mirror of https://github.com/OpenVidu/openvidu.git
bf54f336e0
Every Media Node runs openvidu-media-health, a watchdog that asks the cloud to replace it with a new node (keeping the Media Node count) when its bootstrap failed or never finished, or when LiveKit fails its health check on :7880 for 5 minutes (down) or 10 minutes (errors or timeouts; not counted while the CPU is saturated). Nothing is replaced while no master answers a Redis PING, so a master outage cannot replace the whole media fleet. Drain scripts pause the watchdog, and skip waiting for sessions on a node being replaced. - AWS: the node marks itself Unhealthy in its Auto Scaling group (the lifecycle hook drains it and a new instance is launched). aws-cli is installed before the watchdog starts; set_as_unhealthy.sh is gone. - GCP: the node asks its managed instance group to recreate it. Media autohealing (TCP 7880) now waits 3600 s before judging a new node and needs 5 minutes of failures; HA gets the same autohealing plus the firewall rule for the health check ranges. - Azure: the node reimages itself (new OS disk, bootstrap runs again, same instance, scale-in protection kept) instead of deleting itself, which shrank the scale set. The .json twins are rebuilt from the .bicep. |
||
|---|---|---|
| .. | ||
| output.tf | ||
| tf-gpc-openvidu-ha.tf | ||
| variables.tf | ||
| versions.tf | ||