When your pod is terminating, it takes a while for requests to stop hitting your pod. In your pod definitions, always have a
preStopwithsleep 20to allow Endpoint changes to propagate. AfterpreStopis when Kubernetes hits your containers with aSIGTERM.Always have CPU requests. Be very wary of CPU limits -- don't add them unless you really need to. This allows your container to use more CPU when it's available. But when CPU pressure comes in, your app is guaranteed the amount of CPU cycles it requested.
A corollary to the above: there is always tension between your cluster's utilization and stability. You can increase your utilization rate i.e. by increasing the difference between requests and limits. But that ups the chance of eviction. Reduce utilization, and you'll see less evictions. Find the balance for you.
Automation is not the end-all-be-all. When your automation doesn't work, someone has to fix it after all.
For further actions, you may consider blocking this person and/or reporting abuse
Top comments (0)