Platform and infrastructure questions for DevOps, SRE, and any backend role expected to own its service in production.
37 published across 8 topics.
Processes and signals, file descriptors, DNS and TLS, and debugging connectivity from first principles.
Work up the layers and let each command eliminate a class of cause: resolve the name, check the route, see whether anything is listening, then decide from the failure mode whether packets are being refused, silently dropped, or answered by something that never replies.
df and du disagree because df reads filesystem block allocation while du walks visible directory entries. Deleted-but-open files, inode exhaustion, hidden files under mounts and reserved blocks are the usual causes. Use this LINUX answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects filesystems to the point an interviewer is testing.
Assess the candidate's capability to design sophisticated DVFS algorithms and power management strategies for severely energy-constrained IoT devices operating on harvested energy. Use this hardware answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects power management to the point an interviewer is testing.
Separate the layers that can fail together: routing reachability, name resolution, identity and physical access. Then build an access path that shares none of them, keep controls failing static rather than closed, and rehearse the emergency path often enough that it is known to work.
Evaluate the integration of kernel-bypass techniques to handle millions of simultaneous connections and minimise CPU overhead in MMO architectures.
Namespaces and cgroups, image layering, multi-stage builds, and container security basics.
Roughly 200 near-full copies, because delta compression works on text and a relinked binary differs throughout, so each version costs almost its own size. Deleting the file changes nothing: history is content-addressed and every past commit still reaches the blob. The design fix is that build outputs belong in a registry that can expire them, because version control is built never to forget.
A tag is a mutable pointer in the registry, so re-pushing it makes the same reference resolve to different content over time. Nodes that already cached the old digest keep serving it, so the fix is to deploy by digest and let the tag be a human label rather than the deployed identity.
Container CPU and memory limits fail differently: CPU limits throttle work, while memory limits can trigger OOMKilled restarts. Debug by comparing requests, limits, actual usage and runtime-specific memory settings. Use this containers answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects cgroups to the point an interviewer is testing.
A container image is layered filesystem content plus runtime config, while a container is a host process isolated by namespaces and cgroups. A virtual machine boots its own guest kernel, so the isolation boundary and startup model are different.
Scheduling, controllers and reconciliation, services and ingress, probes, and resource management.
CPU is compressible, so the kubelet throttles it rather than evicting for it. Eviction comes from memory, ephemeral storage or inodes - and which pods die is decided by QoS class, not by who caused the pressure. Check node conditions and eviction thresholds first, then fix the requests and limits so critical workloads are Guaranteed rather than BestEffort.
Kubernetes CrashLoopBackOff debugging starts with the previous container logs, exit code and pod events. Then separate application crashes from liveness probe kills, bad configuration, missing dependencies and OOMKilled restarts.
Removing a pod from its EndpointSlice and sending it SIGTERM happen concurrently, and the data-plane update takes time to propagate, so requests keep arriving at a pod that has already begun shutting down. The fix is a preStop delay plus an application that finishes in-flight work before exiting.
eBPF-based Kubernetes network policy can enforce hard multi-tenancy with faster identity-aware checks than long iptables chains, but it demands verifier-safe rollouts, fail-closed controls, and strong policy observability.
A node image change is a release of the layer underneath Kubernetes, so the control plane cannot stage it, roll it back or reliably report on it. Treat node pools as rings, replace nodes rather than patching them in place, and get the health signal from outside the cluster. It also connects node lifecycle to the point an interviewer is testing.
Pipeline design, artefact promotion, deployment strategies, and safe rollback mechanics.
Build one immutable artefact, promote that same artefact through environments behind ordered quality gates, and deploy so traffic can be shifted back instantly. Rollback only stays possible if schema changes are backward-compatible expand-contract steps decoupled from the code deploy.
Evaluate the candidate's expertise in profiling build systems, configuring distributed caching, and optimising CI/CD pipelines for large-scale mobile projects.
An analysis of the challenges in designing a cost-effective, scalable GitHub Actions runner environment using spot instances, focusing on state management, instance termination handling, and performance tuning.
Treat any payload the fleet loads as a release, because it changes behaviour in production exactly as a binary does. It needs schema and replay validation, a canary ring with its own health comparison, and a rollback that survives the change breaking whatever would apply the rollback.
Replace the stored key with OIDC federation so each job exchanges a short-lived signed token for temporary credentials, then make the trust policy specific to a repository, branch or environment. Also narrow the pipeline's own token, pin third-party steps to commit SHAs, and attest what you built.
Terraform state and drift, module design, and managing environments without duplication.
When a Terraform plan wants to destroy production, stop and identify whether the diff comes from state drift, a provider change, a renamed resource or real configuration intent. Do not approve until the blast radius is understood. It also connects IAC to the point an interviewer is testing.
A deep dive into managing Infrastructure as Code at scale, focusing on the architectural patterns required to handle state drift, enforce compliance, and maintain state integrity.
State is the mapping from a config address to a real resource ID, so it is the only thing that tells Terraform whether to create or update. Drift, imports, refactors and concurrent applies are all consequences of that mapping being separate from both the code and the infrastructure.
Compute, networking, identity, and managed-service selection across Azure, AWS, and GCP.
During the event your only levers are the ones already built: fail over, serve degraded from cache, or shed the dependent feature. Beforehand the work is finding your transitive dependencies on foundational services, making the running system statically stable, and choosing which features are allowed to fail alone.
Managed service vs self-hosted decisions compare operational load, control, lock-in, availability and real cloud cost. The hidden costs are usually egress, cross-zone traffic, idle capacity and the team needed to run the component well. It also connects managed services to the point an interviewer is testing.
An examination of strategic foresight, the realities of vendor lock-in, and the technical necessity of abstracting proprietary dependencies during a large-scale migration. Use this engineering leadership answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects vendor management to the point an interviewer is testing.
Zombie cloud infrastructure should be detected with billing, utilization, ownership, and access signals, then quarantined before deletion so FinOps automation saves cost without destroying critical standby resources. Use this FINOPS answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects cloud infrastructure to the point an interviewer is testing.
Assessing the architectural transition from legacy perimeter models to Zero Trust, focusing on continuous authentication, context-aware access, and the elimination of static credentials. Use this security answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects cloud security to the point an interviewer is testing.
SLIs, SLOs and error budgets, incident command, on-call design, and blameless postmortems.
A single threshold on a single window cannot separate a real outage from a brief spike, so it either pages on noise or misses slow burns. Replace it with multi-window multi-burn-rate alerting: a fast rule that pages and a slow rule that files a ticket, each gated by a short window so a spike that already stopped does not keep paging.
Recovery time is set by how much state diverged, not by how long the fault lasted, so seconds of dual writes can take a day to reconcile. The prevention is a single writer enforced by quorum and fencing; the mitigation is designing the write path so repair is mechanical rather than forensic.
Analyse the leader's approach to blameless post-mortems, identifying root causes over symptoms, and driving accountability for remediation. Use this engineering leadership answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects incident management to the point an interviewer is testing.
Turn it into an experiment: a steady state you can measure in under a minute, a blast radius you chose, an abort that is one command with a named person who can call it, and written approval from whoever owns the revenue. Tuesday afternoon is the right time because everyone who understands the system is awake, and the announcement goes to the channel your own alerts land in.
Evaluate the candidate's crisis management and technical recovery skills in a high-stakes scenario involving a total compromise of Kubernetes administrative control. Use this security answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects incident response to the point an interviewer is testing.
An SLO defines the reliability users should experience, and the error budget is the allowed gap between that target and perfection. Use burn rate to decide when reliability work must outrank feature delivery.
Declare it, take the incident commander role explicitly, and mitigate before you diagnose - roll back or shed load first, understand later. Keep one channel as the record, publish updates on a fixed cadence, hand over rather than push through, and let the review produce owned actions instead of narrative.
Situations to work through: a failing rollout, an exhausted node, a leaked credential, a 3am pager.
Stabilise first and diagnose second, which usually means reverting the change before you understand it. Take one artefact of evidence on the way out, time-box the decision aloud, communicate impact on a fixed cadence, and treat the incident as open until the damage the deploy caused is repaired.
You cannot answer it from configuration, only by performing a restore into an isolated environment and timing it. Establish the recovery targets first, then test each mechanism separately, because backup jobs that report success and replicas that are not backups are the two ways this claim is usually wrong.
Evaluating the engineering leader's capability to drive complex, long-running programs across the finish line when initial momentum has faded and organisational fatigue sets in. Use this engineering leadership answer to show the decision, trade-off, and evidence rather than a memorised definition. It also connects program delivery to the point an interviewer is testing.