Skip to main content
DevOps Insightsguides & answers

DevOps Insights

Start from the error, not the table of contents

Named failure states with every cause we can document, a command that confirms or rules out each one, and how to check the fix held. Written by someone who runs this in production — including the parts that break at 3am.

Known error states resolve instantly. Anything else searches the whole site.

or start fromCrashLoopBackOffOOMKilledexit 137ImagePullBackOffPending

Symptom

ImagePullBackOff

5 documented causes

  1. The image name or tag is wrong

    kubectl describe pod <pod> | grep -A5 Events

  2. Missing or wrong registry credentials

    kubectl get pod <pod> -o jsonpath='{.spec.imagePullSecrets}'

  3. The registry is implied rather than stated

    kubectl get pod <pod> -o jsonpath='{.spec.containers[*].image}'

  4. The node cannot reach the registry

    kubectl describe pod <pod> | grep -i -E 'timeout|no such host|connection refused|i/o timeout'
Open the full page
Read from this site’s own catalogue, not written for this panel. Ordered by how often each cause turns out to be the one — the commands decide it, because nothing here can see your cluster.

Where do you want to start?

New to this

Start here

An ordered path through the fundamentals, so you are not picking articles at random.

Something is broken

Troubleshooting

Named errors with every cause, a command that confirms each one, and how to verify the fix.

Not broken yet

Check a config

Paste a manifest or Dockerfile. Findings link to the failure each one causes, before it causes it.

Want to understand

Learn it properly

Ordered paths, the long guides, and four kinds of explanation — each labelled with the question it answers.

Want depth

Browse by topic

Everything on one subject in one place: articles, questions and guides together.

How every troubleshooting page works

See it on CrashLoopBackOff →

Not a list of articles about Kubernetes. A path from the error you are looking at to a cause you have confirmed — because nothing here can see your cluster, so nothing here asks you to take its word for it.

  1. 01 · SymptomThe string on your screen.CrashLoopBackOff
  2. 02 · EvidenceOne command that shows where you are.kubectl describe pod
  3. 03 · Cause spaceEvery cause it could be, ranked by how often it is the answer.6 documented causes
  4. 04 · ConfirmA command per cause that proves or kills it.Last State: OOMKilled
  5. 05 · FixWhat to change, once you know which one it is.Raise the memory limit
  6. 06 · VerifyHow to know it actually worked.Restart count stops rising

Recently updated

Latest

All articles →
TerraformIntermediateNew3 min read

Terraform Identity Errors After a Provider Upgrade

Missing Resource Identity After Read and Unexpected Identity Change usually start right after a provider upgrade. Why they are provider bugs, and the upgrade-or-pin fix that clears them.

#Terraform#AWS

Neeraj Jha

KubernetesIntermediateNew3 min read

OOMKilled and Exit Code 137: Limit, Leak or Spike?

OOMKilled means the kernel killed a Kubernetes container for using too much memory. How to tell a low limit from a leak or a one-off spike before you raise anything.

#Kubernetes#troubleshooting

Neeraj Jha

ContainersIntermediateNew3 min read

docker push Failing Since Docker 29.7? Two Causes

Since Docker 29.7, docker push can fail with blob unknown to registry on GitLab or with timeout awaiting response headers on slow registries. How to tell which one you have, and what fixes each.

#Docker#CI/CD

Neeraj Jha

KubernetesBeginnerNew3 min read

CrashLoopBackOff: Find the Cause in Three Commands

CrashLoopBackOff means a Kubernetes container keeps exiting and being restarted. Read the previous container's logs, its exit reason and the pod events, and the cause is usually on screen.

#Kubernetes#troubleshooting

Neeraj Jha

What is actually in here

Small on purpose. Every failure state here lists every cause we could document and source, which is only possible for problems that have a finite cause set — so the catalogue grows when there is something complete to add, not to fill a page.