Database Administrator Interview Questions for AI Training Work

I use role-based access control scoped to what each application or user actually needs, rather than granting broad access for convenience, and I audit access permissions periodically to catch permissions that have accumulated beyond what's currently needed.

I check for index fragmentation, growing table sizes without corresponding index or partition adjustments, and outdated statistics, since gradual degradation is often about the database's data profile changing while its configuration stayed static, rather than a single specific event causing a sudden change.

I rely on proper transaction isolation levels and constraints enforced at the database level, rather than trusting each application to enforce consistency independently, since relying on application-level discipline alone leaves the door open for integrity issues if any one application has a bug.

I look at whether the inefficiency comes from how the database itself is configured or indexed versus how the application is querying it, like a query pulling far more data than it actually needs. Some issues genuinely need to be fixed at the application level rather than compensated for indefinitely at the database level.

I test actual restore procedures periodically rather than just confirming backups complete successfully, since a backup that completes without error can still fail to restore correctly, and I make sure recovery time objectives are realistic given the actual database size and restore process.

I'd scope the permissions down to what's actually required after confirming with the application team what functionality depends on current access, rather than leaving overly broad access in place because revoking it feels risky. I'd communicate the change clearly and test it before fully cutting off unused access.

I'd check the query execution plan first to identify the immediate bottleneck, applying a quick fix like an index if that's the clear cause, to restore functionality, then investigate more thoroughly afterward whether the underlying schema or query needs a more durable fix given the new data volume.

I'd start with an audit to understand actual current access and usage patterns, then tighten permissions incrementally starting with the clearest unnecessary access, testing carefully at each step rather than making sweeping changes all at once that risk breaking dependent applications unexpectedly.

I noticed a service account had permissions well beyond what its actual function required, left over from an earlier configuration that had never been revisited. Flagging and correcting it proactively closed a real exposure that hadn't been exploited yet but represented unnecessary risk if that account were ever compromised.

Adding an index that significantly sped up a slow reporting query ended up slowing down write performance on that table more than expected, since the index needed to be maintained on every write. I adjusted the approach to use a more targeted index and reworked the write pattern to reduce the impact rather than removing the fix and losing the read performance improvement.