Salesforce Backend Developer Interview Questions

“In my last role, one internal team called our search service in a tight loop during their nightly job, and search slowed down for everyone. I proposed a limit on requests per client per minute. That team objected, because their job would take longer. I showed them the latency charts and offered a batch endpoint that could answer many lookups in one call. I set the limit from the usage of normal clients, with room above it, and returned a clear error that said when to retry. Their job ended up faster with the batch call. The result was your result.”

“A customer reported that saving an order sometimes took very long, and nobody else had the problem. I could not reproduce it on a standard setup. With their permission, I copied their configuration to a test environment and compared it with a default one. They had added a large number of optional fields, and one of our queries loaded every field for each line of the order. With a normal setup this was cheap, but with theirs it was slow. I changed the query to load only what the save needed, and added a test with a heavy configuration. The result was your result.”

“Checking each deal against the manager chain while reading would be slow for a list with many rows. I would compute access ahead of time in a share table, where each row says that this user or group can see this record, and why. Managers are handled through groups: each manager has a group that contains everyone below them, so a deal shared with the owner is visible to the groups above. A list query then joins deals to the share table for the current user's groups. The cost is on change. When someone moves to a new manager, a background job updates group membership.”

“I would give every run a budget. The runtime counts processor time, memory, the number of database queries and the rows touched. When the code goes over any of them, the run is stopped, the whole transaction is rolled back, and the customer gets an error that names the limit they hit. The limits are published, so developers can design for them. The most common mistake is a query inside a loop over records, so I would hand the code all the records of a save in one list and document how to query once for the whole list.”

“I would not process the file inside one request. The API accepts the file and returns a job id, and the customer asks for the status later. In the background I split the file into batches and process each in its own transaction. Inside a batch, a row that fails validation is written to an error file with its row number and the reason, and the good rows are still saved. At the end the customer can download the successes and the failures. To make a retry safe, each row can carry an outside id, and I update the record if that id already exists.”

“I would keep two layers. The metadata layer says, for each customer, which record types exist, which fields each has, and their data types. The data layer is a shared table where each row has the customer id, the record type and a set of general value columns or one document column. The metadata maps a field name to its place. Every query is translated through that map. The hard part is speed, because a general column cannot have one index per customer field. So for fields that a customer filters on, I would copy the values into a separate index table with typed columns.”

“First I would stop the damage by asking the customer to pause the sync, or by blocking the creating call for a short time. Then I would look at the request logs from the night the duplicates started. A common cause is that the sync timed out, retried, and our side created the record twice because nothing identified the repeat. I would count the duplicates and check which copy has deals or cases attached, because a merge must keep those links. After the cleanup, I would make the create call accept an outside id or a request key. I would write the customer a clear summary.”

“What interests me is that the backend here is not one application. It is a platform where each business adds its own fields, rules and code, and all of them run on shared systems. That means I cannot assume what the data looks like, and I have to design limits that are fair to everyone. I enjoyed a smaller version of this in my current job, where I built configurable rules for our customers. I also like that the users are sales and service teams doing real daily work. I want to learn how to keep such a flexible system fast.”