From 2f891a5d31d3405418f2a4b6a79e5cae25d377a7 Mon Sep 17 00:00:00 2001
From: mr-forust <vzlomdsisma@gmail.com>
Date: Mon, 28 Sep 2026 12:35:16 +0200
Subject: [PATCH] fix(postgres): give probes room on an I/O-bound single node

pg_isready with the 1s default times out under I/O stall and kubelet kills a healthy postgres mid-recovery; each kill restarts a multi-minute fsync from zero and loops forever. readiness/liveness timeout 5s, liveness threshold 5, startup budget 15min.
---
 postgres/k8s/postgres.yaml | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/postgres/k8s/postgres.yaml b/postgres/k8s/postgres.yaml
index ae83540..51112c8 100644
--- a/postgres/k8s/postgres.yaml
+++ b/postgres/k8s/postgres.yaml
@@ -60,16 +60,26 @@ spec:
               command: ["pg_isready", "-U", "postgres", "-d", "postgres"]
             initialDelaySeconds: 10
             periodSeconds: 10
+            # Generous timeout: on an I/O-bound single node even exec can take
+            # seconds, and a 1s default kills a healthy postgres mid-recovery.
+            timeoutSeconds: 5
           startupProbe:
             exec:
               command: ["pg_isready", "-U", "postgres", "-d", "postgres"]
-            failureThreshold: 30
+            # Crash recovery on an I/O-starved single node can fsync for 10+
+            # minutes; killing postgres mid-recovery restarts the fsync from
+            # zero and loops forever. 90x10s = 15 minutes of grace.
+            failureThreshold: 90
             periodSeconds: 10
           livenessProbe:
             exec:
               command: ["pg_isready", "-U", "postgres", "-d", "postgres"]
             initialDelaySeconds: 30
             periodSeconds: 20
+            # Same I/O reasoning as readiness, plus more misses before a kill:
+            # restarting postgres on a loaded node only makes recovery longer.
+            timeoutSeconds: 5
+            failureThreshold: 5
           resources:
             requests:
               memory: "512Mi"