After profiling our event pipeline I noticed a surprising number of short-lived allocations in the deserialization layer. Here's how sync.Pool brought GC pressure down by about 40% without making the code unreadable.
Read more →OOM kills that only happen in production are the worst kind. I spent a weekend tracing exactly how the kernel accounts for memory inside a cgroup v2 hierarchy and what you can do to set sensible limits without starving your services.
Read more →A table that gets heavy UPDATE traffic can accumulate index bloat even when autovacuum is running on schedule. I'll show how to detect it with pgstattuple and when REINDEX CONCURRENTLY is the right call.
Read more →