Redis & Caching for Java Services

09:00 – 10:15

Find the hot path with a measurement, not an opinion

timing at the boundary, reading the SQL log, p95 vs the average

Takeaway A slow endpoint profiled, with the responsible query named and its share of the response time measured.

10:30 – 11:45

Use Redis as a data store, not a magic box

strings and hashes and sets, expiry and eviction, redis-cli

Takeaway Four data structures exercised from the command line, with the trade-offs of each argued.

12:00 – 13:00

Halve an endpoint with cache-aside

cache-aside vs read-through, key design, choosing a TTL

Takeaway The slow endpoint cached, response time halved, and the hit rate visible on a counter.

14:00 – 15:15

Invalidate without serving stale data

invalidate on write, versioned keys, stale-while-revalidate

Takeaway A stale read reproduced deliberately, then closed with a versioned key.

15:30 – 17:00

Stop a cache stampede while it happens

request coalescing, jittered expiry, the thundering herd

Takeaway A stampede triggered under load and then stopped, the difference visible in the database connection count.

09:00 – 10:15

Keep the service up when Redis goes down

timeouts, circuit breakers, degrading honestly

Takeaway Redis switched off mid-load, and the service still answering: slower, but correct.

10:30 – 11:45

Use Redis beyond caching

rate limiting, distributed locks and their limits, pub-sub invalidation

Takeaway A rate limiter that holds across two instances of the same service.

12:00 – 13:00

Operate it: memory, persistence, monitoring

eviction policies, persistence options, hit rate and latency

Takeaway The key that grew found in a memory profile, and a dashboard that answers "is the cache working".

14:00 – 15:15

Test cached code against a real Redis

Testcontainers, asserting invalidation

Takeaway A test suite that fails when a cache key stops being invalidated.

15:30 – 17:00

Cache an unfamiliar service under time pressure

deciding what to cache, proving the improvement

Takeaway Your own before-and-after numbers, and an invalidation strategy that survives being questioned in review.