openvidu-deployment: Media Node health watchdog: withdraw the unhealthy mark when the replacement request fails

The replacement script marks the node unhealthy (/run/openvidu-unhealthy) before it asks
the cloud for a new node, so that the node then leaves without waiting for its sessions.
When the request failed (a cloud API error, a missing permission) the mark stayed: the
node remained a member and, if LiveKit recovered, a later normal scale-in would drop its
sessions instead of draining them. The mark is now withdrawn on any failed exit of the
request.

openvidu-deployment/pro/elastic/aws/cf-openvidu-elastic.yaml

@@ -2353,6 +2353,9 @@ Resources:
                 echo "$(date -u +%FT%TZ) $1" >> /var/lib/openvidu-media-health/replace-requests.log
                 # Let stop_media_node.sh know this node is unhealthy, so it does not wait for sessions
                 touch /run/openvidu-unhealthy
+                # Withdrawn if the request fails: the node then stays, and if it recovers
+                # a later scale-in must still wait for its sessions
+                trap '[ $? -eq 0 ] || rm -f /run/openvidu-unhealthy' EXIT
 
                 # Token for IMDSv2
                 TOKEN=$(curl -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 21600")

openvidu-deployment/pro/elastic/azure/cf-openvidu-elastic.bicep

@@ -1263,6 +1263,9 @@ var media_replaceScriptMediaTemplate = '''
 mkdir -p /var/lib/openvidu-media-health
 echo "$(date -u +%FT%TZ) $1" >> /var/lib/openvidu-media-health/replace-requests.log
 touch /run/openvidu-unhealthy
+# Withdrawn if the request fails: the node then stays, and if it recovers
+# a later scale-in must still wait for its sessions
+trap '[ $? -eq 0 ] || rm -f /run/openvidu-unhealthy' EXIT
 
 RESOURCE_GROUP_NAME=${resourceGroupName}
 VM_SCALE_SET_NAME=${vmScaleSetName}

openvidu-deployment/pro/elastic/gcp/tf-gpc-openvidu-elastic.tf

@@ -1610,6 +1610,9 @@ mkdir -p /var/lib/openvidu-media-health
 echo "$(date -u +%FT%TZ) $REASON" >> /var/lib/openvidu-media-health/replace-requests.log
 # Tell the drain script this node is being replaced so it skips waiting for sessions.
 touch /run/openvidu-unhealthy
+# Withdrawn if the request fails: the node then stays, and if it recovers
+# a later scale-in must still wait for its sessions
+trap '[ $? -eq 0 ] || rm -f /run/openvidu-unhealthy' EXIT
 
 META="http://metadata.google.internal/computeMetadata/v1"
 meta() { curl -s -H "Metadata-Flavor: Google" "$META/$1"; }

openvidu-deployment/pro/ha/aws/cf-openvidu-ha.yaml

@@ -2996,6 +2996,9 @@ Resources:
                 echo "$(date -u +%FT%TZ) $1" >> /var/lib/openvidu-media-health/replace-requests.log
                 # Let stop_media_node.sh know this node is unhealthy, so it does not wait for sessions
                 touch /run/openvidu-unhealthy
+                # Withdrawn if the request fails: the node then stays, and if it recovers
+                # a later scale-in must still wait for its sessions
+                trap '[ $? -eq 0 ] || rm -f /run/openvidu-unhealthy' EXIT
 
                 # Token for IMDSv2
                 TOKEN=$(curl -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 21600")

openvidu-deployment/pro/ha/azure/cf-openvidu-ha.bicep

@@ -1656,6 +1656,9 @@ var media_replaceScriptMediaTemplate = '''
 mkdir -p /var/lib/openvidu-media-health
 echo "$(date -u +%FT%TZ) $1" >> /var/lib/openvidu-media-health/replace-requests.log
 touch /run/openvidu-unhealthy
+# Withdrawn if the request fails: the node then stays, and if it recovers
+# a later scale-in must still wait for its sessions
+trap '[ $? -eq 0 ] || rm -f /run/openvidu-unhealthy' EXIT
 
 RESOURCE_GROUP_NAME=${resourceGroupName}
 VM_SCALE_SET_NAME=${vmScaleSetName}

openvidu-deployment/pro/ha/gcp/tf-gpc-openvidu-ha.tf

@@ -1994,6 +1994,9 @@ mkdir -p /var/lib/openvidu-media-health
 echo "$(date -u +%FT%TZ) $REASON" >> /var/lib/openvidu-media-health/replace-requests.log
 # Tell the drain script this node is being replaced so it skips waiting for sessions.
 touch /run/openvidu-unhealthy
+# Withdrawn if the request fails: the node then stays, and if it recovers
+# a later scale-in must still wait for its sessions
+trap '[ $? -eq 0 ] || rm -f /run/openvidu-unhealthy' EXIT
 
 META="http://metadata.google.internal/computeMetadata/v1"
 meta() { curl -s -H "Metadata-Flavor: Google" "$META/$1"; }