Target DevOps Engineer Interview Questions

“During a holiday freeze we found that a log setting was slowly filling disks on our order servers. Waiting meant a likely outage within days. Releasing meant breaking our own rule. I wrote a short note with the evidence, the smallest possible change, and how to undo it, and took it to the person who owned the freeze. We agreed to release to one server first, watch for an hour, and then continue. The result was your result. I learned that a freeze is a rule about risk, so the case must be made in terms of risk.”

“Before every peak, my team spent days checking by hand that each location's servers had enough disk, memory and the right software version. It was slow and people missed things. I wrote a script that ran the same checks and produced one report with only the failures. For the first season we ran it beside the manual check to prove that it agreed. Then I added it to a schedule, so it ran weekly and not only before peaks. Problems were found earlier, and we saw your result. The team trusted it because they had watched it work.”

“The store must keep selling and keep packing orders it already has. So registers, price lookup and the picking tools run on servers inside the store, with a local copy of prices and orders. While the link is down, sales and pick updates are written to a local queue. New online orders cannot arrive, so the central system should notice the silence and send them elsewhere. When the link returns, the store sends its queue in order, and each record carries an id so nothing is applied twice. I would test this by cutting the link on purpose in a test store.”

“Automatic scaling reacts in minutes, and this spike arrives in seconds, so I would not rely on it. Since we know the start time, I would add capacity in advance and warm caches and database connections before the sale opens. For demand above what we planned, a waiting line in front of the collection pages lets guests in at a rate checkout can handle, with an honest message. I would keep the collection traffic separate, so a guest buying groceries is not slowed by it. Beforehand, I would run a load test shaped like a sudden wall, not a slow ramp.”

“I would create a trace id when the order is placed and pass it along in every call and message, including to store devices. Each service records when it received the order, what it decided and when it handed it on. Devices that go offline save their events and send them later with the original times. Everything lands in one place where support can search by order id and see a timeline. I would keep names and addresses out of traces. For volume, I would keep every trace that had an error or was slow, and only a sample of the normal ones.”

“I cannot see inside the provider, so I measure from our side: how long from sending a message until delivery is confirmed, and what share fails. I alert on growing delay, because a slow provider often returns no errors at all. Messages wait in a queue with retries that back off, and each has an expiry, since a ready message that arrives two hours late only confuses the guest. If the delay passes a limit, we switch to a second channel such as email or text. The app always shows the true order status, so the message is a help and not the only source.”

“In most of my work, an outage meant a web page that failed to load. At Target it can also mean a register that cannot take a sale or a guest waiting in a car. That raises the standard, and I like that. Running software in a large number of stores, each with its own network and equipment, is a different skill from running a cloud service. I have experience with your project and want to grow on the store side. I also like the yearly rhythm, where the holidays are a test you can prepare for.”

“No release does not mean no change. First I would find the scope: one region suggests a network path or a regional service, not the devices. I would follow a sign-in step by step and time each part: device to store network, store to the sign-in service, and that service to its database. Then I would check things that change by themselves, such as a certificate close to expiry, a full disk, or a regional server that has dropped out of rotation. Meanwhile I would send that region's traffic to a healthy site if we can. Afterwards I would add monitoring for the cause.”