9 posts
Grouping 10,000+ notifications makes chunk jobs finish out of order. waiting-children keeps one group and one send; invariants check the five races closed.
Quota was 100 messages per 5 seconds, average 20 req/s, yet 429s still fired. Misaligned fixed windows and a shared account forced a new sending path.
Sending AlimTalk, email, and push as HTTP calls hits limits fast. Why BullMQ takes on rate limits, scheduling, and delivery history.
Kubernetes judges a Pod with liveness and readiness probes: when each restarts a container or holds back traffic, Spring and NestJS setup, and pitfalls.
The first 1,000 tickets go through a web app behind a distributed lock. Does that lock actually keep concurrency under control when the rush hits?
Lettuce is not limited to spin locks. How Spring RedisLockRegistry's Pub/Sub lock mode builds an efficient distributed lock without Redisson.
Like-button races, solved with database locks. How pessimistic locking works, and how next-key locks slow the check for whether a review was already liked.
1,000 people request 100 items at once. How a single UPDATE gives one to exactly 100 of them, instead of the locks usually proposed for a first-come event.
Loading 40–60 delivery plans per screen from an external API was slow. Redis and local caches made it 50–100x faster, and Redis Pub/Sub kept servers in sync.