JPMorgan Chase DevOps Engineer Interview Questions

“We were releasing a new version of a checkout service late in the evening, and the team wanted to finish before a deadline. After the first group of servers took the change, I saw that failed requests were slightly higher there than on the old servers. It was small, and one colleague said it was noise. I compared the two groups over the same minutes and the gap stayed. I called the rollback and wrote in the channel what I saw. Next morning we found a connection setting that failed under load. We then agreed a written rule for when to stop, and your result.”

“A reviewer asked us to prove that every production change in the last quarter had been approved by a second person. Because all our changes went through the pipeline, I could show each deployment linked to a merge request, its approver and its ticket. That covered almost everything. The gap was a handful of manual changes made during incidents, which were only described in chat. I gathered those by hand, which took days. Afterwards I added an emergency path in the pipeline that still records who acted and why, and requires a review after the fact. The next review your result.”

“I would send a small share of traffic to the new version and compare it with the old one on error rate, response time and, most important, the share of payments that complete. If any of these is worse, the pipeline moves traffic back by itself. Servers being replaced stop taking new requests and finish the ones in progress. Database changes are split: first add the new column, then release code that writes both, then remove the old one later, so either version can run at any moment. I would also avoid starting a rollout just before a known busy period.”

“For account balances, losing a committed entry is not acceptable, so I would copy each write to the second site before confirming it, and accept the extra delay. For less critical data, such as preferences, delayed copying is fine. Failover needs a clear decision maker, human or automatic, and a guard so that the old site cannot keep accepting writes if it is only cut off, not dead. If some data was copied late, I would keep the old site's log and reconcile it after recovery instead of discarding it. We would practice the switch on a schedule, including the way back.”

“Secrets live in a central vault, and each service proves its own identity to read only what it needs. Nothing is stored in the code, the build image or the pipeline output. For rotation, I create the new key while the old one still works, let services pick up the new one on their next refresh, confirm that nothing is still using the old one, and then remove it. Services must reload without a restart. Every read from the vault is recorded, and an alert fires if something reads a secret it has never read before. People rarely need to see a value at all.”

“Server health alone would miss it, because servers can be up while sign-in fails. I would track successful sign-ins per minute and the share of attempts that fail, split by app version and region. The normal level differs by hour and day, so the alert compares with the usual pattern instead of a fixed line. I would also run test sign-ins from outside our network every minute with a test account. A sudden drop in attempts matters too, since it may mean the app cannot reach us at all. Each alert names an owner and links to first steps.”

“I would declare an incident so one person leads and others know where to look. The change from last night is the first suspect, and reverting it is cheap, so I revert it while someone else checks capacity and dependencies. I save the current logs and settings before the revert, for the review. Then I watch successful sign-ins, not just error counts, to see if customers are really getting in. Support and the status page get a short honest update at regular intervals. If the revert does not help, we move to the next suspect. Later we ask why the change was not tested under peak load.”

“I enjoy reliability work most when the service truly matters to people. At JPMorgan Chase the systems behind accounts, cards and payments are used at all hours, and a bad release can stop someone buying food. That raises the standard for everything I do, from pipelines to alerts. I am also comfortable with approvals and records. I have seen how a clear trail makes incidents shorter, not longer. I would like to work on making safe releases more automatic, so that the careful way is also the easy way for every team.”