SRE & Production Engineering Interview Prep

Observability interview questions cover the three pillars (metrics, logs, traces) plus the engineering culture around them — Micrometer + Prometheus + Grafana, structured logging (JSON, MDC, correlation IDs), distributed tracing (OpenTelemetry, Zipkin, Jaeger), SLI/SLO/SLA definitions, error budgets, the four golden signals, alerting design (avoiding alert fatigue), and the pragmatic question every senior interview asks: "what would you instrument first in a brand-new service?". For incident response and on-call see the SRE page.

Three pillars + one outcome:

• Metrics — counter, gauge, histogram (Micrometer + Prometheus).
• Logs — structured JSON + correlation ID; ship to ELK / Loki / Splunk.
• Traces — OpenTelemetry, Jaeger / Zipkin / Tempo.
• Outcome = SLI / SLO / error budget — that's the language ops talks.

Four golden signals: latency, traffic, errors, saturation.

The interview surface for this module breaks into these focus areas:

• Alerting — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Apm tools — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Comparisons — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Distributed log aggregation — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Distributed tracing — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Health checks and probes — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Metrics and micrometer — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Prometheus and grafana — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Structured logging — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Scenario-based — applied questions tying multiple topics together. Drill these last, after the fundamentals are sharp.

• Cardinality kills metrics — never put userId in a label.
• MDC propagates correlation IDs through Logger; lose it on async hops.
• Sampling is a trade-off — head-based easy, tail-based smarter.
• Alert fatigue — page on SLO burn-rate, not raw thresholds.
• Health checks ≠ readiness — liveness restarts, readiness gates traffic.

"I'd instrument the four golden signals with Micrometer → Prometheus, structured logs with correlation IDs in MDC, distributed tracing via OpenTelemetry, and define SLOs that drive burn-rate alerts — paging humans only on user-visible regressions."

SRE and production-operations interview questions cover the role that owns reliability for every modern Java backend — incident response and runbooks, postmortems and the blameless culture, on-call rotation design, error budgets and SLO-driven engineering, capacity planning, chaos engineering, the production-debug toolkit (jstack, jmap, async-profiler, JFR, eBPF), and the questions interviewers ask to test calm under pressure ("P0 fires — walk me through the first 15 minutes"). Pair this with the Observability page for the full production-engineering surface.

Production engineering revolves around incident → postmortem → SLO.

• Incident — declare early; severity guides response, not ego.
• On-call — runbooks per alert, paired rotation, fair distribution.
• Postmortem — blameless, timeline + 5-whys, action items with owners.
• Toolkit — jstack, jmap, async-profiler, JFR, eBPF.
• Capacity planning — model peak + headroom + chaos drills.

The interview surface for this module breaks into these focus areas:

• Capacity planning — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Chaos engineering — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Debugging production — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Incident management — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Incident response — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Memory leaks — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• On call — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Performance troubleshooting — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Postmortems — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Root cause analysis — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Runbooks — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Sre practices — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Thread and heap dumps — drill the named patterns, the failure modes, and the recommended Java / framework idiom.
• Scenario-based — applied questions tying multiple topics together. Drill these last, after the fundamentals are sharp.

• MTTR > MTBF for modern systems — fast recovery beats rare failure.
• Error budget — depleted = freeze risky deploys; pure SRE doctrine.
• Memory leaks in JVM — heap dump + diff retained sets; LinkedHashMap cache without bounds is the classic.
• Thread dumps — look for BLOCKED clusters, that's the contention.
• Chaos engineering — start small, gameday quarterly.

"For a P1 I'd declare, mitigate first (rollback / feature flag), preserve evidence, run a blameless postmortem within 48 h with concrete action items, track them to closure, and feed the lesson back into runbooks + alerts."