route-health.sh: add default route management based on gateway health

router/route-health.sh

 #!/bin/sh
 # Health-check failover for route groups via Telegraf/InfluxDB
 # Monitors gateway health and flushes dead routes when live alternatives exist
+# Manages default route for groups with set-default option
 #
 # Usage: route-health.sh [--show]
 
 cd "$(dirname "$(realpath "$0")")" || exit
 
+. ./functions
+
 ROUTES_DIR=routes.d
 ROUTES6_DIR=routes6.d
 STATE_DIR=.state
@@ -52,6 +55,32 @@ if [ -n "$SHOW" ] ; then
     echo
 fi
 
+# Get health status for a monitor tag
+# Usage: get_health TAG
+# Returns: healthy, degraded, dead
+get_health()
+{
+    local tag="$1"
+    local loss=$(grep "^${tag} " "$HEALTH_DATA" | awk '{print $2}')
+
+    if [ -z "$loss" ] ; then
+        echo "dead"
+        return
+    fi
+
+    echo "$loss" | awk '{
+        if ($1 >= 100) print "dead"
+        else if ($1 < 50) print "healthy"
+        else print "degraded"
+    }'
+}
+
+# Check if option is set in group's options file
+has_option()
+{
+    [ -f "$1/options" ] && grep -q "^$2$" "$1/options" 2>/dev/null
+}
+
 # --- Evaluate health for all monitored groups ---
 healthy_count=0
 dead_count=0
@@ -68,20 +97,12 @@ for routes_dir in "$ROUTES_DIR" "$ROUTES6_DIR" ; do
         tag=$(grep -v '^#' "$gwdir/monitor" | grep -m1 .)
         [ -n "$tag" ] || continue
 
+        status=$(get_health "$tag")
         loss=$(grep "^${tag} " "$HEALTH_DATA" | awk '{print $2}')
-        state_path="$routes_dir/$name"
+        loss_display="${loss:-no_data}%"
+        [ -z "$loss" ] && loss_display="no_data"
 
-        if [ -z "$loss" ] ; then
-            status="dead"
-            loss_display="no_data"
-        else
-            status=$(echo "$loss" | awk '{
-                if ($1 >= 100) print "dead"
-                else if ($1 < 50) print "healthy"
-                else print "degraded"
-            }')
-            loss_display="${loss}%"
-        fi
+        state_path="$routes_dir/$name"
 
         [ -n "$SHOW" ] && log "[$name] $tag: loss=$loss_display -> $status"
 
@@ -103,6 +124,74 @@ done
 # Nothing monitored — exit
 [ "$healthy_count" -eq 0 ] && [ "$dead_count" -eq 0 ] && exit 0
 
+# --- Manage default route for groups with set-default ---
+for routes_dir in "$ROUTES_DIR" "$ROUTES6_DIR" ; do
+    [ -d "$routes_dir" ] || continue
+    ipcmd="ip"
+    [ "$routes_dir" = "$ROUTES6_DIR" ] && ipcmd="ip -6"
+
+    for gwdir in "$routes_dir"/*/ ; do
+        [ -d "$gwdir" ] || continue
+        has_option "$gwdir" "set-default" || continue
+
+        name=$(basename "$gwdir")
+        best_gw=""
+
+        # Read gateways in priority order, pick first healthy
+        while IFS= read -r line ; do
+            [ -z "$line" ] && continue
+            echo "$line" | grep -q '^#' && continue
+
+            gw_ip=$(resolve_gw "$line" "$ipcmd")
+            [ -z "$gw_ip" ] && continue
+
+            # Find monitor tag for this gateway IP
+            # Look through all groups for one whose gateway matches
+            gw_tag=""
+            for check_dir in "$routes_dir"/*/ ; do
+                [ -d "$check_dir" ] || continue
+                [ -f "$check_dir/monitor" ] || continue
+                check_gw=$(grep -v '^#' "$check_dir/gateway" 2>/dev/null | grep -m1 .)
+                check_ip=$(resolve_gw "$check_gw" "$ipcmd" 2>/dev/null)
+                if [ "$check_ip" = "$gw_ip" ] ; then
+                    gw_tag=$(grep -v '^#' "$check_dir/monitor" | grep -m1 .)
+                    break
+                fi
+            done
+
+            if [ -z "$gw_tag" ] ; then
+                # No monitor for this gateway — assume healthy (first unmonitored wins)
+                [ -n "$SHOW" ] && log "[$name] $gw_ip: no monitor, assuming healthy"
+                best_gw="$gw_ip"
+                break
+            fi
+
+            status=$(get_health "$gw_tag")
+            [ -n "$SHOW" ] && log "[$name] $gw_ip ($gw_tag): $status"
+
+            if [ "$status" = "healthy" ] ; then
+                best_gw="$gw_ip"
+                break
+            fi
+        done < "$gwdir/gateway"
+
+        if [ -z "$best_gw" ] ; then
+            log "[$name] No healthy gateway found, keeping current default"
+            continue
+        fi
+
+        cur_default=$($ipcmd route show default | awk '/default/ {print $3; exit}')
+        if [ "$cur_default" != "$best_gw" ] ; then
+            log "[$name] Switching default: $cur_default -> $best_gw"
+            if [ -z "$SHOW" ] ; then
+                $ipcmd route replace default via "$best_gw"
+            fi
+        else
+            [ -n "$SHOW" ] && log "[$name] Default already via $best_gw"
+        fi
+    done
+done
+
 # --- Process dead groups ---
 for state_path in $dead_groups ; do
     name=$(basename "$state_path")
@@ -110,9 +199,13 @@ for state_path in $dead_groups ; do
 
     mkdir -p "$STATE_DIR/$state_path"
 
-    # Read table number from group config
+    # Read table number from group config or rt_tables
     table=""
-    [ -f "$state_path/table" ] && table=$(grep -v '^#' "$state_path/table" | grep -m1 .)
+    if [ -f "$state_path/table" ] ; then
+        table=$(grep -v '^#' "$state_path/table" | grep -m1 .)
+    else
+        table=$(awk -v n="$name" '$2 == n { print $1; exit }' /etc/iproute2/rt_tables 2>/dev/null)
+    fi
 
     # Flapping protection: require DOWN_THRESHOLD consecutive checks
     down_count=0