update 06/28/24 12:42:47

Qovery · Jun 28, 2024 · 0ec0018 · 0ec0018
1 parent c608c1e
commit 0ec0018
Show file tree

Hide file tree

Showing 2 changed files with 34 additions and 4 deletions.
diff --git a/charts/qovery/values-aws.yaml b/charts/qovery/values-aws.yaml
@@ -122,8 +122,13 @@ ingress-nginx:
       default-ssl-certificate: "qovery/letsencrypt-acme-qovery-cert"
     updateStrategy:
       rollingUpdate:
-        # set the minimum acceptable number of unavailable pods during a rolling update
-        maxUnavailable: 1
+        # AWS LB is slow to catchup change in the topology, so we go 1 by 1 to not have any downtime
+        maxSurge: 1
+        maxUnavailable: 0
+    # AWS LB is slow to catchup change in the topology, so we go slowly to let AWS catchup change before moving to the next instance
+    # LB healthcheck is 6, and need 2 rounds to consider the instance as (un)healthy. Double the time to be safe
+    readinessProbe:
+      initialDelaySeconds: 30
     # enable autoscaling if you want to scale the number of replicas based on CPU usage
     autoscaling:
       enabled: false # set-by-customer
@@ -144,6 +149,16 @@ ingress-nginx:
       externalTrafficPolicy: "Local"
       sessionAffinity: ""
       healthCheckNodePort: 0
+    # force a connection for 30 seconds before shutting down, to avoid exiting too early
+    # and let time to AWS LB to catchup change in the topology
+    # When /wait-shutdown is called, the LB healthcheck /healthz endpoint return an error, but nginx keep processing request
+    lifecycle:
+      preStop:
+        exec:
+          command:
+            - sh
+            - -c
+            - (sleep 30 | nc localhost 80)&  sleep 1 ; /wait-shutdown
     topologySpreadConstraints:
       - labelSelector:
           matchLabels:

diff --git a/charts/qovery/values-demo-aws.yaml b/charts/qovery/values-demo-aws.yaml
@@ -122,8 +122,13 @@ ingress-nginx:
       default-ssl-certificate: "qovery/letsencrypt-acme-qovery-cert"
     updateStrategy:
       rollingUpdate:
-        # set the minimum acceptable number of unavailable pods during a rolling update
-        maxUnavailable: 1
+        # AWS LB is slow to catchup change in the topology, so we go 1 by 1 to not have any downtime
+        maxSurge: 1
+        maxUnavailable: 0
+    # AWS LB is slow to catchup change in the topology, so we go slowly to let AWS catchup change before moving to the next instance
+    # LB healthcheck is 6, and need 2 rounds to consider the instance as (un)healthy. Double the time to be safe
+    readinessProbe:
+      initialDelaySeconds: 30
     # enable autoscaling if you want to scale the number of replicas based on CPU usage
     autoscaling:
       enabled: false # set-by-customer
@@ -144,6 +149,16 @@ ingress-nginx:
       externalTrafficPolicy: "Local"
       sessionAffinity: ""
       healthCheckNodePort: 0
+    # force a connection for 30 seconds before shutting down, to avoid exiting too early
+    # and let time to AWS LB to catchup change in the topology
+    # When /wait-shutdown is called, the LB healthcheck /healthz endpoint return an error, but nginx keep processing request
+    lifecycle:
+      preStop:
+        exec:
+          command:
+            - sh
+            - -c
+            - (sleep 30 | nc localhost 80)&  sleep 1 ; /wait-shutdown
     topologySpreadConstraints:
       - labelSelector:
           matchLabels: