diff --git a/charts/krateo-sse-proxy/templates/deployment.yaml b/charts/krateo-sse-proxy/templates/deployment.yaml
index c61ba48..b79d308 100644
--- a/charts/krateo-sse-proxy/templates/deployment.yaml
+++ b/charts/krateo-sse-proxy/templates/deployment.yaml
@@ -47,18 +47,31 @@ spec:
                   optional: {{ .Values.clickhouse.passwordSecret.optional }}
             - name: LISTEN_ADDR
               value: {{ .Values.listenAddr | quote }}
+          # Lenient liveness tolerance (org standard): the Kubernetes defaults
+          # (timeoutSeconds: 1, failureThreshold: 3) kill HEALTHY pods under CPU
+          # contention — a slow-but-alive /health reply counts as a failure and the
+          # resulting restart storm worsens the contention. Only a truly dead process
+          # should die: 10s period x 6 failures = 60s of continuous failure before
+          # a restart.
           livenessProbe:
             httpGet:
               path: /health
               port: 8080
             initialDelaySeconds: 5
             periodSeconds: 10
+            timeoutSeconds: 5
+            failureThreshold: 6
+          # Readiness stays prompt (5s period, 3 failures) so a genuinely unhealthy
+          # endpoint is removed from the Service quickly, but timeoutSeconds: 5 keeps
+          # load alone from flapping the endpoint in and out.
           readinessProbe:
             httpGet:
               path: /health
               port: 8080
             initialDelaySeconds: 3
             periodSeconds: 5
+            timeoutSeconds: 5
+            failureThreshold: 3
           resources:
             {{- toYaml .Values.resources | nindent 12 }}
           securityContext:
diff --git a/charts/otel-collector-daemonset/values.schema.json b/charts/otel-collector-daemonset/values.schema.json
index 99bf053..8897d0d 100644
--- a/charts/otel-collector-daemonset/values.schema.json
+++ b/charts/otel-collector-daemonset/values.schema.json
@@ -528,6 +528,48 @@
               }
             }
           }
+        },
+        "livenessProbe": {
+          "type": "object",
+          "additionalProperties": true,
+          "properties": {
+            "initialDelaySeconds": {
+              "type": "integer"
+            },
+            "periodSeconds": {
+              "type": "integer"
+            },
+            "timeoutSeconds": {
+              "type": "integer"
+            },
+            "failureThreshold": {
+              "type": "integer"
+            },
+            "terminationGracePeriodSeconds": {
+              "type": "integer"
+            }
+          }
+        },
+        "readinessProbe": {
+          "type": "object",
+          "additionalProperties": true,
+          "properties": {
+            "initialDelaySeconds": {
+              "type": "integer"
+            },
+            "periodSeconds": {
+              "type": "integer"
+            },
+            "timeoutSeconds": {
+              "type": "integer"
+            },
+            "successThreshold": {
+              "type": "integer"
+            },
+            "failureThreshold": {
+              "type": "integer"
+            }
+          }
         }
       },
       "description": "Values forwarded to the upstream opentelemetry-collector subchart."
diff --git a/charts/otel-collector-daemonset/values.yaml b/charts/otel-collector-daemonset/values.yaml
index e2e23c9..3ba4ad7 100644
--- a/charts/otel-collector-daemonset/values.yaml
+++ b/charts/otel-collector-daemonset/values.yaml
@@ -176,3 +176,18 @@ opentelemetry-collector:
     requests:
       cpu: 100m
       memory: 256Mi
+
+  # Probe tolerances, passed through to the upstream opentelemetry-collector chart's
+  # templates/_pod.tpl (org standard). The Kubernetes defaults (timeoutSeconds: 1,
+  # failureThreshold: 3) kill HEALTHY pods under CPU contention — a slow-but-alive
+  # health reply counts as a failure and the resulting restart storm worsens the
+  # contention. Liveness is lenient so only a truly dead process dies: implicit
+  # periodSeconds default 10 x 6 failures = 60s of continuous failure before restart.
+  livenessProbe:
+    timeoutSeconds: 5
+    failureThreshold: 6
+  # Readiness keeps prompt endpoint removal (failureThreshold: 3, implicit 10s period)
+  # but timeoutSeconds: 5 keeps load alone from flapping the endpoint in and out.
+  readinessProbe:
+    timeoutSeconds: 5
+    failureThreshold: 3
diff --git a/charts/otel-collector-deployment/values.schema.json b/charts/otel-collector-deployment/values.schema.json
index 83ac2cb..c7aa423 100644
--- a/charts/otel-collector-deployment/values.schema.json
+++ b/charts/otel-collector-deployment/values.schema.json
@@ -515,6 +515,48 @@
               }
             }
           }
+        },
+        "livenessProbe": {
+          "type": "object",
+          "additionalProperties": true,
+          "properties": {
+            "initialDelaySeconds": {
+              "type": "integer"
+            },
+            "periodSeconds": {
+              "type": "integer"
+            },
+            "timeoutSeconds": {
+              "type": "integer"
+            },
+            "failureThreshold": {
+              "type": "integer"
+            },
+            "terminationGracePeriodSeconds": {
+              "type": "integer"
+            }
+          }
+        },
+        "readinessProbe": {
+          "type": "object",
+          "additionalProperties": true,
+          "properties": {
+            "initialDelaySeconds": {
+              "type": "integer"
+            },
+            "periodSeconds": {
+              "type": "integer"
+            },
+            "timeoutSeconds": {
+              "type": "integer"
+            },
+            "successThreshold": {
+              "type": "integer"
+            },
+            "failureThreshold": {
+              "type": "integer"
+            }
+          }
         }
       },
       "description": "Values forwarded to the upstream opentelemetry-collector subchart."
diff --git a/charts/otel-collector-deployment/values.yaml b/charts/otel-collector-deployment/values.yaml
index 8f8fd38..bea82f0 100644
--- a/charts/otel-collector-deployment/values.yaml
+++ b/charts/otel-collector-deployment/values.yaml
@@ -263,3 +263,18 @@ opentelemetry-collector:
     requests:
       cpu: 100m
       memory: 256Mi
+
+  # Probe tolerances, passed through to the upstream opentelemetry-collector chart's
+  # templates/_pod.tpl (org standard). The Kubernetes defaults (timeoutSeconds: 1,
+  # failureThreshold: 3) kill HEALTHY pods under CPU contention — a slow-but-alive
+  # health reply counts as a failure and the resulting restart storm worsens the
+  # contention. Liveness is lenient so only a truly dead process dies: implicit
+  # periodSeconds default 10 x 6 failures = 60s of continuous failure before restart.
+  livenessProbe:
+    timeoutSeconds: 5
+    failureThreshold: 6
+  # Readiness keeps prompt endpoint removal (failureThreshold: 3, implicit 10s period)
+  # but timeoutSeconds: 5 keeps load alone from flapping the endpoint in and out.
+  readinessProbe:
+    timeoutSeconds: 5
+    failureThreshold: 3