I’m Hammer, a backend engineer. I spent over five years at Channel.io building CRM and marketing systems that handled more than 100 million messages a month.
On this blog, I share lessons from running those systems in production, along with notes on what I’m studying.
Selected work
- Built a shared platform for bulk actions such as importing customer records, sending campaign messages, and tagging conversations in bulk. Fair queuing and per-tenant rate limits kept large jobs from monopolizing shared resources.
- Reduced head-of-line blocking in the bulk action queue through logical partitioning. Together with dynamic worker scaling, this cut the time to process 100,000 items from 25 minutes to 4 in a load test.
- Introduced ClickHouse for CRM audience segmentation so marketers could target customers using cart, order, and coupon data in a single query.
- Designed an asynchronously updated index table that isolated the main DynamoDB write path from GSI backpressure.
Start here
In English
- How we built and scaled a shared job queue. Building the bulk processing system · Partitioning the queue and scaling workers
- How we made cart, purchase, and coupon data queryable together. More Precise Customer Targeting with ClickHouse
In Korean
- KVCache부터 PagedAttention까지 · Nov 18, 2026
- 메모리 버퍼풀로 살펴보는 PostgreSQL, RocksDB, ClickHouse · Nov 5, 2026
- 메모리 할당기 — glibc, jemalloc, TCMalloc, Java, Go · Oct 21, 2026
- Go, Rust, Java의 런타임 스케줄링 · Oct 5, 2026