From f937c2589d4f8c178872c44b95982cc2690547b4 Mon Sep 17 00:00:00 2001
From: alkatrazz <alkatrazz@localhost.localdomain>
Date: Thu, 14 May 2026 17:57:12 +0200
Subject: [PATCH] =?UTF-8?q?fix(funk-cluster):=20grep=20-c=20bug=20+=20unco?=
 =?UTF-8?q?rdon=20automatique=20au=20d=C3=A9marrage?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- check_k8s_pods : grep -c retournait "0\n0" (|| echo 0 inutile quand
  grep-c affiche 0 même en cas d'absence de match) — utilise un cache
  intermédiaire pods_output + running=$(grep -c) || running=0
- cmd_start : uncordon automatique si nœuds SchedulingDisabled détectés
  (séquelle d'un funk-stop --k8s), suivi d'une attente pod infra avant
  le check — évite de devoir passer --k8s juste pour débloquer le cluster

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 ansible/roles/litellm/files/funk-cluster | 35 +++++++++++++++++++-----
 1 file changed, 28 insertions(+), 7 deletions(-)

diff --git a/ansible/roles/litellm/files/funk-cluster b/ansible/roles/litellm/files/funk-cluster
index 1c2e89f..b77a4af 100644
--- a/ansible/roles/litellm/files/funk-cluster
+++ b/ansible/roles/litellm/files/funk-cluster
@@ -152,20 +152,20 @@ check_k8s() {
 
 check_k8s_pods() {
     local ns=$1 label=$2
-    local total running
-    total=$($KUBECTL get pods -n "$ns" --no-headers 2>/dev/null | wc -l)
-    running=$($KUBECTL get pods -n "$ns" --no-headers 2>/dev/null | grep -c "^[^ ]* *[0-9]*/[0-9]* *Running" || echo 0)
-
-    if [ "$total" -eq 0 ]; then
+    local pods_output total running
+    pods_output=$($KUBECTL get pods -n "$ns" --no-headers 2>/dev/null)
+    if [ -z "$pods_output" ]; then
         warn "$label — namespace vide ou inaccessible"
         return
     fi
+    total=$(echo "$pods_output" | wc -l)
+    running=$(echo "$pods_output" | grep -c " Running ") || running=0
 
     if [ "$running" -eq "$total" ]; then
         check_ok "$label — $running/$total Running"
     else
         check_fail "$label — $running/$total Running"
-        $KUBECTL get pods -n "$ns" --no-headers 2>/dev/null \
+        echo "$pods_output" \
             | grep -v " Running " \
             | awk '{printf "      %-45s %s\n", $1, $3}' \
             | head -6
@@ -419,11 +419,32 @@ cmd_start() {
 
     if [ $k8s_up -gt 0 ]; then
         local ready
-        ready=$($KUBECTL get nodes --no-headers 2>/dev/null | grep -c " Ready" || echo 0)
+        ready=$($KUBECTL get nodes --no-headers 2>/dev/null | grep -c " Ready") || ready=0
         echo ""
         echo "[ cluster k8s — santé pods ]"
         if [ "$ready" -gt 0 ]; then
             check_ok "Nœuds — $ready/3 Ready"
+
+            # Uncordon automatique si nœuds SchedulingDisabled (séquelle d'un funk-stop --k8s)
+            local uncordoned=0
+            for name in "${K8S_NAMES[@]}"; do
+                if $KUBECTL get node "$name" --no-headers 2>/dev/null | grep -q "SchedulingDisabled"; then
+                    $KUBECTL uncordon "$name" 2>/dev/null && ok "$name — uncordonné" || true
+                    uncordoned=$((uncordoned + 1))
+                fi
+            done
+
+            # Après uncordon : attendre que les pods infra se re-schedulisent
+            if [ $uncordoned -gt 0 ]; then
+                info "Attente déploiement pods après uncordon (jusqu'à 2 min)..."
+                local sched_elapsed=0
+                until [ "$($KUBECTL get pods -n infra --no-headers 2>/dev/null | grep -c " Running ")" -ge 1 ] \
+                      || [ $sched_elapsed -ge 120 ]; do
+                    sleep 10; sched_elapsed=$((sched_elapsed + 10))
+                    echo -ne "  → Attente pods... ${sched_elapsed}s\r"
+                done; echo ""
+            fi
+
             check_k8s_workloads
         else
             warn "Nœuds joignables mais pas encore Ready"