Cisco Backend Developer Interview Questions

“We sent a settings update to a group of remote gateways in your project. The update contained a wrong server address, and the gateways that applied it could no longer reach us. We noticed because the number of connected devices started falling. I stopped the rollout at once, so most devices never received it. For the affected ones, we were lucky that they kept the old settings as a fallback after repeated failures. Since then, every remote change goes to a small group first, and devices must confirm they can still reach the service, or they return to the previous settings on their own.”

“Our cloud team released weekly, while the firmware team released a few times a year, because each of their releases needed long testing on real devices. We once built a feature that needed a new field from the device, and then waited months. After that, I proposed we agree on the message format at the very start and write it down. The firmware team gave us a simulator that spoke the new format, so we could build and test against it. We made the cloud side accept both old and new formats. The feature was ready on the day the firmware shipped, and your result.”

“Since firewalls block incoming connections, each device would open an outbound encrypted connection to the service and keep it alive. The service stores the desired settings for every device with a version number. When an administrator makes a change, the service saves the new version and tells the device over its connection that a new version exists. The device fetches it, applies it and reports which version it now runs. The dashboard can then show desired against actual for every device. A device that is offline simply picks up the latest version when it reconnects, so no change is lost.”

“Devices would send readings in small batches to a front layer that only validates them and appends them to a durable log, so a slow database never pushes back on devices. Stream processors read the log, compute things like averages and write to a time-series store partitioned by customer and time. I would keep full detail for a short period, because that is what people need when fixing a fresh problem, and keep summaries by minute and by hour for much longer. Dashboards read the summaries. I would also protect the service from a single misbehaving device that floods it, with limits per device.”

“I would send the whole new settings as one document with a version and a checksum, not as separate commands that could stop half way. The device stores it next to the current settings and validates it first. Then it switches over in one step. Straight after, the device checks that it can still reach the cloud service. If it can, it confirms the new version and keeps it. If it cannot within a set time, it returns to the previous settings by itself, because the change may have broken its own connection. Applying the same version twice must be harmless.”

“A single missed heartbeat means little, so I would mark a device offline only after several in a row are missed. The harder case is when one of our receiving servers or a whole region fails. Then every device connected there goes quiet at the same moment. I would track which server each device was connected to, and if a large group on the same server falls silent together, treat it as our fault first. For those devices the dashboard should show status unknown, and offline alerts should be held back. Sending thousands of false alarms to administrators teaches them to ignore real ones.”

“In the moment, I would limit the rate of new connections at the front layer, rejecting the extra ones quickly, so the service can come up and accept devices in waves. Rejecting fast is kinder than accepting and timing out. For the lasting fix, the devices must not all retry on the same beat. Each device should wait a random time before reconnecting and double the wait after each failure, up to a limit. I would also make the reconnect itself cheaper, for example by resuming a session instead of a full sign-in. Then I would rehearse this failure in a test environment regularly.”

“In most backend work, the clients are browsers and phones that update themselves. Here the clients are routers, switches and access points in offices around the world, running many software versions, behind firewalls, and sometimes offline for days. A careless change in the cloud can cut an office off from its own network. I find that responsibility interesting. I have built your project, where I worked with long-lived connections and queues. Cisco makes the equipment and the cloud services that manage it, so I would see both ends. I want to learn how to design changes that are safe at that reach.”