|
| 1 | +# Zero downtime deployments |
| 2 | + |
| 3 | +This is a list of things you should consider when dealing with a high traffic production environment if you want to |
| 4 | +minimise the impact of rolling updates and downscaling. |
| 5 | + |
| 6 | +### Deployment strategy |
| 7 | + |
| 8 | +Limit the number of unavailable pods during a rolling update: |
| 9 | + |
| 10 | +```yaml |
| 11 | +apiVersion: apps/v1 |
| 12 | +kind: Deployment |
| 13 | +spec: |
| 14 | + progressDeadlineSeconds: 120 |
| 15 | + strategy: |
| 16 | + type: RollingUpdate |
| 17 | + rollingUpdate: |
| 18 | + maxUnavailable: 0 |
| 19 | +``` |
| 20 | +
|
| 21 | +The default progress deadline for a deployment is ten minutes. |
| 22 | +You should consider adjusting this value to make the deployment process fail faster. |
| 23 | +
|
| 24 | +### Liveness health check |
| 25 | +
|
| 26 | +You application should expose a HTTP endpoint that Kubernetes can call to determine if |
| 27 | +your app transitioned to a broken state from which it can't recover and needs to be restarted. |
| 28 | +
|
| 29 | +```yaml |
| 30 | +readinessProbe: |
| 31 | + exec: |
| 32 | + command: |
| 33 | + - wget |
| 34 | + - --quiet |
| 35 | + - --tries=1 |
| 36 | + - --timeout=4 |
| 37 | + - --spider |
| 38 | + - http://localhost:8080/healthz |
| 39 | + timeoutSeconds: 5 |
| 40 | + initialDelaySeconds: 5 |
| 41 | +``` |
| 42 | +
|
| 43 | +If you've enabled mTLS, you'll have to use `exec` for liveness and readiness checks since |
| 44 | +kubelet is not part of the service mesh and doesn't have access to the TLS cert. |
| 45 | + |
| 46 | +### Readiness health check |
| 47 | + |
| 48 | +You application should expose a HTTP endpoint that Kubernetes can call to determine if |
| 49 | +your app is ready to receive traffic. |
| 50 | + |
| 51 | +```yaml |
| 52 | +livenessProbe: |
| 53 | + exec: |
| 54 | + command: |
| 55 | + - wget |
| 56 | + - --quiet |
| 57 | + - --tries=1 |
| 58 | + - --timeout=4 |
| 59 | + - --spider |
| 60 | + - http://localhost:8080/readyz |
| 61 | + timeoutSeconds: 5 |
| 62 | + initialDelaySeconds: 5 |
| 63 | + periodSeconds: 5 |
| 64 | +``` |
| 65 | + |
| 66 | +If your app depends on external services, you should check if those services are available before allowing Kubernetes |
| 67 | +to route traffic to an app instance. Keep in mind that the Envoy sidecar can have a slower startup than your app. |
| 68 | +This means that on application start you should retry for at least a couple of seconds any external connection. |
| 69 | + |
| 70 | +### Graceful shutdown |
| 71 | + |
| 72 | +Before a pod gets terminated, Kubernetes sends a `SIGTERM` signal to every container and waits for period of |
| 73 | +time (30s by default) for all containers to exit gracefully. If your app doesn't handle the `SIGTERM` signal or if it |
| 74 | +doesn't exit within the grace period, Kubernetes will kill the container and any inflight requests that your app is |
| 75 | +processing will fail. |
| 76 | + |
| 77 | +```yaml |
| 78 | +apiVersion: apps/v1 |
| 79 | +kind: Deployment |
| 80 | +spec: |
| 81 | + template: |
| 82 | + spec: |
| 83 | + terminationGracePeriodSeconds: 60 |
| 84 | + containers: |
| 85 | + - name: app |
| 86 | + lifecycle: |
| 87 | + preStop: |
| 88 | + exec: |
| 89 | + command: |
| 90 | + - sleep |
| 91 | + - "10" |
| 92 | +``` |
| 93 | + |
| 94 | +Your app container should have a `preStop` hook that delays the container shutdown. |
| 95 | +This will allow the service mesh to drain the traffic and remove this pod from all other Envoy sidecars before your app |
| 96 | +becomes unavailable. |
| 97 | + |
| 98 | +### Delay Envoy shutdown |
| 99 | + |
| 100 | +Even if your app reacts to `SIGTERM` and tries to complete the inflight requests before shutdown, that |
| 101 | +doesn't mean that the response will make it back to the caller. If the Envoy sidecar shuts down before your app, then |
| 102 | +the caller will receive a 503 error. |
| 103 | + |
| 104 | +To mitigate this issue you can add a `preStop` hook to the Istio proxy and wait for the main app to exist before Envoy exists. |
| 105 | + |
| 106 | +```bash |
| 107 | +#!/bin/bash |
| 108 | +set -e |
| 109 | +if ! pidof envoy &>/dev/null; then |
| 110 | + exit 0 |
| 111 | +fi |
| 112 | +
|
| 113 | +if ! pidof pilot-agent &>/dev/null; then |
| 114 | + exit 0 |
| 115 | +fi |
| 116 | +
|
| 117 | +while [ $(netstat -plunt | grep tcp | grep -v envoy | wc -l | xargs) -ne 0 ]; do |
| 118 | + sleep 1; |
| 119 | +done |
| 120 | +
|
| 121 | +exit 0 |
| 122 | +``` |
| 123 | + |
| 124 | +You'll have to build your own Envoy docker image with the above script and |
| 125 | +modify the Istio injection webhook with the `preStop` directive. |
| 126 | + |
| 127 | +Thanks to Stono for his excellent [tips](https://github.com/istio/istio/issues/12183) on minimising 503s. |
| 128 | + |
| 129 | +### Resource requests and limits |
| 130 | + |
| 131 | +Setting CPU and memory requests/limits for all workloads is a mandatory step if you're running a production system. |
| 132 | +Without limits your nodes could run out of memory or become unresponsive due to CPU exhausting. |
| 133 | +Without CPU and memory requests, |
| 134 | +the Kubernetes scheduler will not be able to make decisions about which nodes to place pods on. |
| 135 | + |
| 136 | +```yaml |
| 137 | +apiVersion: apps/v1 |
| 138 | +kind: Deployment |
| 139 | +spec: |
| 140 | + template: |
| 141 | + spec: |
| 142 | + containers: |
| 143 | + - name: app |
| 144 | + resources: |
| 145 | + limits: |
| 146 | + cpu: 1000m |
| 147 | + memory: 1Gi |
| 148 | + requests: |
| 149 | + cpu: 100m |
| 150 | + memory: 128Mi |
| 151 | +``` |
| 152 | + |
| 153 | +Note that without resource requests the horizontal pod autoscaler can't determine when to scale your app. |
| 154 | + |
| 155 | +### Autoscaling |
| 156 | + |
| 157 | +A production environment should be able to handle traffic bursts without impacting the quality of service. |
| 158 | +This can be achieved with Kubernetes autoscaling capabilities. |
| 159 | +Autoscaling in Kubernetes has two dimensions: the Cluster Autoscaler that deals with node scaling operations and |
| 160 | +the Horizontal Pod Autoscaler that automatically scales the number of pods in a deployment. |
| 161 | + |
| 162 | +```yaml |
| 163 | +apiVersion: autoscaling/v2beta1 |
| 164 | +kind: HorizontalPodAutoscaler |
| 165 | +spec: |
| 166 | + scaleTargetRef: |
| 167 | + apiVersion: apps/v1 |
| 168 | + kind: Deployment |
| 169 | + name: app |
| 170 | + minReplicas: 2 |
| 171 | + maxReplicas: 4 |
| 172 | + metrics: |
| 173 | + - type: Resource |
| 174 | + resource: |
| 175 | + name: cpu |
| 176 | + targetAverageValue: 900m |
| 177 | + - type: Resource |
| 178 | + resource: |
| 179 | + name: memory |
| 180 | + targetAverageValue: 768Mi |
| 181 | +``` |
| 182 | + |
| 183 | +The above HPA ensures your app will be scaled up before the pods reach the CPU or memory limits. |
| 184 | + |
| 185 | +### Ingress retries |
| 186 | + |
| 187 | +To minimise the impact of downscaling operations you can make use of Envoy retry capabilities. |
| 188 | + |
| 189 | +```yaml |
| 190 | +apiVersion: flagger.app/v1alpha3 |
| 191 | +kind: Canary |
| 192 | +spec: |
| 193 | + service: |
| 194 | + port: 9898 |
| 195 | + gateways: |
| 196 | + - public-gateway.istio-system.svc.cluster.local |
| 197 | + hosts: |
| 198 | + - app.example.com |
| 199 | + appendHeaders: |
| 200 | + x-envoy-upstream-rq-timeout-ms: "15000" |
| 201 | + x-envoy-max-retries: "10" |
| 202 | + x-envoy-retry-on: "gateway-error,connect-failure,refused-stream" |
| 203 | +``` |
| 204 | + |
| 205 | +When the HPA scales down your app, your users could run into 503 errors. |
| 206 | +The above configuration will make Envoy retry the HTTP requests that failed due to gateway errors. |
0 commit comments