Blog
The blog is where we share what we're learning as we build the platform. New posts as we have something worth saying.
What a technical skills assessment should measure, four formats compared by evidence quality and candidate cost, and how to score one so it holds up.
Context deadline exceeded means a caller-imposed timeout elapsed before the work finished. How to find whose deadline fired in Kubernetes, and what was slow under it.
Exit code 127 means command not found; 126 means found but not executable. Both are shell verdicts on your entrypoint. Where containers hit them, verified live.
Exit code 139 means SIGSEGV: the process touched memory it did not own. Container-specific causes, a fast attribution path, and a live repro the naive test gets wrong.
ErrImagePull and ImagePullBackOff are one retry sequence, not two errors. Four root-cause families and the exact event line each leaves, verified on a live cluster.
OOMKilled is the kernel's memory verdict, on your limit or the node's. What the cgroup really counts, and the observed oom_score_adj numbers behind who dies first.
Exit code 143 means SIGTERM: your container was asked to stop and complied. When 143 is healthy, when it is a probe kill, and why Kubernetes still labels it Error.
CrashLoopBackOff is a symptom, not a cause. The last exit code and the time-to-death route you to the real failure. A diagnosis path with a runnable repro.
The definitional debate is settled and useless. What SRE and DevOps titles actually signal in job listings, which skills transfer, and how to read a JD past its title.
Exit code 137 means SIGKILL. The kernel OOM killer, a failed liveness probe, or an outside actor: a diagnosis path that tells the three apart, with a runnable repro.
Most Kubernetes interview questions reward memorised definitions. These probe operating skill: what each reveals, strong-answer signals, and the follow-ups that matter.
Most DevOps job ads are requirement soup that repels the engineers they need. How to write a JD around provable capabilities: structure, wording, lines to delete.
No CS degree? Not the blocker you think. What actually filters candidates out, the signals that substitute for credentials, and a working plan.
Most portfolio advice produces identical toy projects nobody reads. What hiring managers actually look for, five projects worth building, and how to present them.
Question lists test recall, not operating skill. What each interview stage can honestly tell you, and how to build a scenario that predicts performance.
Practise diagnosing CrashLoopBackOff, ImagePullBackOff, Pending pods, unreachable Services and OOMKilled containers through real troubleshooting paths.
Dropping the degree requirement is the easy part. The evidence ladder, the process changes, and the honest costs nobody puts in the framework posts.
Support engineers already have production instincts most career switchers lack. Map support experience onto SRE competencies, close the real gaps, make the move.
Resumes reward prestige, not skill. SkillBricks replaces the CV with proof from real environments, so the people with the skills get the interviews.
Coding-puzzle screens filter for test-takers, not engineers. Here is what they miss, why senior candidates skip them, and what to measure instead.
The CKA is not the only way to show you can run Kubernetes in production. Here is what actually signals DevOps competence, and five things you can start doing this week.
The architecture behind browser-based Kubernetes labs: xterm.js, Socket.IO, an assessment gateway, gVisor-sandboxed pods, and tmux for disconnect survival.
What we are building, why we are building it, and how the blog will work during the launch build.
How DevOps and SRE profiles actually get screened: where the six seconds go, what hiring managers care about, and how to stop optimising for signals nobody reads.