From a69ddbde56dd59671de70a6307b70a3d92917008 Mon Sep 17 00:00:00 2001
From: kisfenyo <nagyfenyvesi.viktor@gmail.com>
Date: Tue, 27 Oct 2026 21:45:58 +0100
Subject: [PATCH] =?UTF-8?q?hub=20manifest=20(R-1326):=20a=20startupProbe?=
 =?UTF-8?q?=20gives=20the=20hub=20up=20to=20180=20s=20to=20start=20(measur?=
 =?UTF-8?q?ed=2031=E2=80=9336=20s);=20liveness=20unchanged=20after=20start?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 manifests/hub.yaml | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/manifests/hub.yaml b/manifests/hub.yaml
index 99c35698..856b78a4 100644
--- a/manifests/hub.yaml
+++ b/manifests/hub.yaml
@@ -249,6 +249,18 @@ spec:
             - name: agent-poke
               mountPath: /etc/hub-secrets/agent-poke
               readOnly: true
+          # R-1326 (2026-10-27): the hub took 31–36 s to start („starting" → „Listening on :8080", 0.174.0 and 0.174.1),
+          # and the liveness probe alone killed it after ~65 s — a growing database would have turned a deploy into a
+          # restart loop. The startup probe allows up to 180 s (36 × 5 s, 5× the measured start) and holds the liveness
+          # probe off until its first success; after a normal start the liveness probe watches as before. A longer
+          # initialDelaySeconds would leave the hub unwatched for that long on EVERY start instead.
+          startupProbe:
+            httpGet:
+              path: /healthz
+              port: 8080
+            periodSeconds: 5
+            timeoutSeconds: 3
+            failureThreshold: 36
           livenessProbe:
             httpGet:
               path: /healthz