Database administrator screening questions: engines, backups, restore drills and on-call
On this page
- Production, development, cloud or reliability DBA: place the candidate
- Engines, versions and the size of the estate
- Backups and restores: the scenario that separates DBAs
- High availability, replication, upgrades and migrations
- Performance and security questions with strong answers
- Certifications: what Microsoft's shows, and checking Oracle's
- How DBA resumes overstate the work
- On-call, logistics, knockout checklist and scorecard
- Questions people ask
Good database administrator screening questions establish four things quickly: which engines the candidate has run in production, how large the estate was, whether they have actually restored from backup under pressure, and what on-call looked like. Ask for dates, sizes and durations. A DBA who has done the work answers with numbers; one who has watched it answers with product names.
This page is for recruiters and hiring managers screening production, development and cloud DBAs. It is not a data engineering screen: if the role moves and models data between systems, use the data engineer phone screen questions instead. If the role is mostly servers, directory and patching with some databases on the side, the systems administrator screening questions fit better.
Production, development, cloud or reliability DBA: place the candidate
"DBA" covers several jobs. Ask the candidate to split a normal month between them, in percentages, before anything else. It tells you which of the later sections to push hardest.
| Flavor | What they spend the day on | Question that confirms it |
|---|---|---|
| Production (operational) DBA | Backups, restores, patching, replication, monitoring, access requests, on-call | "What woke you up last time you were paged, and what did you do?" |
| Development (application) DBA | Schema changes, query and index design, reviewing developers' SQL, release scripts | "Tell me about a schema change you reviewed and pushed back on." |
| Cloud or managed-service DBA | Amazon RDS or Aurora, Azure SQL, Cloud SQL: sizing, parameter groups, failover settings, cost | "What did you still have to manage yourself on the managed service?" |
| Database reliability engineer | Automation, infrastructure as code for databases, fleet-wide tooling, capacity | "What did you automate that a DBA used to do by hand?" |
A mismatch here is the most common reason DBA submittals fail. A strong development DBA placed in a role that is 70% on-call operations will struggle, and the reverse is just as true.
Engines, versions and the size of the estate
These questions set the level. Write the numbers down exactly as the candidate gives them; they are what the hiring manager will ask you about first.
| Question | What a strong answer sounds like | Red flags |
|---|---|---|
| Which engines have you run in production, and which was primary? | One primary engine named with versions: "Mostly SQL Server 2019 and 2022, some PostgreSQL 15 on RDS." | A list of eight engines with no primary, or versions they cannot name. |
| How many instances and databases were you responsible for? | A count and a split: "About 60 instances, 400 databases, 12 of them business-critical." | "Lots" or "the whole company's," with no number. |
| How big was the largest database, and how fast did it grow? | A size and a growth rate: "Four terabytes, growing about 100 GB a month." | Cannot give a size for the database they call their main one. |
| Were you the only DBA, or one of a team? | A team size and their place in it, including who covered nights. | Says "we" throughout and cannot say what they owned personally. |
| What did your team own, and what did others own (storage, OS, network)? | A clear boundary: "We owned the instance and up; infrastructure owned VMs and storage." | No idea where their responsibility stopped. |
The engine vocabulary helps you check fluency without testing depth. A SQL Server DBA will mention Always On availability groups, log shipping, recovery models and Query Store. An Oracle DBA talks about RMAN, Data Guard, RAC and archive log mode. A PostgreSQL DBA mentions WAL archiving, streaming replication, pg_basebackup or pgBackRest, and often Patroni. A MySQL DBA talks about the binary log, replicas and tools such as Percona XtraBackup. Hearing the right words does not prove skill, but hearing the wrong engine's words for the engine they claim is a real flag.
Backups and restores: the scenario that separates DBAs
Every DBA says they "managed backups." Far fewer have restored one under pressure. Start with the plain question, then read the scenario.
| Question | What a strong answer sounds like | Red flags |
|---|---|---|
| When did you last restore a production backup, and why? | A date, a reason and a duration: "In June, a bad release corrupted an orders table. Restored 800 GB to a side server in about 90 minutes." | "We never needed to," or only restores of test databases. |
| How often did you test restores, and how? | A schedule and a method: monthly restore of each critical database to a test server, with an integrity check and a record of the time taken. | Backups are "monitored" but never restored on purpose. |
| What were your recovery point and recovery time objectives? | Numbers per tier: "15 minutes of data loss, four hours to recover, for tier-one systems." | Does not know the terms, or says "zero" for both without explaining how. |
| Where were the backups stored, and could an attacker with admin rights delete them? | Off-server copies, a separate account or immutable storage, and a clear view of who could delete them. | Backups on the same server or share as the data. |
Read this to the candidate
"At 2:14 this afternoon a developer ran a DELETE without a WHERE clause on a production customer table. It's now 2:40, and the application is still taking orders. Walk me through what you do."
Tick each item they mention:
- Find out the scope first: which table, how many rows, what else depends on it.
- Check that the backup chain covers 2:14 (full backup plus log or WAL backups, or point-in-time recovery on the managed service).
- Restore to a separate database or instance, to a point just before 2:14, not over production.
- Recognize that orders taken since 2:14 must be kept, so the fix is copying the missing rows back, not rolling the whole database back.
- Tell the application owner and the incident channel what is happening and how long it will take.
- Verify the result with the business owner before closing.
- Afterwards: ask why a developer could run that statement in production and change the access.
Scoring: 0–2 items is a junior answer. 3–4 is a working DBA. 5 or more, in roughly this order, is someone who has done it.
The item that separates candidates most is the fourth. A DBA who proposes restoring the whole database over production to 2:13 has just thrown away 26 minutes of orders. You do not need to know the commands to hear that.
High availability, replication, upgrades and migrations
| Question | What a strong answer sounds like | Red flags |
|---|---|---|
| How was high availability set up for your most important database? | The mechanism by name (availability group, Data Guard, streaming replication with Patroni, Multi-AZ) and whether failover was automatic. | "The cloud handles that," with no idea what happens during a failover. |
| Tell me about a failover, planned or not. | What triggered it, how long the application was affected, and one thing that went wrong. | Has never seen one, in a role that lists high availability as a must-have. |
| What was the biggest version upgrade you ran? | From which version to which, how they tested it, the rollback plan and the actual downtime. | "The vendor did it," or an in-place upgrade with no rollback plan. |
| Have you migrated a database between platforms or into the cloud? | Source and target, size, method (backup and restore, replication, a migration service) and the cutover window. | Attended the migration project but cannot describe the cutover. |
| How often did you patch, and who decided when? | A cadence tied to the vendor's releases, a test environment first, and an agreed window. | Patches applied "when there is time," or never on production. |
Replication is also where cloud cost questions start. If the role is heavily on a managed service, the cloud engineer screening questions cover accounts, networking and spend in more depth than this page does.
Performance and security questions with strong answers
A slow-query story anyone can score
Ask: "Tell me about the worst performance problem you fixed. What was slow, how did you find the cause, and what changed?" A strong answer has a symptom with a number, a method, a cause and a measured result:
"The nightly billing job went from 40 minutes to six hours. Query Store showed one statement had switched to a new plan after statistics updated. We fixed the index it needed, and the job went back to 35 minutes."
Listen for the tool that found the cause: execution plans, Query Store, wait statistics, Oracle AWR reports, PostgreSQL's pg_stat_statements and EXPLAIN ANALYZE, or the MySQL slow query log. Weak answers jump straight to "we added more memory" or "we added an index" without saying how they knew that was the problem, and without a before-and-after number.
Access, encryption and audit
| Question | What a strong answer sounds like | Red flags |
|---|---|---|
| How did developers and applications get access to production? | Named service accounts or roles with only the permissions needed, requests approved and reviewed, no shared admin logins. | Everyone uses the same admin account "because it's easier." |
| How was sensitive data protected? | Encryption at rest (for example Transparent Data Encryption), encrypted connections, masking in non-production copies. | Production copies with real customer data handed to developers without masking. |
| Have you supported an audit (SOX, PCI DSS, HIPAA, SOC 2)? | Which one, what evidence they produced (access reviews, change records, backup logs). | Claims audit experience but cannot name a single piece of evidence they supplied. |
Certifications: what Microsoft's shows, and checking Oracle's
Certifications show that someone studied an engine. They say little about restores at 3 a.m. Treat them as a tiebreaker unless the client lists one as a requirement, and then check status, not just the name on the resume.
- Microsoft. The Microsoft Certified: Azure Database Administrator Associate is earned with exam DP-300 and covers Azure SQL Database, Azure SQL Managed Instance, SQL Server on Azure virtual machines and SQL Server on-premises. Microsoft lists a renewal frequency of 12 months, and renewal is a free online assessment on Microsoft Learn (as of September 2026). A candidate can send a transcript share link from their Learn profile, which shows what they hold.
- Oracle. Oracle University offers associate and professional database administration credentials. We could not load Oracle's certification pages to confirm current exam numbers and versions for this article, so check the specific credential on Oracle University's own site before you tell a client which one to require, and ask the candidate for Oracle's verification link rather than a screenshot.
For PostgreSQL and MySQL roles, open-source experience usually matters more than any credential. Ask which version they ran and what extensions or tools they depended on.
How DBA resumes overstate the work
| Resume says | Often means | Question that checks it |
|---|---|---|
| "Managed backup and recovery" | Scheduled backup jobs and watched the alerts | "When did you last restore one, and how long did it take?" |
| "Performance tuning" | Ran an index advisor and applied its suggestions | "How did you find the cause of your worst slowdown?" |
| "High availability and disaster recovery" | The availability group existed; someone else built it | "Who configured it, and have you tested failover?" |
| "Oracle, SQL Server, PostgreSQL, MySQL, MongoDB" | Deep in one, read-only access to the rest | "Which would you be comfortable being on call for alone?" |
| "Led the cloud migration" | Moved a few databases in a program others planned | "How many did you move yourself, and what was the cutover window?" |
| "24x7 production support" | Was on an escalation list that rarely called | "How often were you paged in a typical on-call week?" |
None of these are lies. They are the normal gap between a team's work and one person's part in it, and the questions close it in a minute each.
On-call, logistics, knockout checklist and scorecard
DBA roles fail late on on-call and maintenance windows more often than on skills. Ask directly.
| Question | What a strong answer sounds like | Red flags |
|---|---|---|
| What did your on-call rotation look like, and how often were you paged? | Rotation length, team size and a typical number: "One week in four, two or three pages a week, most resolved in under an hour." | Refuses any on-call for a role that requires it. |
| This role patches on Saturday nights once a month. Does that work? | A direct yes, or a clear condition. | A vague yes that becomes a no at offer stage. |
| This role involves access to regulated data and a background check. Any concerns? | A direct answer. | Surprise or discomfort, where the client is in finance or healthcare. |
| Are you legally authorized to work in the US, and will you now or in the future need visa sponsorship? | A direct answer to both. | None. Ask everyone the same two questions. |
Must-ask on every DBA screen
- Flavor split: production, development, cloud or reliability, in percentages.
- Primary engine and version, instance and database counts, largest database size.
- Last production restore: date, reason, size, duration.
- The DELETE scenario, scored against the checklist.
- High availability mechanism and one failover story.
- One performance fix with a before-and-after number.
- Certifications with status and a verification link, if the client requires them.
- On-call, maintenance windows, background check, salary, notice, other processes, work authorization.
Knock out, or flag to the client before submitting, if:
- The role is production-heavy and the candidate has never restored a production backup.
- The client's primary engine is a must-have and the candidate has only read-only or course-level exposure to it.
- The candidate scores two or fewer on the restore scenario for a mid-level or senior role.
- The candidate will not do on-call or weekend windows the role requires.
| Area | 1 | 2 | 3 | 4 |
|---|---|---|---|---|
| Estate and ownership | Assisted on a few databases | Owned some instances | Owned a critical estate | Set standards across the estate |
| Backup and recovery | 0–2 scenario items | 3–4 items | 5+ items and a real restore | All items, tested restores on a schedule |
| Availability and change | Watched others | Ran patches | Ran upgrades and a failover | Designed HA and led a migration |
| Performance and security | Generic answers | One fix, no number | Fix with method and number | Several, plus access and audit work |
| Logistics fit | Deal-breaker | Two open questions | One open question | All aligned |
Write down the candidate's exact figures: "restored 800 GB to a side server in about 90 minutes" tells a data platform manager more than "strong backup and recovery skills." Interview Signal attaches quotes like that to each score from the call, so the submittal can use the candidate's own numbers.
Questions people ask
What is the single most useful question for a DBA screen?
Ask when they last restored a production backup, what it was for, and how long it took. A DBA who runs restores answers with a date, a size and a duration. One who only schedules backups talks about the backup job.
Can a non-technical recruiter screen a database administrator?
Yes, if you score what the candidate says against a checklist rather than judging the answer yourself. The restore scenario and the estate-size questions on this page can be scored by anyone; leave deep tuning and design questions to the technical interview.
Does a DBA need a certification?
Rarely as a hard requirement. A current Microsoft DP-300 certification or an Oracle credential shows study of a specific engine, but restore history, on-call experience and the size of the estate tell you far more about whether someone can run production.
How is a DBA screen different from a data engineer screen?
A DBA keeps databases available, recoverable, fast and secure: backups, replication, patching, tuning and access. A data engineer moves and models data between systems. The overlap is SQL; the day-to-day risk is different, so the questions are too.