Detect the actual failure
Automation should distinguish between a missing source, a stopped encoder process, a rejected destination and a local Internet problem. Restarting the encoder repeatedly will not fix invalid credentials or a platform outage, and it can hide the condition operators need to diagnose.
Use bounded reconnect logic
Reconnect after a short delay and increase the delay when failures continue. Record timestamps and reasons. Unbounded rapid retries waste CPU and network resources. A healthy design returns to normal promptly after a brief interruption but remains observable during a persistent fault.
Keep the upload path resilient
StarBonding can reduce restarts caused by a single local link dropping because working links remain available inside the bonded session. The encoder still needs appropriate reconnect settings for server, application or destination interruptions outside the local access links.
Test recovery deliberately
Disconnect a modem, disable venue Ethernet, stop the source and temporarily use an invalid destination in separate tests. Confirm which component recovers each condition, how long recovery takes and whether the operator receives enough information to act.