where the whole encoder is gone, not just the video.

How this works on Enhanced IRL

• With a Cloud OBS plan, it is included. Every Cloud OBS plan ships with automatic scene switching and a preconfigured offline scene, at no extra cost and with nothing to add to your bill. Because it happens inside a real OBS, you design that moment however you want.
• With the Stream Key plan (€9/month), it drives the OBS you already run. We manage the scene switching for the OBS on your own machine: the switch is decided from our ingest, where the bitrate is actually measured, and applied to your local OBS. No port forwarding and no VPS.
• The standalone Failover relay is for people who want only the splice. €9, €19 and €29 a month for 5, 10 and 15 Mbps. It exists for someone broadcasting from their own OBS who wants the platform connection held open and nothing else from us. If you have a Cloud OBS plan you do not need it.

The other half of the problem is making the drop less likely in the first place: SRTLA relay hosting: what the bonded receiver has to do.

Before you buy anything: the checklist

Half the drops we are asked about are fixed by the encoder, for free. Work through this first:

• Measure your real upload where you stream. Not at home, not on Wi-Fi. Walk the route with a speed test and write down the worst number you see.
• Set the bitrate to about 70% of that worst number. The headroom is not waste: it is what absorbs the sag when the cell fills up.
• Use SRT or SRTLA instead of RTMP. The retransmission buffer is the single biggest improvement available to a mobile stream, and it costs nothing.
• Raise the latency setting until it stops helping. A bigger buffer hides longer gaps at the cost of delay. For IRL, two to four seconds is usually the sweet spot.
• Add a second link if you can. Two carriers rarely fail in the same place at the same time. Bonding turns two mediocre connections into one usable one.
• Test the fallback on purpose. Put the phone in airplane mode mid-stream and watch what your viewers would see. Whatever surprises you there is the thing to fix.

Frequently asked questions

Related

• IRL streaming server: what it is and which one you need
• SRTLA relay hosting: the bonded receiver you do not have to run
• How failover keeps your stream on air
• What to do when your bitrate is unstable
• How automatic scene switching works
• Recommended thresholds for scene switching