Oracle DevOps Engineer Interview Questions

“We had to patch the operating system on a few hundred database hosts within a fixed time. I first applied it to test hosts and ran the database test suite. Then I planned waves: a handful of low-risk hosts, then larger groups, with a stop rule if any host failed health checks after the patch. Each database was failed over to its standby before its host was patched, and the owning team got a notice with their window. I kept a shared page with progress updated after each wave. One wave found a driver problem, we paused, fixed it and continued. The result was your result.”

“During a routine drill I tried to restore a customer database from the previous night's backup, and the restore failed because one archived log file was missing. The backup job had reported success, since it checked only the data files. That meant we could not have recovered to a point in time for weeks. I immediately took a fresh full backup and verified it by restoring it to a separate host. Then I changed the job to verify that the full chain of logs is present and to fail loudly if not, and I set up an automatic weekly restore of a random database. The result was your result.”

“Point-in-time recovery needs two things. Regular backups of the data files, say a full one weekly and incrementals daily, give the starting points. Continuous archiving of the transaction logs gives every minute in between, because a restore replays the logs from the backup up to the requested time. Both must be kept for at least the month, in storage in another location, and protected from deletion. I would monitor the age of the last archived log, because a gap breaks recovery. And I would restore to a random time on a schedule, to prove it works and to know how long a recovery takes.”

“The first step is detection: hardware signals from the host raise an alert, and the scheduler marks the host so that no new machines are placed there. Then each customer machine is moved. Where live migration is supported, the machine moves to a healthy host with only a short pause. Where it is not, the customer is told and given a window, or the machine is restarted elsewhere at a planned time. Customer disks should live on replicated storage outside the host, so a dying disk does not take data with it. The whole flow should be automatic, and the region must keep spare hosts ready for it.”

“First I would buy time without risk, for example by removing archived logs that are already safely backed up, or old trace files. Then I would extend the storage while the database runs, which most platforms allow, and confirm the database sees the new space. Next I would find what is growing, since a table that doubled in a week is a sign of an application change or a job that stopped cleaning up. For the future, I would track growth per day and alert on the number of days left, because a fixed threshold gives the same warning for a slow disk and a fast one.”

“I would start from what users feel: how long queries take, and how many fail. Then the signals that predict trouble before users feel it: transactions open for a long time, sessions waiting on locks, the lag of the standby, the age of the last archived log, free space, and connection counts near their limit. Each has a normal shape by time of day, so I would alert on departures from that, not on one fixed number. Only a few things page a person at night, like replication stopping or space nearly gone. The rest go to a daily review.”

“First I would confirm which side is primary now and that it is healthy and has all the data. Then I would repoint the application, ideally by changing one connection setting rather than touching servers, and watch the errors stop. The real work comes next. The application should never hold the primary's address directly. It should connect through a name or a service that moves with the primary, and its connection pool should drop dead connections. I would also ask why nobody was alerted at night. Then I would schedule regular failover drills that include the application, not only the database.”

“When a company puts its database in your cloud, it is trusting you with the thing it cannot replace. I take that seriously and I find it motivating. The problems here are the ones I like most: backups that must restore, failovers that must work at three in the morning, and fleets of hosts that must be patched without anyone noticing. In my current job I look after a few dozen databases, and the drills and the forecasting I set up were the work I was proudest of. I want to learn how the same care is applied to many thousands.”