I'm building a simple application with mostly static data, although new records will be added over time. For verification requests, I only need to check whether a requested value exists. Would it be a good idea to load every record into memory when the server starts so requests can check the cache instead of querying the database? I'm considering deploying on a VPS and possibly running Redis or Memcached in Docker. I'm also wondering whether the cache should be refreshed whenever a new record is added, and whether treating a cache miss as "not found" is safe for this use case.
4 Answers
If the data is genuinely static for most users, serving generated files through object storage and a CDN could be simpler than maintaining a persistent application cache. However, if the application still needs to accept new database records, you’ll still need server-side write logic and a reliable way to publish those changes.
It can work if the entire dataset fits comfortably in memory and you only run one server instance, but the difficult part is keeping the cache accurate. Whenever a record is added or removed, the corresponding cache entry needs to be updated or invalidated. Also, measure before adding Redis or Memcached—an indexed database lookup may already be extremely fast because the database keeps frequently used pages in memory.
Would refreshing the cache whenever a new entry is added be enough? For this app, a cache miss can be treated as “not found” because I only need to verify whether the requested data exists.
I’d avoid loading everything during startup unless you have a measured performance reason. A startup warm-up process adds maintenance work and can make restarts slower. A simpler approach is to update or invalidate the relevant cache entry when data changes, then let normal reads use the cache. Be especially careful if you later run multiple server instances, since each one could hold a different copy.
A Redis or Valkey cache can be populated before startup, but that is essentially an ETL or cache-warming job you’ll need to maintain. For a small dataset, an in-process set or map may be enough, provided you update it whenever a record is created or deleted and accept that it must be rebuilt after a restart.
That may be more infrastructure than I need right now, since I’m still learning and plan to deploy the application on a VPS. I’m mainly trying to decide whether a local in-memory cache is worthwhile.