DevOps and SRE Interview Prep for OPT and H-1B Candidates: Incident Rounds, Linux, Kubernetes and On-Call

The eight SRE and DevOps round types, the scenario questions inside each one, and the two visa questions most prep guides skip, from job titles to where on-call work happens.

SRE interview questions for OPT and H-1B candidates fall into eight round types: Linux and networking troubleshooting, practical coding, infrastructure as code and CI/CD, Kubernetes, observability and SLOs, a live incident simulation, reliability-focused system design, and an on-call behavioral round. Interviewers grade reasoning under pressure, not memorization. Two visa questions sit beside the loop: whether the role's written duties, not the word "DevOps", meet the H-1B specialty-occupation test, and whether on-call or remote work happens in a place the employer's Labor Condition Application (LCA) covers.

Status as of November 1, 2026: OPT unemployment limits (90 days, 150 on STEM OPT) are in effect and unchanged; the F-1 60-day grace period still applies because a federal court has blocked the duration-of-status rule since 2026-09-14; ending the separate H-1B worker 60-day grace period is only proposed (comments close 2026-11-10).

What is asked in an SRE interview?

An SRE interview asks you to keep a production system healthy out loud. Expect a Linux or networking troubleshooting scenario, a scripting exercise, Terraform, CI/CD and Kubernetes questions, error-budget math, a live incident simulation, and system design graded on failure modes, capacity and safe rollouts rather than features.

DevOps loops tend to weight pipelines and infrastructure as code; SRE loops weight SLOs and incidents; many companies use the titles interchangeably. Still choosing? Our guide to becoming a DevOps or SRE engineer as an international student maps the skills and the F-1 to H-1B sequence.

Linux troubleshooting interview questions

The classic opener is "the server is slow, where do you start?" Interviewers want a method: for each resource (CPU, memory, disk, network), check utilization, saturation and errors, then follow whatever looks abnormal.

Narrate it with uptime (load trend), vmstat 1 (run queue, I/O wait), free -m, iostat -x 1 (disk latency), ss -s (sockets) and dmesg -T (kernel messages), saying what you expect before each command.

Scenario questions that come up again and again

• "The disk is full, but du adds up to far less." A process holds deleted files open. lsof +L1 finds it; restarting it frees the space.
• "df shows free space, but nothing can create files." Inodes are exhausted, usually by millions of small files. Check df -i.
• "Load is 40 on an 8-core box, but CPUs are idle." Linux load counts tasks in uninterruptible sleep, which usually points at storage or a hung network mount. Look for processes in state D.
• "A process vanished with no error logged." Check dmesg for the OOM killer. Exit code 137 means SIGKILL (128 + 9).

"Connection refused" versus "timed out" separates candidates quickly. Refused means the host answered with a reset: reachable, but nothing listening, or a firewall rejecting. Timed out means packets are dropped somewhere, so check security groups, routes and firewalls with mtr, curl -v and tcpdump. For DNS, dig +trace shows where resolution breaks.

Infrastructure as code and CI/CD questions

Know what Terraform state is, why it lives in a remote backend with locking, why secrets in state are a risk, and how plan exposes drift. For "someone changed a security group in the console," detect it with a plan in CI, decide whether to revert or codify, then tighten permissions so it cannot recur silently.

For "design a pipeline from commit to production," build one immutable artifact per commit, promote that same artifact through environments, run a canary against the baseline, roll back automatically when it burns error budget, and run database migrations in expand-then-contract steps so a rollback never meets an incompatible schema.

Kubernetes interview questions

Kubernetes rounds are debugging rounds. Practice these on a local cluster:

• CrashLoopBackOff: kubectl describe pod for events and the exit code, then kubectl logs --previous. Usual causes: a missing config or secret, a bad command, or a liveness probe killing a slow starter.
• Pending: the scheduler explains itself in events. Insufficient requested CPU or memory, an unmatched node selector, a taint without a toleration, or an unbound volume claim.
• OOMKilled: the container exceeded its memory limit. Explain requests versus limits and the QoS classes (Guaranteed, Burstable, BestEffort).
• Probes: readiness removes a pod from Service endpoints, liveness restarts it, startup holds both for slow boots. A liveness probe that checks a downstream database can restart your whole fleet during that database's outage.
• 502s during rolling updates: pods get traffic before they are ready or die before connections drain. Discuss a preStop delay, the termination grace period, maxUnavailable and PodDisruptionBudgets.

SRE interview questions on observability, SLOs and error budgets

Google's SRE book puts the core in one line: "The four golden signals of monitoring are latency, traffic, errors, and saturation" (Monitoring Distributed Systems). An SLI is the ratio of good events to valid events, the SLO is the target for that ratio, and an SLA is a contract with consequences. Know the budget math cold:

A burn rate of 1 spends the budget exactly over the window; a burn rate of 14.4 for one hour spends 2% of a 30-day budget (14.4 x 1 / 720). Page on fast burns, ticket slow ones, and alert on symptoms users feel, not causes such as high CPU.

How do I answer an incident postmortem question?

Tell it in order: impact in numbers, detection, mitigation, root cause and contributing factors, the action items you shipped, and what you would change, all in blameless language. Many loops also run an incident live, so practice both.

The live incident simulation

A typical prompt: "At 02:10 you are paged. Checkout errors jumped to 8% against a 99.9% SLO. Payments deployed version 2.31 at 01:55." Work in this order:

1. Acknowledge the page and declare an incident, naming who leads and who communicates.
2. Scope the impact: which users, which regions, since when, how fast the budget burns.
3. Mitigate before diagnosing. A deploy 15 minutes before the spike means roll back first.
4. Post an update with what you know, what you are doing and when the next update comes.
5. Verify recovery on the SLI itself.
6. Hand off to a blameless postmortem with owned action items.

The interviewer reveals data only when you ask, so ask, and think out loud.

The postmortem story

Google's Postmortem Culture chapter says a blameless postmortem must focus "on identifying the contributing causes of the incident without indicting any individual or team." Say "the pipeline let an unreviewed config change reach every region at once," not "a teammate pushed bad config." If English is your second language, rehearse from a six-line skeleton of your two best incidents; our guide to STAR answers for non-native English speakers shows how.

Reliability-focused system design

The structure matches any backend system design interview: clarify requirements, estimate load, sketch components, go deep. Prompts include a metrics and alerting pipeline, a deployment system with progressive rollout, failover for a stateful service, a log pipeline or a distributed cron. For each component, say how it fails, how you detect that, its blast radius and how the system degrades, then how you would monitor and safely change what you designed.

How is a DevOps interview different from a software engineer interview?

A DevOps or SRE loop usually trades some algorithm rounds for troubleshooting, platform and incident rounds, and grades design on reliability. Formats vary, so ask the recruiter.

Do DevOps roles qualify for H-1B? Title versus duties

The title is not the test; the written duties and the degree requirement are. Since the H-1B modernization rule took effect on 2025-01-17, 8 CFR 214.2(h)(4)(ii) requires a bachelor's degree or higher "in a directly related specific specialty" as the minimum for entry, and says: "Directly related means there is a logical connection between the required degree, or its equivalent, and the duties of the position." A general degree without further specialization is not enough.

O*NET lists "DevOps Engineer" and "Infrastructure Engineer" among sample titles for Software Developers (15-1252.00), and "Systems Engineer" appears there and under Network and Computer Systems Administrators (15-1244.00), alongside Network Administrator. The BLS Occupational Outlook Handbook (last modified 2026-08-27) says administrators typically need a bachelor's degree in a field related to computer or information science, though some employers may require only a postsecondary certificate or an associate's degree. Our reading, not a USCIS statistic: duties that look like routine administration invite more questions about whether a specific degree is the normal minimum than duties built around designing, automating and writing software. The occupation code also sets which prevailing-wage figures apply.

The job description and offer, not your interview answers, go into the petition, and they must describe the job you will actually do. If the written duties differ from what the team described, raise it with HR before you accept; never ask for a title that doesn't match the work. Our SRE H-1B sponsorship guide covers which employers sponsor the role.

On-call, remote work and the LCA worksite

DOL ties the LCA to a place. Under 20 CFR 655.715, "Place of employment means the worksite or physical location where the work actually is performed by the H-1B, H-1B1, or E-3 nonimmigrant." DOL's August 24, 2023 OFLC FAQ says a move within the same area of intended employment generally needs no new LCA, though notice must still be provided, and that telework offered to H-1B workers must be offered to similarly employed US workers. Outside that area, the short-term placement rule (20 CFR 655.735) allows up to 30 workdays a year (60 in narrow cases), with the employer paying lodging, travel and meals. At the limit, the employer must file an LCA for that area or end the placement, and the FAQ says a new LCA for worksites outside the original area means an amended or new H-1B petition.

Worksite scenarios for on-call and remote infrastructure roles, as of 2026-11:

Our post on remote work rules for OPT and H-1B goes further. In the interview, "Is on-call expected from home?" is simply a normal engineering question.

Your interview loop runs against a visa clock

Postings are thin: Indeed Hiring Lab's October 22, 2026 snapshot put its software development postings index at about 78-79 as of October 16, the lowest of the five sectors it charted.

Worked example. Your post-completion OPT started 2026-09-01 and no employer is reported. November 1 is unemployment day 62 and November 29 is day 90. A recruiter screen on 11-04, a technical screen on 11-10 and an onsite on 11-18 put an offer near Thanksgiving (11-26) with no margin. The SEVIS unemployment counter looks back to your actual OPT start date and counts "each day without employer information," so a signed offer does not stop the count until the employer is reported. If the start date is weeks away, ask your DSO how those days will be counted.

Rules that touch an infrastructure job search, as of 2026-11-01:

If a hiring manager raises the $100,000 payment, USCIS's H-1B page (updated 2026-09-21) says it does not apply to a change of status, amendment or extension USCIS grants for someone inside the US; it would apply to a consular-notification petition or a denied change of status.

Before you accept an infrastructure offer:

1. Confirm your unemployment-day count with your DSO, who can see the SEVIS counter.
2. Search the LCA disclosure files on DOL's OFLC performance page for the employer's past filings under similar titles; our guide to searching LCA filings walks through it.
3. Check the USCIS H-1B Employer Data Hub for the employer's petition history.
4. Ask HR in writing for the worksite address, remote policy and on-call expectations, which shape what the LCA (Form ETA-9035) must cover.
5. Ask whether you would be employed directly or placed at a client site.
6. Tell your DSO about any planned STEM OPT worksite or duty change; a material change needs a modified, signed Form I-983.
7. Bring classification and worksite questions to the employer's immigration attorney, or your own.

Common mistakes

• Prepping only LeetCode. Most SRE loops spend more time on troubleshooting, platform and incident rounds.
• Jumping to root cause in the simulation. Mitigate first; an explained rollback beats a perfect diagnosis mid-outage.
• Reciting commands without hypotheses. Say what each command should show and what you will do next.
• Treating "DevOps" as a visa-safe title. Duties and the degree requirement decide specialty occupation.
• Accepting "on-call from anywhere" unasked. Where you work matters for the LCA; ask the employer's attorney.
• Letting one loop drain your OPT clock. Run parallel processes and keep your DSO informed.

Frequently asked questions

What is asked in an SRE interview?

Most SRE loops combine Linux or networking troubleshooting, a practical scripting round, infrastructure as code, CI/CD and Kubernetes questions, SLO and error-budget reasoning, a live incident simulation, reliability-focused system design and an on-call behavioral round. Interviewers grade how you narrow down a problem and protect users, not how many commands you remember. Ask the recruiter which rounds your loop includes.

How is a DevOps interview different from a software engineer interview?

A software engineer loop leans on algorithm coding and feature-oriented system design. A DevOps or SRE loop usually has less pure algorithm work and adds troubleshooting, platform, incident and on-call rounds, with system design graded on failure modes, capacity and safe rollouts. Some SRE tracks keep full algorithm rounds, so confirm the format.

How do I answer an incident postmortem question?

Tell it in order, starting with impact in numbers, then detection, mitigation, root cause and contributing factors, and the action items you shipped with owners. Use blameless language that describes what the system allowed rather than who made a mistake, and end with what you would do differently.

Do DevOps roles qualify for H-1B?

It depends on the written duties and degree requirement, not the title. Under 8 CFR 214.2(h)(4)(ii), as amended effective 2025-01-17, a specialty occupation must require a bachelor's degree or higher in a directly related specific specialty, and a general degree without further specialization is not enough. Ask the employer's immigration attorney how the role is classified, and never ask for a title or duties that do not match the real job.

What Linux troubleshooting questions come up?

Expect a slow server, a disk that reads full while du disagrees, free space but no new files, a high load average with idle CPUs, a process killed without a log line, and a host you cannot reach. Interviewers want a method, such as checking utilization, saturation and errors for each resource, plus the commands that prove each step.

Sources

• 8 CFR 214.2, 20 CFR 655.715 and 20 CFR 655.735 (Cornell LII)
• DOL OFLC H-1B, H-1B1 and E-3 FAQs, August 24, 2023 and OFLC performance data
• USCIS Form I-129, H-1B registration and Employer Data Hub
• SEVIS Help Hub, unemployment counter
• Federal Register, proposed end of the 60-day grace period; D/S case docket (D. Mass.); Unified Agenda, RIN 1615-AD00
• USCIS H-1B specialty occupations page (who the $100,000 payment would cover)
• BLS Occupational Outlook Handbook; O*NET 15-1252.00 and 15-1244.00
• Indeed Hiring Lab, US labor market snapshot, October 2026
• Google SRE book: Monitoring Distributed Systems and Postmortem Culture

How we checked this: on November 1, 2026 we read the regulations, the DOL FAQ, and the USCIS, SEVIS Help Hub, Federal Register, BLS, O*NET and Indeed Hiring Lab pages above, and took every policy status and date from those sources and our dated research brief. The interview questions reflect common loop formats, not any one employer's process.

This is general information, not legal or tax advice; your own I-20, I-94 and notices, and your DSO or immigration attorney, govern your situation.

If you want support on an infrastructure job search while your OPT clock runs, F1Jobs works with international candidates targeting employers that sponsor.