Resource Quotas and Resource Limits Slides

rq_rl-slides.html

<section data-background-transition='zoom' data-transition='concave' data-background='https://pixel.nymag.com/imgs/daily/vulture/2016/07/29/29-steve-brule-check-it-out.w710.h473.2x.jpg' data-state='blackout'>
  <h2>Kubernetes Resource Quotas and Resource Limits</h2>
  </br>
  <h3>For your health</h3>
</section>
<section data-background-transition='zoom' data-transition='linear' id='try-it'>
  <section data-transition='concave'>
    <h2>Resource Limits vs Resource Quotas</h2>
    <p>Memory and CPU constraints for a container are configured with resource requests and resource limits</p>
    <p><b>Resource Quotas:</b> are resource limits that apply to a namespace</p>
  </section>
  <section>
    <h2>Why are these things important?</h2>
    <p>When Containers have resource requests specified, the scheduler can make better decisions about which nodes to place Pods on. And when Containers have their limits specified, contention for resources on a node can be handled in a specified manner.</p>
    <p>When well configured resource quotas are set for a namespace, it is much harder for overall cluster resource contention to affect other namespaces</p>
  </section>
  <section data-transition='concave'>
    <h2>Determining appropriate values</h2>
    <img src="https://i.imgur.com/tovzLOL.png" width="80%"></img>
  </section>
  <section data-transition='concave'>
    <h2>Setting resource quotas for your namespace</h2>
    <pre><code>
    Name:            platform-rq
    Namespace:       platform
    Resource         Used    Hard
    --------         ----    ----
    limits.cpu       1760m   2
    limits.memory    1924Mi  2500M
    requests.cpu     1120m   1500m
    requests.memory  1275Mi  2G
    </pre></code>
  </section>
  <section data-transition='concave'>
    <h2>Setting resource limits in your deployments</h2>
    <p>Because we use resource quotas, any container that does not have requests.cpu, requests.memory, limits.cpu, limits.memory will cause the pod to fail to deploy</p>
    <pre><code>
    resources:
      limits:
        cpu: "100m"
        memory: "100Mi"
      requests:
        cpu: "55m"
        memory: "75Mi"
    </pre></code>
  </section>
  <section data-transition='concave'>
    <h2>Avoid overcommitting</h2>
    <p>This is a situation that we want to avoid if possible</p>
    <pre><code>
  Namespace                  Name                                  CPU Requests  CPU Limits   Memory Requests  Memory Limits
  ---------                  ----                                  ------------  ----------   ---------------  -------------
  ingestion                  ingestion-service-6d549f7669-gvr4m    0 (0%)        0 (0%)       0 (0%)           0 (0%)
  platform                   freshtracks-agent-744656f5d-7zdqf     510m (39%)    1010m (77%)  520Mi (18%)      1044Mi (36%)
  platform                   freshtracks-node-exporter-9wxjv       0 (0%)        0 (0%)       0 (0%)           0 (0%)
  platform                   heimdall-74b56cfb4c-mg8fr             120m (9%)     150m (11%)   135Mi (4%)       160Mi (5%)
  platform                   p8s-prometheus-node-exporter-6prrq    0 (0%)        0 (0%)       0 (0%)           0 (0%)
  platform                   tiller-deploy-5664745cb7-qp8c9        0 (0%)        0 (0%)       0 (0%)           0 (0%)
  reporting                  token-service-786b7d9c4d-fftmh        250m (19%)    500m (38%)   64Mi (2%)        128Mi (4%)
  waffleio-dev               tiller-deploy-58c9d747cc-zpkzm        0 (0%)        0 (0%)       0 (0%)           0 (0%)
  waffleio-production        tiller-deploy-dc99644dc-bnxbk         0 (0%)        0 (0%)       0 (0%)           0 (0%)
Allocated resources:
  (Total limits may be over 100 percent, i.e., overcommitted.)
  CPU Requests  CPU Limits    Memory Requests  Memory Limits
  ------------  ----------    ---------------  -------------
  880m (67%)    1660m (127%)  719Mi (25%)      1332Mi (46%)
    </pre></code>
  </section>
  <section>
    <h2>Pod eviction</h2>
    <p>--kube-reserved and --evicition-threshold kubelet flags will evict pods from an overstuffed node in favor of preserving the kubelet</p>
  </section>
  <section data-transition='concave'>
    <h2>Troubleshooting failed deploys</h2>
    <p><code>kubectl get ev -n dana</code></p>
    <p>Loggly</p>
    <code>
      syslog.host:*.dev "exceeded quota"
    </code>
  </section>
  <section>
    <h2>Tweaking your deployments for resource limits</h2>
    <p>No limits - see error, set limits</p>
    <p>Changing replicas</p>
    <p>Changing max surge</p>
    <p>Bump the quota</p>
  </section>
</section>